SCROLL NEWS / DISCOVERY
Search headlines
Find the stories shaping the conversation.
Results for " maxim " 249 found 🔀 AI-powered Shuffle
AbCellera women’s health data top ‘most aggressive upside scenario,’ putting Astellas and Bayer on notice
AbCellera has reported phase 2 results “beyond even the most aggressive upside scenario” that its CEO dared to consider, boosting the biotech’s chances of challenging Astellas and Bayer for a women’s health market. The phase 2 clinical trial compared ABCL635, an antibody targeting NK3R, to placebo in patients with hot flashes caused by menopause. Investigators enrolled 92 postmenopausal women experiencing a mean of about 10 moderate or severe hot flashes per day. By Week 4, the number of moderate or severe hot flashes had fallen by a mean of 8.8 on ABCL635 and by 3.5 on placebo. That statistically significant, placebo-adjusted reduction of 5.3 hit the trial’s primary endpoint. The level of efficacy also cleared the bar set by existing drugs. While cross-trial comparisons can be misleading, AbCellera CEO Carl Hansen, Ph.D., had seen enough to start hyping the program. “We now view ABCL635 as a potential best-in-class non-hormonal therapy with strong differentiation against approved NK3R competitors across all important dimensions, including efficacy, safety, dosing and convenience,” Hansen said on a Monday morning conference call with investors to discuss the data. “We believe ABCL635 is now a highly de-risked program.” Last week, Hansen used AbCellera’s earnings call to set expectations for readout, with the CEO defining success as a clean safety profile and efficacy comparable to the two approved small molecules. The data vaulted over that bar. Astellas linked (PDF) one of the small molecules, Veozah, to placebo-adjusted reductions in hot flash frequency of 2.1 and 2.6 at Week 4 of its two phase 3 trials. Bayer reported (PDF) placebo-adjusted reductions of 3.04 and 3.29 at Week 4 of pivotal studies for its small molecule, Lynkuet. At 5.3, the placebo-adjusted reduction in AbCellera’s trial comfortably cleared the standard set by the incumbents. Veozah and Lynkuet came to market based on evidence of their ability to reduce the frequency and severity of hot flashes. AbCellera’s phase 2 study covered both endpoints, with the Canadian biotech reporting a mean reduction in severity of 1.4 points at Week 4 on ABCL635, compared to 0.3 points for placebo. Again, the placebo-adjusted result compares favorably to data on Veozah and Lynkuet. Analysts shared Hansen’s enthusiasm for the data. The phase 2 results “firmly validate ABCL635 as a potentially best-in-class treatment,” BMO Capital Markets analysts said in a note to investors Monday. ABCL635’s apparent efficacy edge “is likely attributable to the higher dosing/target engagement the asset is able to reach vs. the oral agents due to its wider therapeutic index,” the analysts added. The analysts, who model $2.3 billion in unadjusted peak sales for ABCL635, flagged the safety profile as another positive of the phase 2 readout. While Veozah carries a boxed warning for liver damage, nobody on ABCL635 in the phase 2 trial had elevated liver enzymes or bilirubin values. If replicated in phase 3, the lack of a liver signal could allow approval without the monitoring requirements imposed on Veozah. Investors appeared equally enthused, sending AbCellera's stock up 32% to $9.21 in premarket trading Monday from a Friday closing price of $6.93. Next, AbCellera will collect the Week 12 data and talk to its advisors about further development. The results open up “a wide variety of possibilities of how we might take this forward into phase 3 studies to truly maximize the value of the asset and make sure that ABCL635 is able to demonstrate its differentiation potential,” Sarah Noonberg, M.D., Ph.D., AbCellera’s chief medical officer, told investors.
Starwatch: Deep partial solar eclipse in UK to precede the Perseids
The eclipse will begin just after 6pm on 12 August, reaching its maximum just over an hour later
Ukraine uses business principles to maximize Russia’s casualties and economic costs, potentially crushing its will to keep fighting, RAND says
Ukraine has KPIs for the military—just like corporations, which use them to track output from employees or subsidiaries.
Federal Policy Priorities To Strengthen Rural Health and Maximize Rural Health Transformation Program Investments
Executive Summary Rural health care in the United States is at an inflection point. The Rural Health Transformation Program (RHTP), enacted as part of H.R. 1, is a $50 billion, five-year investment designed to help states improve access, quality, and outcomes by transforming rural health care delivery. This investment takes place alongside an estimated $137 … Continued
They made a living on Russia’s Amazon. Ukrainian attacks have sent it up in flames.
Attacks have hit warehouses belonging to the company almost daily — a new chapter in Kyiv’s campaign of long-range drone strikes designed to inflict maximum pain on the Kremlin.
3 Blockbuster Trade Targets Phoenix Suns Should Monitor
As Phoenix navigates a pivotal crossroads, targeting high-impact stars like Zion Williamson and Anthony Davis could bridge the gap between developing youth and maximizing Devin Booker’s prime.
Praxis Profit launches new executive advisory service to help Tahoe business owners strategically maximize profitability
SOUTH LAKE TAHOE, Calif. – Praxis Profit, a boutique strategic chief financial officer advisory firm, announced today its full-suite of financial decision-making consultancy service including cash flow visibility, margin protection and executive level financial decision-making...
Why losing weight may not be enough to prevent type 2 diabetes
A healthy lifestyle can prevent type 2 diabetes for many people, but new research suggests that weight loss alone may not be enough for one particularly vulnerable group. Even after losing about 8% of their body weight and keeping it off for years, people in a high-risk category known as cluster 5 continued to experience rising blood sugar, weakening insulin production, and a strong risk of developing diabetes.
Trek Voluntarily Imposes E-Bike Speed Limits
All Trek e-mountain bikes sold in the US will have a 20mph maximum pedal assist speed.
KGK Science secures investment to expand nutraceutical clinical research
North American nutraceutical CRO KGK Science will use growth investment from Maxim Partners to expand clinical trials, decentralised research and regulatory services for dietary supplement and consumer health companies
High School Science That Teaches an ID Perspective? Yes, Please!
Kristin Marais and Summer Edwards explain why they’re so passionate about teaching their subject and how the courses are designed with maximum flexibility.
Falcons Reacts Survey: What is Tua Tagovailoa’s outlook?
How will the former top-five pick maximize his fresh start in Atlanta?
Smart Tips to Maximize Your NASCAR Weekend at New Hampshire Motor Speedway
At New Hampshire Motor Speedway, fans get more than just racing — they get great value. New England's only NASCAR Cup Series weekend features affordable tickets, free parking, flexible camping options and inclusive entertainment like live music and fan experiences.
Should there be political age limits?
The case for and against age maximum limits for our political leaders. And should we also have health transparency rules?
New Jersey behavioral health company acquires independent medical examination provider
Red Bank, N.J.-based PAX Health acquired Ho-Ho-Kus, N.J.-based Makraf IME, a statewide psychology and neuropsychology independent medical examination (IME) provider group operating across 17 counties. Makraf IME provides independent medical examination opinions on causality, need for treatment, maximum medical improvement, permanency and fitness for duty. The practice also provides psychology and neuropsychology IMEs, disability evaluations, […]
How to maximise the mental health benefits of your vacation
Don’t just hit the beach. Research is revealing how to have a holiday that leaves you mentally restored and refreshed – with long-lasting effect
Shohei Ohtani pitching start scratched, no set timetable for return
The original plan was for Shohei Ohtani to start on Wednesday’s series finale in Philadelphia, putting him in the sixth rotation spot after the break, maximizing his time off from pitching to rest his left knee, while also pitching directly before an off day. But those plans changed over the weekend, after Ohtani felt discomfort […]
Blues duo to perform at Earlville Opera House
Blues musicians Maximilian Eyle and John Driscoll will perform July 25 as part of the Earlville Opera House Courtyard Series. Tickets are on sale now.
Doing More With Less: How Counties Can Successfully Address Behavioral Health Crisis Response Needs
Join Bamboo Health for a practical discussion on strategies county health departments can use to maximize resources and improve care outcomes.
Maximum Entertainment’s U.S. Subsidiary Agrees to Cathay Bank’s Request for Stipulation of a Receiver
Stockholm, Sweden, July 17, 2026 - Maximum Entertainment AB (Nasdaq: MAXENT B) (“Maximum Entertainment” or the “Company”) today announces that its U.S. subsidiary, Maximum Entertainment, LLC (the “Borrower”), has agreed to stipulate to the appointment of a court-supervised receiver as requested by…
Sports without politics is a fantasy and the World Cup proves it
From ancient Greece to Nazi Germany, sport and politics have long been intertwined. Nevertheless, we still imagine that international sport can be kept above the fray and serve as a neutral arena where nations compete on merit alone. In this article, diplomacy scholar Geoffrey Allen Pigman argues the 2026 FIFA World Cup shows this is a dangerous illusion. When international sport is left entirely to revenue-maximizing private bodies like FIFA, it doesn't escape politics, but rather surrenders to its worst expressions. With the 2028 Los Angeles Olympics fast approaching, the IOC must decide whether to learn that lesson or repeat the same mistakes. The largest and most lucrative FIFA World Cup ever draws toward its climactic conclusion amidst a swirl of controversy. Team members, staff, fans, and spectators have been denied their right to enter the United States to participate, and US President Donald Trump intervened with FIFA boss Gianni Infantino to overturn a referee’s call that Trump believed disadvantaged the US national team. It is a moment to step back and reflect on what the purposes of international sport mega-events like the World Cup, and international sport more broadly, are, and what they should be. Does international sport serve public purposes? Is there a public interest? Or is it primarily a private endeavor organized by commercial enterprises to entertain the global public (and generate profit in the process), which is occasionally used by governments and leaders for raisons d’etat? Central to the idea of sport is the principle of fair play, which in the international context demands that games be free from political interference. However, it is impossible to separate sport from politics, all claims to the contrary notwithstanding. The most positive political use of international sport through the millennia has been to facilitate diplomacy. The use of international sport for diplomatic objectives has existed since at least the time of the ancient Olympic Games, around which was instituted the custom of the Ekecheiria, or Olympic Truce, between participating states to facilitate fair competition and diplomatic achievements by ensuring that participants could travel safely to the Olympiad. In his seminal work On Diplomacy (1987), diplomacy scholar James Der Derian characterized the objectives of diplomacy as overcoming alienation and mediating estrangement between governments and between peoples. One increasingly important diplomatic tool in the technology-infused media environment over the past century has been public diplomacy, under which governments and non-state diplomatic actors (e.g. transnational corporations) communicate to global publics about their international policy objectives and to shape their global brand image. Political scientist Michał Kobierecki, in his important book Sports Diplomacy (2020), views sports diplomacy as falling broadly within the ambit of public diplomacy, as international sport intimately involves the participation of global publics as spectators and consumers of global media. Diplomacy scholar Stuart Murray and this author, in a 2013 article, distinguish between international sport used for diplomatic purposes and international sport-as-diplomacy, which encompasses the range of diplomacy that occurs as teams and competitors, spectators and fans, and government officials all converge for international sporting events like the Olympic Games and FIFA World Cup. Diplomacy scholar H.E. Chehabi, in a 2001 article on US-Iran sports diplomacy, describes the interaction between spectators and fans at international sporting events as “people-to-people diplomacy,” in which relationships develop independently of governments and their objectives. Diplomacy by the aforementioned definitions is often unsuccessful. Moreover, all too often the tools of diplomacy may be employed by governments and other actors to thwart or oppose the objectives of diplomacy as characterized by Der Derian. Governments routinely use the tools of public diplomacy to disseminate propaganda, spreading misinformation in the attempt to deceive global publics about their objectives and to create misleading impressions about their brand identity. Within sports diplomacy, the term “sportswashing” has been coined recently to describe how governments invest in and promote international sport to create positive global public impressions of their nations and, in doing so, to distract the global public from their deficient records in areas such as human rights and democratic governance. Murray, in his 2018 book Sports Diplomacy: Origins, Theory and Practice, introduces the concept of sports anti-diplomacy, wherein governments intentionally use international sport to advance aggressive and nationalistic objectives. Obvious historical examples include Hitler’s use of the 1936 Berlin Olympics to glorify his Third Reich and the 1969 El Salvador-Honduras football “war.” SUGGESTED READING The beautiful game made ugly By Emily Ryall FIFA argues that it is and seeks to be apolitical in its stewardship of international Association Football (soccer). FIFA policies governing bids to host the World Cup require uninhibited and non-discriminatory access for players, team staff, and spectators of all participating countries to host nations and venues. It became clear soon after Trump returned to power in the United States in January 2025 that Trump’s political intentions and objectives to limit non-US citizens from living and working in the United States and, in numerous cases, even visiting the country, would collide with FIFA’s host nation policies. Beginning soon after the start of the second Trump administration, US federal authorities imposed restrictions on entry to the United States of citizens of between 20 and 40 countries for any purpose, with only limited exceptions. Federal Immigration and Customs Enforcement (ICE) began arresting non-US citizens in the United States, including persons with and without legal status, holding them in jails, concentration camps, and other detention facilities until they were deported to their home or third countries, where they faced substantive risks to their personal safety. FIFA boss Gianni Infantino, in an effort to ensure the most successful and lucrative World Cup in history, engaged diplomatically with President Trump in an attempt to facilitate entry to the US for World Cup participants. Infantino did all in his power to ingratiate himself with Trump, to the extent of creating a FIFA “Peace Prize,” which he awarded to Trump in December 2025 at a ceremony in the then Trump-controlled John F. Kennedy Center for the Performing Arts in Washington, DC, on the occasion of the World Cup draw. The prize was an ersatz consolation for someone who aspired in vain to win the rather more prestigious and well-known Nobel Peace Prize. And in the run-up to the World Cup, conditions became worse. In February 2026, Trump and Israeli Prime Minister Benjamin Netanyahu launched a war of aggression against Iran, illegal under international law, under the false pretence that Iran’s government was preparing to attack the United States and Israel. During the US aerial bombardment of Iran, US planes bombed Azadi Stadium, the facility where Iran’s national football team trained. As the 2026 FIFA World Cup began, it was clear that Infantino’s efforts had been for naught: the Trump administration would not adhere to FIFA policies governing access to World Cup host nations. World Cup referee Omar Artan, a citizen of Somalia, was denied a US visa to enter the United States to participate in the World Cup. Trump’s government only granted US visas to Iran’s football team ten days before they were scheduled to arrive for the tournament. Numerous members of the support staff for Iran’s team were denied US visas. Iran’s team, which had been scheduled to train in Arizona, was directed to shift its training site to Tijuana, Mexico, notwithstanding that all three of their group stage matches took place at US venues. The US Department of Homeland Security only permitted the Iranian team to enter the US on the day before each scheduled match and then to return to Mexico immediately after playing. Members of Iraq’s and Switzerland’s football teams were delayed for hours by US officials at the US border, and an Iraqi team photographer was refused entry. Spectators and fans from Haïti and Iran were denied entry to the United States altogether. A large number of fans from a range of countries had their applications for US visas rejected. There were multiple off-ramps through which participants could have avoided the negative consequences of Trump’s decisions for the 2026 World Cup. Once it was clear that the Trump government intended to violate the established expectations for World Cup hosts, FIFA could have shifted the matches scheduled to take place in the United States to Mexico and Canada. The three nations were already designated as co-hosts, and host cities and venues had been selected. Once it was clear that FIFA, governed by FIFA boss Gianni Infantino’s dependent and dysfunctional relationship with Trump, would never consider this option on its own, it fell to the national football federations to act. The national federations, particularly some combination of federations of the leading contenders (France, Spain, Brazil, Argentina, Morocco, England, Germany, etc.), could have refused to participate under the US-imposed conditions or, at the least, demanded an ironclad agreement with the Trump government that FIFA policies on access would be respected. ___ International sport is inextricably entangled with diplomacy of different sorts, for good or ill, it cannot be considered to be in the purely private realm and left to revenue-maximizing private actors. ___ Why did neither FIFA nor the national football federations act to protect the interests of players, team staff, referees, fans, and spectators? In essence, the relentless engines of capitalism prevailed over norm observance, ethics, and fair play. The revenue interests of FIFA, sponsors, global media, and host cities won out. FIFA’s managers viewed hosting another World Cup principally in the United States as a veritable pot of gold for FIFA and its collaborators. FIFA has estimated its own revenue over the 2023-26 budget cycle, which includes the 2026 World Cup, at $13 billion, revised up from an initial projection of $11 billion, and 72% higher than the previous revenue cycle. Of that total, FIFA is estimated to have generated $8.9 billion from the Cup itself in revenue from TV rights ($3.92 billion), sponsorships ($2.69 billion), ticket sales ($3.1 billion), licensing rights, and other sources. A joint study by FIFA and the World Trade Organization estimates that the World Cup will contribute over $40 billion to global GDP, including nearly $14 billion in direct visitor spending in the three host countries (but principally in the United States, which hosted the lion’s share of the matches). The FIFA World Cup is now reportedly “the single most lucrative recurring commercial property in global sport.” And yet, notwithstanding the Trump government’s uses of the World Cup for sports anti-diplomacy, and the climate of anxiety that this created for participants and spectators, diplomacy that did not depend upon governments was able to thrive at the World Cup. Examples of successful people-to-people diplomacy were legion, as citizens of towns and cities across the States, as well as in Mexico and Canada, welcomed team members, their families and supporters, to training sites across North America. Fans of different teams joined together in person in stadiums for matches, at watch parties and bars, and at welcome events sponsored by host cities. The joy and camaraderie shared between the citizenry of Boston and Scotland’s “Tartan Army” of fans is but one example of how the “Beautiful Game” brings people together. Friendships forged across continents are likely to endure well beyond the Cup final. Even those achievements cannot be seen as unalloyed, however. As South African football fan Byron Pillay commented to Al Jazeera, “Football is called the Beautiful Game for a reason, for its ability to unite people. But it’s hard to believe in that magic with the politics and war rhetoric off the field of play, especially when one of the tournament hosts is central to that.” Ultimately, the experience of the 2026 FIFA World Cup illustrates that, as international sport is inextricably entangled with diplomacy of different sorts, for good or ill, it cannot be considered to be in the purely private realm and left to revenue-maximizing private actors. As long as there is the possibility of diplomatic progress in mediating estrangement between governments or between peoples, there is a broad and overarching public interest in international sport. Likewise, as it would be difficult to put on an international sport mega-event without international sporting organizations like FIFA and the International Olympic Committee (IOC), global media firms, and companies purchasing sponsorships, public and private interests in international sport must coexist, most likely somewhere on a shifting spectrum between partnership and rivalry or cross purposes. As the World Cup wraps up, the IOC needs to consider its lessons before the scheduled 2028 Los Angeles Olympics if they want to avoid a media-hyped Trump version of the 1936 Berlin Olympiad. The IOC and the National Olympic Committees need to set conditions on and make demands of Trump’s government with clearly articulated consequences for non-compliance, up to and including cancellation of the Games. Time is running out for the IOC, as unlike with the FIFA World Cup, there is no at-the-ready alternate venue for Los Angeles.
The Moment Victor Wembanyama Decided to Pass Up $50 Million
Victor Wembanyama signed a five-year, $251 million maximum contract with the San Antonio Spurs instead of a $300 million supermax deal with individual performance incentives.
When regulation is good for business
By now, I think I have established both my view that I am a fan of ESG investing and the benefits of regulating what is and isn’t a sustainable investment, and my view that in Europe, ESG regulation has gone too far and become too costly. It almost feels like the old Aristotelian maxim of doing things in moderation that leads to the best outcomes is valid in regulatory efforts as well. To give you an example, let’s look at a study by the National Bank of Belgium on the impact of the EU’s Sustainable Finance Disclosure Regulation (SFDR) on fund flows of conventional and ESG funds. The SFDR was announced on 27 November 2019 and came into effect on 10 March 2021. It was implemented in several stages and split funds marketed in the EU into three broad categories: (i) Art. 6 funds that are basically conventional funds without any specific ESG integration, (ii) Art. 8 funds that ‘promote environmental or social characteristics’, which must provide additional reporting how they do that, and (iii) Art. 9 funds that are impact funds. The gaps between the announcement and the phased implementation give us an idea of how fund flows were affected by the regulation. It is clearly visible in the figure below that once the new regulation was announced and when it was initially implemented, fund flows in Art. 8 and Art. 9 funds increased significantly. Conventional Art. 6 funds also saw more inflows, but the increase was not statistically significant. I should add that the data below is pure fund flow, cleaned of performance effects and of any reclassified funds that moved from one classification to another. Note also that once the initial rush in ESG funds was over, additional, stricter regulatory requirements didn’t create that many new flows. This implies that with additional regulation, the funds incurred higher costs to comply but didn’t benefit in terms of new inflows and higher assets under management. Fund flows as the SFDR regulation was implemented in the EU Source: Emiris et al. (2026) When examining the drivers of these flows, the authors of the study found that institutional investors were more likely to switch to ESG funds and particularly so in countries where concerns about sustainability are culturally more accepted. In a society where buy-in into environmental and social risks is higher, private investors were shifting more of their assets into ESG funds as the regulation made it clear which funds are really green and which ones are not. And institutional investors in these countries did so as well, given the societal pressures and reputational risks of sticking with conventional funds. Meanwhile, fund managers reacted to the changing regulatory environment as well. Many funds that were mildly green before the regulation was announced became greener and more sustainable as the regulation was introduced to attract new assets and new investors. Cleverly, and unsurprisingly, these ESG funds also charge on average higher fees, which means that not only did fund managers gather fresh assets, but their profitability increased as well.
Inside the Record-Setting Business of the Los Angeles Dodgers with President & CEO Stan Kasten
Stan Kasten, Los Angeles Dodgers President, CEO and Part Owner, joins Sports Business Radio to discuss the record-setting business of the Dodgers. From superstar Shohei Ohtani's impact on attendance, sponsor sales and merchandise sales, to improvements to the fan experience at Dodger Stadium, to the Dodgers surpassing 4 million fans last season and being on track to draw that many fans again this season, we discuss how in many regards, the Dodgers are maximizing revenues better than any other pr
Inside the Record-Setting Business of the Los Angeles Dodgers with President & CEO Stan Kasten
Stan Kasten, Los Angeles Dodgers President, CEO and Part Owner, joins Sports Business Radio to discuss the record-setting business of the Dodgers. From superstar Shohei Ohtani's impact on attendance, sponsor sales and merchandise sales, to improvements to the fan experience at Dodger Stadium, to the Dodgers surpassing 4 million fans last season and being on track to draw that many fans again this season, we discuss how in many regards, the Dodgers are maximizing revenues better than any other pr
Green MP Hannah Spencer aims to pass law on maximum workplace temperatures
Byelection winner says heatwaves are causing ‘absolute chaos’ and workers need protection from unsafe conditions
NAVER D2SF Invests in 23i, a Virtual Entertainment Startup
NAVER D2SF, the corporate venture capital arm of NAVER, has made a new investment in 23i (CEO, Hyungmin Kim), a virtual entertainment startup. As the global virtual entertainment market continues to grow rapidly, particularly in Korea and Japan, NAVER D2SF decided to invest in the team based on its ability to create expanded entertainment IP through strong content production capabilities and an in-house technology pipeline designed to maximize the strengths of virtual artists.
DePaul Athletics and Playfly Sports Enter Multimedia Rights Partnership
CHICAGO, IL / NEW YORK, NY, July 8, 2026 – DePaul University Athletics and Playfly Sports, the sports industry's leading revenue maximization company, today ...
Ubiquiti warns of new max severity UniFi OS vulnerability
Ubiquiti has released security updates to patch seven critical vulnerabilities in UniFi OS, including a maximum-severity flaw that can be exploited in command injection attacks.
MLB Players Propose Limiting Optional Assignments To Minor Leagues, More
MLB players proposed that optional assignments to the minor leagues be cut from a maximum of five per season to three.
Could 'humanmaxxing' actually help you live longer? Here's what experts say
Humanmaxxing is the latest wellness trend focused on maximizing human potential through biohacking, longevity science and advanced medical treatments.
Could 'humanmaxxing' actually help you live longer? Here's what experts say
Humanmaxxing is the latest wellness trend focused on maximizing human potential through biohacking, longevity science and advanced medical treatments.
Making Our Own Luck
We are cross-posting a piece that former NIH official Kris Willis wrote for Macroscience. Kris is now the President and Founder of the Woodley Park Institute, for which I’m a board member. Picking winners or hedging bets? Federal science agencies have long wrestled with the question of how to ensure their billions of dollars of grants and contracts result in maximum benefits for Americans. On the surface, it might seem that a goal-oriented, interventionist style of management would be the best policy: prioritize our most pressing problems and distribute funding accordingly.This top-down strategy creates a unique hazard, though: bad choices can send resources in the wrong direction, resulting in slow progress that impedes the delivery of tangible benefits to citizens. Moreover, if decision makers pursue an applied advance when the underlying scientific principles remain poorly understood, they risk costly failures and a loss of support for the research enterprise at large. Examples include Nixon’s 1971 War on Cancer, undertaken on the premise that a cure could be achieved in less than a decade, the drive to build an operational commercial fusion plant by the mid-1990s, and the National Plan to Address Alzheimer’s Disease, which failed to produce the effective treatments and prevention strategies promised by 2025. The contrasting laissez-faire approach is to hedge your bets. If the path to an advance is impossible to predict, then the wisest choice might be to distribute funding to as widely as possible. The main risk of this strategy is that spreading funding thinly may limit the resources available to the most promising research, resulting in the same negative outcomes as the interventionist approach: limited support for the most promising topics, slow progress, and delayed returns. In reality, at science agencies like the National Science Foundation and National Institutes of Health, program managers navigate between these two extremes, setting some funds aside to target high-priority areas and using the rest to cover the widest possible range of meritorious proposals. This middle path seems like the most rational way to proceed, but it still requires making choices. How much should be set aside for high priority topics? What should those topics be? How should we define merit? Expert opinion can provide a helpful guide, but leading experts can have remarkably different opinions about what matters and what should be prioritized. Moreover, experts and administrators alike can be risk averse, giving an edge to high-profile, well-established concepts. What’s really needed is a data-driven framework to help guide decision-making, one that doesn’t simply repackage prestige or incumbency. Recognizing breakthroughs years in advance Many breakthroughs fail to attract funding or recognition at the earliest stages of development. Katalin Karikó, for example, spent years struggling to win grants for work on mRNA that would eventually win her and Drew Weissman a Nobel Prize. Similar stories can be told about Jim Allison’s groundbreaking studies leading to the development of cancer immunotherapy, Stan Pruisner’s demonstration that prions can self-replicate without DNA, or Robin Warren and Barry Marshall’s proof that Helicobacter pylori, not stress, causes ulcers. Yet at some point, the scientific community recognizes and responds to transformative breakthroughs like these. How does that change happen? And what if science funders could identify the areas that are on the cusp of a breakthrough? Those questions led my collaborators and me to study how scientists responded to past breakthroughs. Based on patterns in that data, we developed a means of predicting future breakthroughs, using biomedicine as a proof of concept. It’s worth taking a moment to describe how our approach works. Briefly, our prediction process begins by grouping the 18 million or so peer-reviewed papers in PubMed, the authoritative database of biomedical research, into roughly 50,000 unique topics. A small subset of these topics are so widely acclaimed as breakthroughs that they have been recognized with a major prize like a Nobel or a Lasker Award. We theorized that studying the development of these outliers over time, especially before they gained acclaim, was the key to identifying future instances of them. After defining topics, we chose 21 representative breakthroughs and asked what they looked like before they won prestigious prizes. To answer that question, we re-ran our topic-mapping algorithm repeatedly, winding back the clock one year each time, stringing together the years to follow the progress of each field as it grew and advanced. We found that in each case, the years leading up to a discovery followed a pattern: a burst of papers on a new, rapidly evolving topic, many of which quickly became influential. These characteristics—the percent of papers that are new, the percentage of papers that are brought in from other topics, and the influence of each individual paper—can all be measured separately for any given topic. Together, they act as a predictive signal that is detectable an average of five years before a transformative discovery is made, and up to thirty years before the discovery receives a major prize. Scanning the current research landscape for topics that display this signal allows us to predict what work will likely produce a future breakthrough. Each year we analyzed includes four or five signals, a number that appears to have remained constant across a twenty year time frame. In their classic work on the sociology of science, Bruno Latour and Steve Woolgar established that the desire to participate in a major discovery is an important motivator for scientists as they consider whether they should take up (or abandon) a research problem. Although further work is needed, the simplest explanation for our results is that this behavior is widespread enough to detect at scale. A foundational advance, or something like it, draws the attention of scientists and causes them to change the direction of their research. They vote with their feet, staking their own reputation and careers by publishing on and citing the new idea. That upswell of interest is the foundation of our breakthrough signal. What kind of research does our model identify as a breakthrough? After building a model based on the patterns in our 21 preselected breakthroughs, we tested it by identifying all early signals of discovery from 1994 through 1997. Of the 18 signals we found in that time period, 17 can be traced forward in time to a breakthrough. Examples include super-resolution fluorescence microscopy (see the image above), the directed evolution of proteins and enzymes (Chemistry Nobel, 2018), and sequencing the human genome (National Medal of Science, 2008). The discovery of the role of leptin signaling in obesity is a particularly interesting case study. Around 1950, scientists noticed the existence of a type of lab mouse that suffered from a unique inherited form of obesity. When given the freedom to choose their own diet, these mice ate more than normal and preferred food that was high in fat. Researchers had few clues as to why the animals overate; early experiments pointed to the existence of a soluble ob factor found circulating in the bloodstream that regulated appetite, but its identity was unknown. In 1994, Jeffrey Friedman and colleagues identified and sequenced the ob gene, noting that the protein it encoded appeared to fit the profile of a circulating factor. A year later, a trio of high-profile publications demonstrated that ob was a hormone that regulated body weight and fat deposition by regulating appetite. One of these was co-authored by Friedman, who christened it leptin from the ancient Greek word for thin. Using our model, the data from 1996 make it clear this advance would become a breakthrough. Scientists had flocked to study the biology of the new hormone, generating the flurry of new papers and citations required to produce a signal. One of these demonstrated that serum leptin concentrations reflected the amount of adipose tissue in the human body, providing clinical validation of the earlier mouse studies. Before the identification of leptin, scientists were unable to point to a specific molecule that controlled appetite and adiposity, so that it was impossible to rule out non-physiological causes of obesity; its characterization fundamentally changed our concept of weight gain. In 2010, Friedman and fellow pioneer Douglas Coleman were awarded a Lasker for the discovery. Our work also identifies breakthroughs in clinical practice and behavioral and social sciences, even though these areas are less likely to draw the attention of major prize committees. One example that falls into this category is the introduction of endovascular aneurysm repair, a minimally invasive surgical procedure for the treatment of abdominal aortic aneurysm, a leading cause of death among Americans over the age of 65. Relative to the previous standard of care, it reduces in-hospital mortality by almost four-fold. In 2017, Timothy Chuter received the American College of Surgeons’ prestigious Jacobsen Innovation award for his multiple refinements to the technique. Another under-recognized breakthrough identified by our model is the development of standardized survey instruments to assess the quality of life for HIV+ patients. The antiretroviral cocktails introduced in the mid-1990s reduced AIDS mortality, but treatment came with serious side effects. The new evaluations showed that patients whose disease progressed had worse physical functioning than Americans with other chronic diseases, while those who were asymptomatic maintained a health-related quality of life on par with the overall US population. This evidence made it possible for patients and activists to argue that even in the absence of a total cure, and in spite of the side effects, alleviation of symptoms was a worthwhile priority for clinicians. My colleagues and I have failed to identify any significant recognition for the physicians and health policy experts who pioneered this advance, although in 2016, researchers called for the World Health Organization to set targets for good health-related quality of life as part of its framework to end the AIDS pandemic. From forecasting to funding How should science funders use our predictions? Any serious discussion of what we should do with this new capability needs to begin with an understanding of its limitations. We can’t (and shouldn’t) invest all of our resources in the small number of topics predicted to produce breakthroughs. Innovative new fields are born from existing ones; they rarely arise de novo. If funders don’t maintain a diverse portfolio, the breakthroughs of tomorrow will have no antecedents. Think of this as eating the seed corn, or killing the goose that lays the golden egg. There are also multiple goals of funding beyond basic scientific discovery, including translating discoveries into clinical practice, developing new technologies, supporting economic development, and training new scientists. The investments needed to accomplish these goals are likely different from those required to support emerging breakthroughs. Finally, we must be aware that we may not capture every instance of a breakthrough. The concept is fuzzy, and not every significant discovery is recognized with a major award. Further, breakthroughs are rare, meaning that our dataset is, by necessity, small. This makes formal estimates of accuracy challenging. Although our overall success rate appears to be high, any one signal may turn out to be a false positive. We found one apparent example: the development of the third generation COX-2 inhibitors, Vioxx and Celebrex, which were much celebrated in the early 2000s as effective but non-addictive pain relievers, meets all the algorithmic criteria of a breakthrough. Five years after it received FDA approval, researchers demonstrated that Vioxx was associated with serious cardiovascular complications, and the manufacturer voluntarily pulled the drug from the market. Celebrex remains available, although like other NSAIDs, it carries an FDA warning for increased cardiovascular risk. Decision makers who were overly focused on the breakthrough potential of this research might have wasted substantial resources nurturing an area that ultimately proved to be a failure—but so would those who relied on expert opinion at the time. That said, the potential benefits of funding the next mRNA vaccine or cancer immunotherapy years earlier than we might have are large. When we can’t assume the outcome of an action, it’s worth conducting an experiment. Can targeted support for areas that are poised to produce a breakthrough increase the return on scientific investment without cannibalizing future advances? With the necessary administrative infrastructure in place, conducting this experiment would be straightforward. First, identify all the breakthrough signals between 2018 and 2023, select half at random, and commit to a decade of investment. For biomedicine, a six year window should yield about 30 topics with the potential to produce a breakthrough. Providing strong support to half of these might be accomplished for $300 million a year. That’s roughly half of the recent annual budget of the NIH Common Fund or around 20% of the fiscal year 2025 budget for ARPA-H. After five years, then again at ten, compare to see which group — the one that was actively managed, or the control — produced more breakthroughs, and on what timescale. Although we used biomedicine as an example, the same experiment could just as well be run for physics, materials science, or any other discipline. Any attempt to undertake such an experiment should keep three points top of mind. First, we can predict the topic of future breakthroughs, but the people and vision for how to move the work forward are still important. In practice, this means success requires active program managers with the expertise to develop the science and the authority to do so, including by recruiting investigators. Second, we would be wise to learn from the failures of past attempts to fund transformative research. High among these are funding modestly repackaged work, confusing the genuinely novel with the merely unfamiliar, and relying too much on incumbent investigators. All of these risks can be mitigated by thoroughly and intelligently integrating high quality data into portfolio management. Finally, we need to have patience. Even with strong funding support, producing a breakthrough takes time. If supporting breakthroughs is the best way to nurture the birth of new fields, we also need to accurately describe and properly manage the other two stages, so that we invest wisely throughout the productive lifetime of ideas and divest once they reach the point of diminishing returns. We would benefit from more study of all of these phases, especially the transition from one to another. More research could answer questions like: When does a breakthrough become an established field? How do the availability of funding, the size and characteristics of the available workforce, and the accumulation of evidence that disagrees with prevailing models affect a field’s decline? Exactly how do old ideas give birth to new ones, and can we speed up the process? The more we understand about the dynamics of scientific progress, the better we can maximize the odds that our curiosity leads to results. I’d call that finding a way to make our own luck.
How New Hampshire Motor Speedway Delivers Maximum Entertainment for the Price
New Hampshire Motor Speedway delivers more than just races — it's an immersive NASCAR weekend featuring multiple race series, live music, fan events and unforgettable atmosphere. Perfect for fans wanting excitement and great value, NHMS provides an action-packed experience that brings motorsports to life at "The Magic Mile."
To age better, we need to rethink health itself
Challenging conventional thinking has historically brought new ways to maximize health and improve quality of life while ageing. Here's what's needed now.
MASTER CLASS: Tap in to the power within to make healthy choices
Intention is a very powerful psychological attribute contributing toward any behavior change. Without an intent to change, there's almost zero chance it will occur – especially with respect to health behaviors. This week, I'll present some tips for maximizing intention in an effort to make good behaviors "stick." Plus, I'll introduce an exercise that fits perfectly into physical activity program.
Germany’s Deniz Undav: Maximum impact from minimum time
The brace against Ivory Coast puts the German on top of the tournament’s goalscorer list.
Some butterflies live 25 times longer than their relatives. Scientists are unlocking their secrets
A tropical butterfly from the Heliconius tribe has evolved to live a maximum lifespan of 348 days, nearly 25 times longer than its relatives.
Harvard Business Review Brings Back The World’s Dumbest Idea
HBR revives “maximizing shareholder value”—the dumbest idea in the world—treating customers as mere constraints. True value creation for customers, not metrics, wins long-term.
Analyst Perspectives on Oracle’s Strategy to Help Customers Secure Business Data against AI Threats
Oracle today announced its AI security strategy based on three principles: secure at source, secure at speed, and secure through resilience. To help customers accelerate and strengthen their security posture Oracle is making several of its most widely deployed and battle-tested database security, patching, testing, and lifecycle management tools available at no cost for a limited time, or at a 90% discount on one-year term licenses. This move has been positively received by leading industry analysts and reaffirms Oracle’s leadership in equipping enterprises to maximize their resilience in the face of increasing AI-era security threats.
Stephen King shares his thoughts on 'Widow's Bay' and 'Maximum Pleasure Guaranteed'
Two reviews for the price of one.
Is SMIC N+3’s Metal Pitch Smaller than Intel 18A’s?
Almost four years ago, we published that SMIC had started shipping 7 nm (N+1) chips. Now, SMIC is shipping its third-generation 7 nm (N+3) process in Huawei’s Kirin 9030, with a minimum metal pitch of 32.5 nm, about 10% tighter than the 36 nm minimum metal pitch shipping in Intel’s latest Panther Lake CPUs on 18A. The headline is true, but incomplete cherry picked metric. N+3 reaches the density of TSMC N6 through aggressive DUV multi-patterning and design-technology co-optimization (DTCO), but it pays for that in complexity, efficiency and process control. We found this and more in our reverse engineering and teardown where we cover SMIC’s N3 process technology, Huawei’s packaging, memory, architecture, and more. SemiAnalysis has been building a state-of-the-art teardown lab in Oregon capable of analyzing the world’s most advanced and important chips over the last year and half. We have already generated revenue on advanced datacenter chip teardowns including our recent reverse engineering of a major TSMC customer’s COUPE CPO optical engine + EIC 3D stack. This is the first public report from the SemiAnalysis Teardown Engineering & Evaluation Lab, or STEEL for short. The lab is aggressively scaling up and out and we’re excited to announce it publicly. This is a bit of inconvenient timing for TechInsights as they are private equity owned and currently being sold while having enjoyed virtually no credible competition for decades. This has led to TechInsights underinvesting in CAPEX. SemiAnalysis exceeds TechInsights in revenue despite no venture or private equity ownership and being founded only 6 years ago. Because we have no external investors and are founder led, we move faster, build faster, and we can release client chip teardowns for free regularly, while focusing on datacenter for our major clients. Here’s the first public image from our lab, the HiSilicon Kirin 9030 Pro SoC: This report will detail our teardown of the Kirin 9030 and our findings on SMIC’s N+3 process, the most advanced in China. For comparison, we’ll show our teardown of the MediaTek Helio G99, made on TSMC N6. Through this comparison, we can look at the effect of export controls – SMIC N+3 and TSMC N6 are comparable nodes, but one is heavily export-controlled, the other free to use the West’s most advanced equipment. Here we see both China’s progress and constraints. SMIC N+3 reaches TSMC N6-class logic density, but it requires far more aggressive DUV multi-patterning, so it does not match N6 on process maturity or cost. The Kirin 9030 Pro performs similarly to three-year-old Android flagships, and trails far behind the current flagship SoCs from Apple, Qualcomm, MediaTek, and Samsung. The efficiency gap is even wider. Export controls have not stopped Huawei and SMIC from shipping advanced silicon, but they have forced a different path. Without EUV, SMIC is leaning harder on DUV multi-patterning, DTCO, and increasingly complex integration. The roadmap continues forward through tighter design rules and backside power, but each step adds cost and process risk. Huawei’s τ scaling and LogicFolding show another path: stacking active logic and recovering density through advanced packaging and system-technology co-optimization (STCO). To understand the Kirin 9030, we must first understand Huawei’s SoC history. HiSilicon is Huawei’s chip design arm, responsible for the Kirin smartphone SoCs, Kunpeng server CPUs, Ascend AI accelerators, and switch/router networking silicon. Before export controls, Huawei was TSMC’s largest customer – the only customer on TSMC’s first EUV node, N7+, and among the first on N5, alongside Apple. That ended in late 2020. Huawei switched to Qualcomm SoCs in its flagship smartphones, though export controls limited them to 4G-only variants. In late 2023, Huawei returned to in-house silicon with the Kirin 9000s, a successor to the Kirin 9000, fabricated on SMIC N+2 instead of TSMC N5. In the following years, they released the Kirin 9010 and 9020 on the same N+2 process. These chips used Huawei’s in-house TaiShan CPU cores and Maleoon GPU. We have not torn down a Kirin 9020 ourselves, so the predecessor die shot is from Kurnal. The die shots show how Huawei spent its silicon budget: which functional blocks are where, and how their areas compare to the predecessor. First, a quick guide to the major blocks on the die. The total die area is nearly identical, but the 9030 uses that area more aggressively. A denser process lets Huawei fit an extra middle CPU core, more GPU and NPU cores, and larger caches into the same footprint. In contrast, the Helio G99 is a much smaller, low-cost SoC, built for budget smartphones rather than a flagship device. While the Kirin 9030 is ~140 mm², the G99 is only ~29 mm², roughly one-fifth the area. The underlying TSMC process technology, however, is directly comparable as a baseline for analyzing SMIC’s. The Kirin 9030 is an evolutionary refresh, not a clean-sheet design. Its CPU, GPU and NPU cores carry over the 9020’s families, and the gains come from three levers: the SMIC N+2-to-N+3 process step, DTCO and floorplan work, and incremental microarchitecture. Area is where the first two show up, and the 9030 scales well here. Performance and efficiency are the harder test. Huawei’s design holds up better than its node would suggest, but the chip still trails, both because N+3 sits behind the leading-edge nodes and because its cores, while competent, remain a few generations behind the newest designs. The new prime core is an incremental update. The main changes are a 10% frequency increase from 2.5 GHz to 2.75 GHz and a doubling in the L2 cache from 1 MiB to 2 MiB. Despite the increased cache, the core size decreased by 7.6%. Excluding the private L2 cache, the core size decreased by 21%. This is a large reduction for an incremental node. Compared with the TaiShan New V120 core in the Kirin 9020, the Kirin 9030’s middle core is almost unchanged architecturally, yet each core shrinks by ~22%. Most of that comes from the move from N+2 to N+3, with layout likely accounting for the rest. Visually, the most noticeable change is the increase from 3 middle cores to 4. There is also a 20% increase in the shared L3 cache of the big cluster. This helps to improve multi-core performance without sacrificing much in terms of area. Even with each core shrinking, the big CPU cluster’s total area is essentially unchanged. The per-core savings went back into an additional middle core and larger caches. The tiny cores shrank less than the prime core (excluding its L2 cache) and less than the middle cores. This is likely because fixed overhead is a larger share on a small core. We cannot resolve any architectural changes from the die shot alone, but the per-clock and efficiency gains shown below point to more than pure process and layout scaling. The area reduction was offset by a doubling of the shared L2 cache from 2 MiB to 4 MiB, leaving the total tiny CPU cluster area slightly larger. Area is the easiest improvement to see from a die shot, but it is only one part of PPA (power, performance, area). For modern logic, power and performance matter just as much, and often more. Since Dennard scaling broke down in the mid-2000s, voltage and frequency have not scaled in step with transistor dimensions, so each node has had to fight harder for gains in performance and efficiency. The starkest comparison is not Kirin 9020 versus Kirin 9030 Pro. Apple’s efficiency cores run circles around Huawei’s prime core. Apple’s low-power core delivers 20% higher integer performance while drawing only 1 W, compared with 4.5 W for Huawei’s prime core. N+3 matches TSMC N6, but N6 is several generations old. Apple and Qualcomm build on N4 and N3P, which are denser and sit on a better voltage-frequency curve, giving them a larger transistor budget and more performance per watt. The 9030’s own cores did improve. The middle and tiny cores gained 17% and 14% in per-clock integer performance over the 9020, with floating-point flat on the middle core and up 11% on the tiny. The tiny core improves cleanly, with performance rising while power falls and efficiency increases by 45% in integer and 24% in floating point. The middle core is mixed: integer performance rises but power rises faster, cutting integer efficiency by 7%, while lower power lifts floating-point efficiency 16%. Per-clock gains at the same or lower frequency are microarchitectural, so the cores are tuned, not just shrunk. Both also failed to hold their rated maximum frequencies, pointing to thermal, power, or stability limits. Per clock, the middle core sits around Arm Cortex-A720 and the tiny core near the Cortex-A520; absolute performance trails because Huawei clocks them much lower. The prime core is roughly Cortex-X2 class per clock, a 2021 design. Apple’s 2020 M1 Firestorm core is still 35% higher per clock and 57% faster in absolute integer performance at a similar 4.5 W. The current leading edge is further ahead again: the Apple M5 P-core is 60% higher per clock and 2.7× faster, the Arm C1 Ultra 45% higher and 2× faster. Matching older high-end cores per clock is a genuine design achievement. What Huawei cannot match is the voltage-frequency curve and transistor budget of leading-edge nodes, which let Apple, Qualcomm and others spend more transistors in the same area on wider cores, larger caches and deeper buffers while running at lower voltage. Huawei’s LogicFolding roadmap is one answer, stacking active logic to recover density and shorten signal paths. We return to it later. The GPU compute units (CUs) changed more visibly than the CPU cores, moving to a more rectangular layout for both the arithmetic logic unit (ALU) clusters and the CU overall. Even with ray-tracing support added, a CU shrank ~28%. However, that shrink is offset by the increase from 4 to 6 CUs and the area outside the CUs grew 33%. Overall, the GPU cluster is larger by ~10%. The GPU is where Huawei makes its biggest gains. The Maleoon 935 is not competitive with current flagships, but it is a large step up from the 920 and reaches older-flagship territory. In 3DMark it is 70% faster in Wild Life Extreme (WLE) and 79% faster in Steel Nomad Light (SNL) than the 920; with 11% higher clocks and 50% more CUs, the ~67% theoretical uplift roughly matches WLE and is beaten by SNL. It edges ahead of the Snapdragon 8+ Gen 1 in WLE and SNL, and the Dimensity 9200 and Apple A16 in WLE, but stays far behind newer parts: the Snapdragon 8 Elite Gen 5 and Dimensity 9500 are ~2.4–2.6× faster in WLE and ~3.2× faster in SNL. The Maleoon 935 is Huawei’s first GPU with hardware-accelerated ray tracing; there it lands slightly ahead of the Exynos 2200, and on par with the Apple A16, with current flagships up to 3.7× faster. The Neural Processing Unit (NPU) saw the largest structural changes of any block, moving from a Lite and a Tiny core in the Kirin 9020 to a Lite and two Tiny cores in the Kirin 9030. Both core types also show significant layout changes. This is a reversal in Huawei’s NPU design. The Kirin 9000 5G, its last flagship chip on TSMC N5, used two Lite and one Tiny core. The series of SoCs on SMIC N+2 moved to one Lite and one Tiny core, likely for area savings. With the Kirin 9030, Huawei has shifted back toward a larger multi-core NPU cluster, but with the additional area going to a Tiny core rather than a Lite core. We’re diving deep into the most advanced datacenter and AI hardware hitting the market. To learn more about what’s in the pipeline or to commission a custom teardown, contact sales@semianalysis.com. Interested in joining us on this ride and think you can be a difference maker? Check out our Careers page. Before diving into the process stack, the package and memory are worth separating from the SoC itself. The Pro variant of the Kirin 9030 carries 12 GB of Samsung DRAM, with two stacks of four dies each. The dies were identified as the K4L2E165YD, a 12 Gb LPDDR5X-9600 device fabricated on Samsung’s 1a node, the fourth generation of its 10 nm-class DRAM after 1x, 1y and 1z. 1a has shipped in volume since 2022, so this is current memory rather than older-node inventory. The 16 GB Pro Max variants we obtained were found with both CXMT and Samsung packages. The CXMT package is marked CXDD7JEDM, with two stacks of four dies, packaged in week 45 of 2025. The inferred die dimensions from X-ray computed tomography (CT) are consistent with a known density of ~0.3 Gib/mm² for the CXMT G4 process, roughly equivalent to other manufacturers’ 1z processes. The Kirin 9030 uses a typical integrated package-on-package (iPoP) stack: multiple DRAM dies in a memory package sit above an organic redistribution layer (RDL) interposer, which sits above the SoC and package substrate. The full package is then mounted to the printed circuit board (PCB) through ball-grid array (BGA) solder bumps. The memory package substrate is a thin bismaleimide-triazine (BT) laminate carrying the LPDDR5X stack. The organic RDL interposer over the SoC routes the PoP signals around the die and carries possible dummy thermal copper pillars. The package substrate, a thicker Ajinomoto Build-up Film (ABF) build-up over a BT core, fans the flip-chip bumps out to BGA pitch and embeds the power planes. The whole stack is organic. The only silicon is the SoC and the LPDDR5X dies; there is no silicon interposer. Keeping it all-organic brings the package’s coefficient of thermal expansion (CTE) close to the PCB’s, reducing board-level warpage, and avoids the cost of a silicon interposer the SoC’s bandwidth does not need. In an iPoP stack, the memory package connects to the organic RDL interposer through an array of solder bumps. Underfill fills the gap around those bumps, adding stiffness and protecting the joints from mechanical stress. The Pro and Pro Max variants differ here, which we cover behind the paywall. The die shot and architecture tell us how Huawei allocated its silicon budget. The process tells us what SMIC can manufacture. We use the Helio G99 as the process reference for TSMC N6. Both SMIC N+3 and TSMC N6 are evolutions of previous 7 nm-class nodes. We used targeted TEM cross-sections through logic and memory regions, imaged in both fin-cut and gate-cut directions. Each cross-section caption gives its horizontal field width (HFW), the real width of the imaged area. We start at the transistor fins, then move up through standard cells, local interconnect, and SRAM. SMIC has not overtaken Intel or TSMC. It uses aggressive DUV scaling and DTCO to reach N6-class density, but that density doesn’t translate into comparable performance and efficiency, for two reasons: the node gap to leading-edge nodes, and Huawei’s core designs. Fin Profile One of the most important knobs in a FinFET process is the fin profile: the shape of an individual fin and the channel where current passes from source to drain. The ideal fin is tall, narrow, and nearly vertical. A taller fin increases effective channel width, while a narrower fin improves electrostatic control by thinning the body the gate must control. Push either too far, and the process pays for it: weaker drive current, fragile fins, taper, footing and line-edge variation that hit yield and device variability. The Intel 22 nm, 14 nm, and 10 nm fin cross-sections show how FinFET nodes have improved over time. 22 nm fins were a first-generation structure, relatively short, wide and strongly tapered. The shape limits current density and reduces gate control uniformity across the height of the fin. At 14 nm and 10 nm, Intel pushed the fins taller and narrower while also making the sidewalls more vertical. Rather than shrinking the device, these changes increase the effective channel width per fin and improve electrostatic control. The trade-off is that taller fins at tighter pitches are much more difficult to manufacture. Now, let’s compare the Helio G99 on TSMC N6 with the Kirin 9030 on SMIC N+3. Both processes are in the same class, with fin pitch of 30-32 nm on N+3 and 34 nm in our N6 cross-section. The pitch for N6 is especially interesting as N7’s HD library is generally listed with a 33 nm fin pitch, and N6 did not shrink pitches directly. Its density gains came from DTCO instead of tighter pitches. The 34 nm pitch was stable across our sampled region and serves more as a comparison against the SMIC N+3 we have not investigated further. Pinning down N+3’s fin patterning scheme takes more than one core unit. The CPU cores show a dense ~32 nm pitch, with the pitch between N-P fin pairs alternating between 78 and 88 nm. Logic alone may be consistent with dual-pitch mandrels of 120 and 110 nm, but this is a complex and unusual approach. Combining the pitch from the the 8T SRAM, which has more complex repeat unit, with the CPU core sequence allows us to reverse engineer the patterning steps with more confidence. As both the logic and SRAM should share the same base grid, a single CD mandrel lithography pattern with 128 nm pitch undergoing SAQP produces a die-wide ~32 nm grid (128 nm/4), which supports the pitch sequencing seen in both logic and SRAM cells. In the sampled cross-sections, N+3 shows a taller, narrower, higher-aspect-ratio fin than N6. The measured fin aspect ratio is ~9.5:1 on N+3 versus 7.8:1 on N6. N+3 also shows less top rounding, with an estimated radius of ~2 nm, compared with 2.8 nm on N6. Even though the fin widths differ, the ratio of top rounding to fin width tells the same story, with N+3 at 0.37 and N6 at 0.44. In a geometric sense, lower is better; a perfectly rectangular fin would have no top-rounding penalty. These are single-digit-nanometer features measured from a handful of cuts, so treat the absolute numbers as approximate. The important result is the relative gap: N+3’s fins are consistently taller, narrower and less rounded than N6’s. We’re diving deep into the most advanced datacenter and AI hardware hitting the market. To learn more about what’s in the pipeline or to commission a custom teardown, contact sales@semianalysis.com. Interested in joining us on this ride and think you can be a difference maker? Check out our Careers page. Standard Cell A standard cell is the basic building block of chip layout: a fixed-height row pairing one NMOS and one PMOS transistor that share a gate, tiled in a grid to build logic blocks. The key dimensions are contacted gate pitch (CGP), cell height (CH), fin count, and the lower-metal routing grid. To measure density, we use the Bohr metric: a weighted average of NAND2 gate area (60%) and scan flip-flop area (40%). This represents a realistic mix of combinational and sequential logic. This metric has its limitations, especially for complex cell layouts like TSMC’s FinFLEX, which alternates cells with different fin counts. Even so, it is the best metric for a pure process-level comparison. Another important measurement is the fin pitch; it refers to the distance between two fins of the same transistor. In a FinFET process, multiple fins are used in each transistor to increase the drive current and thus performance. TSMC N6 ships both a high-density (HD) library with 2 PMOS and 2 NMOS fins per cell, and a high-performance (HP) library with 3 of each. More fins under the shared gate mean more effective channel width. HP cells switch harder at the cost of area. Designers mix the two on a die, primarily spending HP cells on timing-critical paths, and matching their PPA targets. In the Cortex-A55 core of the Helio G99, we found a cell height of 240 nm for the HD cell. MediaTek has used HD cells in the G99 to minimize die size and thus cost. As an SoC for budget smartphones at ~$100, this is essential. By contrast, we found only one library in the Kirin 9030, with 2 NMOS and 2 PMOS fins. This suggests a narrower library strategy than TSMC N6, where both HD and HP libraries are widely used. This likely reflects the smaller customer base and the more constrained domestic design and electronic design automation (EDA) ecosystem. In all three CPU cores of the Kirin 9030, we found cell heights of 228 nm, 5% smaller than on N6. This is also a reduction of 9.5% over SMIC N+2’s cell height of 252 nm. SMIC N+3 and TSMC N6’s HD library both feature a CGP of 57 nm. For SMIC, this is a 9.5% shrink over N+2. In the past, CGP and cell height alone may have been enough to compare transistor density. Now, however, we must consider scaling boosters and DTCO as well. SMIC’s density gain does not come from EUV. It comes from using every available DTCO booster aggressively. First is fin depopulation: reducing the number of NMOS and PMOS fins in each cell. The first FinFET nodes started with 3 or 4 fins for each transistor. SMIC N+3 and TSMC N6 HD both use only 2 fins per transistor, trading drive strength for density. Next is contact over active gate (COAG). By landing the gate contact directly over the active gate, instead of out over the isolation region, the cell height drops. N+3 integrates COAG while N6 does not. Our N+3 gate-cut cross-sections indicate COAG, with the gate contact sitting over the active region, while N6 shows an off-gate contact. Last is single diffusion break (SDB). Diffusion breaks are inserted between cells in the same row to provide electrical isolation, but they also introduce local layout effects (LLE), layout-dependent shifts in electrical characteristics. In the past, a double diffusion break was used, consuming the space of two CGPs. SMIC N+3 and TSMC N6 instead use SDB, saving area but increasing LLE sensitivity. This must be controlled at the process level and accurately modeled in the process design kit (PDK) so that EDA tools can account for it. Overall, SMIC N+3 has a transistor density of 113.4 MTr/mm², slightly above TSMC N6 at 107.7 MTr/mm². Even without EUV, SMIC has achieved density beyond TSMC’s mature N6 node which utilizes EUV. Metal Stack The smallest critical dimension in the teardown is M0; SMIC N+3 uses a 32.5 nm local metal pitch. That is smaller than the 36 nm M0 pitch on Intel 18A in Panther Lake. However, this does not mean that SMIC has a better process than Intel 18A or TSMC N3P. M0 is a local intra-cell routing layer. Its usefulness depends on the full interconnect stack: M1 and M2 pitch, track count, via and line resistance, design rules, mask count, overlay control, and routing flexibility. The 32.5 nm M0 is consistent with self-aligned quadruple patterning (SAQP), whose four-population line-width loading we read coarsely as alternating widths of 21.5 to 24 nm; M1 and M2, at 38 and 40 nm, are consistent with self-aligned double patterning (SADP), a single A/B split. On TSMC N6, M0, M2, and M3 sit at a relaxed ~40 nm and are consistent with SADP-class double patterning, with no need for quadruple patterning. That said, we measure M2 for example at ~43 nm, likely inflated by sparse routing. We do not assign any specific layer to EUV from our cross-sections; the distinction we can draw is double versus quadruple patterning, not lithography wavelength. Transistor-level density in the front-end-of-line (FEOL) sets an upper bound, but the design is ultimately limited by what the interconnect stack can route. The lowest metals are the most important for standard-cell density, but the semi-global and global layers determine how usable that density is at the block and chip level. Two axes are commonly used for chip cross-sections: the fin-cut and the gate-cut. The micrograph above is a fin-cut and shows metals 0 through 3. This axis lets us see the even-numbered metals, with M0 right above the fins. There are two kinds of M0 lines. The first are the power rails; these are wide wires for the VDD and VSS running horizontally at the top and bottom edges of each standard cell. The wide wires measure 55 nm across, more than double the other M0 lines. Their width minimizes resistance and reduces IR drop. The second kind are intra-cell wires, short segments within the cell that connect terminals to M1. These have alternating widths between 21.5 and 24 nm. The M0 pitch is 32.5 nm, a 19% reduction versus N+2 and N6. At this pitch, DUV patterning requires more aggressive multi-patterning, increasing mask count, overlay sensitivity, process complexity and cost. M0 is below what a single DUV-defined spacer (SADP) can resolve, so SMIC cascades a second spacer step (SAQP). The cross-section reflects the cost: the M0 trenches are visibly more re-entrant (narrower at the bottom than the top) than M1 or M2 on the same chip and carry a bright barrier-rich foot where the trench meets the etch-stop layer. That shape is partly the intended damascene profile, as a slightly narrow bottom helps void-free copper fill, but its magnitude at M0 is driven by the tight pitch and the higher trench aspect ratio. Intel 18A supports an M0 pitch of 32 nm, although Panther Lake has only shipped with a looser 36 nm pitch. This is due to Intel’s heavy usage of HP libraries. Among leading-edge nodes, 18A has the loosest M0 pitches due to PowerVia. With power routing moving to the backside, congestion is reduced, and the entire front-side metal stack can be used for signal routing. M2 is the first true inter-cell routing layer. It runs horizontally like M0 but spans across multiple cells to carry block-level signals. The M2 pitch sets the cell’s track height – the number of M2 tracks that fit between the VDD and VSS rails, defining what the library calls a 6-track or 7.5-track cell. This layer is the most important, limiting the routing of entire blocks. SMIC N+3 features a 5.7-track cell. The M2 pitch is 40 nm, a 5% decrease over N+2 and the same as N6. This shrink keeps the pitch at the edge of what is possible with double patterning. Future nodes will need to increase the number of masks for M2 as reducing the number of tracks is much harder due to the limitations in routing. The micrograph above is in the perpendicular direction, the gate-cut, and shows metals 0 through 4. This allows us to see and measure the odd-numbered vertical metal layers. The M1 pitch is 38 nm, 9.5% less than N+2 and 33% less than N6. The M1-to-gate ratio matters because it sets local routing flexibility. N+2 and N+3 use a 3:2 ratio, while N6 uses a 1:1 ratio, explaining the huge differences in M1 pitch. The more M1 lines there are compared with the gates, the more flexibility there is for power and signal crossing within the cell. Routing flexibility enables more complex and better cells. Clean fractional ratios are also preferred as a grid is periodic and improves layouts. The 3:2 ratio gives SMIC more local routing flexibility than a strict 1:1 grid, but it also complicates layout and patterning. This is a DTCO choice, with SMIC increasing process complexity to recover density and routability without EUV. This 3:2 ratio is not very popular in the leading-edge nodes. TSMC has only used it on N7+, the N5 family, and the short-lived N3(B). They have switched back to a 1:1 ratio for N3E. Intel has only used it on the 10 nm/Intel 7 family, with Intel 4, 3 and 18A all using a 1:1 ratio. Samsung is the only one still using a 3:2 ratio at the leading edge, using it in the SF4 and SF3 families. It remains to be seen if SMIC will remain at a 3:2 ratio or move to a 1:1 ratio with its future nodes. The industry is still actively exploring these local-routing ratios. At VLSI 2026, imec will be presenting work on even higher ratios, including a 2:1 scheme that can reduce area by up to 14%. We will be covering the conference in a future newsletter article. Subscribe We’re diving deep into the most advanced datacenter and AI hardware hitting the market. To learn more about what’s in the pipeline or to commission a custom teardown, contact sales@semianalysis.com. Interested in joining us on this ride and think you can be a difference maker? Check out our Careers page. The final local interconnect layer for N+3 is M3, with a pitch of 44 nm. The M3 pitch is the same as on N+2 and 10% larger than on N6. The semi-global layers carry the majority of block-level signal routing. They have a coarser pitch than the lower local layers. On leading-edge nodes, they are designed to sit at the limit of DUV single patterning. M4 through M11 pitches were found divided between 80–82 nm (M4–M6), 128 nm (M7–M10), and 148 nm (M11). Given limited sampling, it is possible these are divided further in dense routing areas. At the top are two giant metal layers, M12 and M13. These have kept the same pitches as N+2 at 1920 nm and 4600 nm respectively. While the lower layers’ pitches are generally fixed by the process and library, the upper layers vary much more in pitch and count, depending on the design. Even two smartphone SoCs on the same process can have wildly different metal stacks. The Helio G99 carries fewer routing layers, reaching coarse metal pitches of 850 nm by M9, while the larger and higher-performance Kirin 9030 keeps fine pitches until M11. SRAM At the leading edge, SRAM is much more difficult to scale than logic. TSMC’s latest nodes have seen little to no bitcell scaling, while logic still has more DTCO levers to pull. While looking for other logic libraries in the GPU compute units, we stumbled upon the SRAM. The most common type of SRAM has 6 transistors (6T), but this cell had 8 transistors (8T) instead. 8T SRAM adds two transistors to form a dedicated read port. Unlike a 6T cell, where reading disturbs the storage, the decoupled read port removes read-disturb, improving read stability and letting the cell be pushed harder for performance. At first glance, the cut looked like an unusual logic library, with each cell row having 3 fins of one polarity and 5 fins of another. The rows also alternated in orientation. Energy-dispersive X-ray spectroscopy (EDS) resolved our confusion. The cut had not landed on the GPU logic, but on the SRAM macro beside it. The unusual fin pattern was due to the SRAM library. We return to EDS in the process flow analysis behind the paywall. SRAM libraries are not like traditional logic libraries. Due to the unequal number of PMOS and NMOS transistors, they require specialized rules and layout libraries. They do not need the flexibility of logic libraries, so they are hyper-optimized for one purpose: dense, reliable memory. The SRAM cell we found is a 1:2:2-2:2 cell. This means there is 1 fin per pull-up (PU) PMOS transistor, and 2 fins per pull-down (PD) and pass-gate (PG) NMOS transistor. These 2 PU, 2 PD and 2 PG transistors would usually form a single 6T high-current cell (HCC). An 8T HCC adds a read-pull-down (RPD) and a read-pass-gate (RPG) NMOS transistor, each with two fins. We measured a cell height of 406 nm, which brings the bitcell size to 0.0463 µm². That is a theoretical peak density of 21.6 Mib/mm². We estimate that a 6T HCC would have a cell height of 292 nm and a size of 0.0337 µm². This is ~12% larger than a 6T HCC on Intel 3 and 4. We also estimate the 6T high-density cell (HDC) to have a cell height of 228 nm and a size of 0.0260 µm². This is coincidentally the same as the logic standard-cell height measured earlier. The estimate puts the cell near Samsung 7LPP/5LPP and slightly below TSMC N7/N6. That is a theoretical peak density of 38.5 Mib/mm². 6T HDC is arguably the most important cell as it is used for the largest caches in a chip, the L3 caches and system-level cache (SLC). Both the Kirin 9020 and 9030 have split the SLC into 4 banks to raise total SLC bandwidth. In the Kirin 9030, the SLC increased from 2 MiB to 3 MiB per bank. Correspondingly, the number of arrays within the bank also increased by 50%, from 16 to 24. Each array can store 128 KiB and forms an orderly pattern on the die shot. From the Kirin 9020 to the Kirin 9030, the area of a 128 KiB SLC array decreased from 0.0477 mm² to 0.0392 mm², an 18% shrink. The achieved density is 25.5 Mib/mm², 66% of the theoretical maximum. While the SLC was quite similar across both chips, the L3 has seen some major changes, particularly in terms of its layout. The total capacity also went up from 10 MiB to 12 MiB. Much like the SLC, the L3 is also split into 4 banks. In the Kirin 9020, an L3 bank consisted of 16× 128 KiB arrays and 16× 32 KiB arrays. However, an L3 bank in the Kirin 9030 instead consists of 48× 64 KiB arrays. In the Kirin 9020 L3, a 128 KiB array was 0.0513 mm² and a 32 KiB array was 0.0154 mm². The size of the 128 KiB array is different on the L3 and SLC as the assist circuitry for the two arrays differs depending on their purpose. In the Kirin 9030 L3, a 64 KiB array is 0.0210 mm². Although not a like-for-like comparison, normalized for capacity, it is 18% smaller than the 9020’s 128 KiB L3 array and 31% smaller than its 32 KiB L3 array. The achieved density is slightly lower than the SLC, at 23.8 Mib/mm², 62% of the theoretical maximum. Unlike the L3 and SLC, the prime cores’ private L2 cache uses a 2-bank design. As the prime cores’ L2 is latency-critical, it likely uses 6T HCC instead of 6T HDC. The 9020 has 16 arrays in each bank while the 9030 has 32. Each array has a capacity of 32 KiB. A 32 KiB array in the L2 shrank from 0.0171 mm² to 0.0142 mm², ~17% smaller. The density is 17.6 Mib/mm², ~59% of the theoretical maximum for 6T HCC. SRAM scaled well from N+2 to N+3, shrinking by ~19%, close to the theoretical logic shrink. The caveat is that N+2’s bitcells were unusually large, bigger than comparable 7 nm-class nodes, so part of the gain is catch-up rather than true scaling. With the insights from STEEL’s teardowns, we will be doing a deep dive into SRAM in a future newsletter article. Subscribe Everything above came out of a single STEEL teardown: die annotation, block-level area analysis and TEM cross-sections through logic and SRAM. We’re diving deep into the most advanced datacenter and AI hardware hitting the market. To learn more about what’s in the pipeline or to commission a custom teardown, contact sales@semianalysis.com. Future Roadmap The same cross-sections that pin down N+3 also show where SMIC can go next. Although N+3 is already close to the practical limits of DUV multi-patterning in several layers, SMIC still has a few scaling levers left. A theoretical N+4 would likely start with cell height. N+3 uses 5 M0 tracks between its power rails. Moving to 4 M0 tracks, as on SMIC N+2 and TSMC N6, could reduce cell height by roughly 15%. The routing grid is only one side of the shrink; the front end also must fit into the smaller cell. One possible FEOL lever is reducing the p-to-n isolation spacing from two diffusion grid units to one. Intel used this scaling booster on Intel 4, and TSMC did so on its N3 family. This path trades layout flexibility for density. Fewer M0 tracks reduce local routing resources, while tighter p-to-n spacing raises integration and design-rule difficulty. M2 is also constrained by the cell height shrink. For SMIC to maintain a ~5.7-track cell, M2 would need to move toward ~35 nm. That would move another layer into SAQP territory. SMIC could also reduce the CGP from 57 nm to 54 nm. Intel reached a similar CGP on Intel 10 nm/Intel 7 without EUV. The local interconnect is also tougher. If SMIC keeps the 3:2 M1-to-gate ratio, M1 would need to shrink to 36 nm and would likely require SAQP as well. If SMIC moves to a 1:1 ratio, M1 could relax to 54 nm, but it would give up routing flexibility. Under this theoretical path, we estimate that SMIC N+4 could reach a cell height of 198 nm and a CGP of 54 nm, implying a Bohr density of 137.8 MTr/mm², on par with TSMC N5 or Samsung SF4. However, the difficulty is cumulative. Each step is individually plausible, but together they make N+4 harder than the transition from N+2 to N+3. It will likely take longer, cost more, and carry less process margin. A theoretical N+5 would require a larger integration shift. One possible path is backside contacts (BSCon), moving power routing and source/drain contacts to the backside, which would reduce front-side routing pressure and enable another reduction in cell height. The front-side metal pitches could relax to reduce the process complexity. M0 would likely relax slightly to ~34 nm, and M2 and M4 pitches could relax further. CGP is unlikely to shrink much further. Even with EUV, 48 nm has been the practical limit for yield and process control. This approach would allow N+5’s cell height to fall to 170 nm and its CGP to 53 nm. This implies a Bohr density of 163.6 MTr/mm², on par with Intel 18A’s HP library. However, this would not make N+5 cost-competitive with the leading edge. It would reach a similar density through a much more expensive route. The integration difficulty rises sharply, with new process flows for backside alignment, wafer thinning, contact reveal, and backside metallization. Past this point, standard density and interconnect scaling become increasingly unattractive. That is where Huawei’s roadmap stops looking like a normal foundry roadmap and starts looking like a packaging roadmap. At ISCAS 2026, Huawei unveiled its tau (τ) scaling law, reframing process scaling in the time domain. τ is the time cost of data movement and processing: switching delays in transistors, RC signal propagation delays in circuits, compute, memory, and networking latency. Outside Huawei’s terminology, this is called system-technology co-optimization. This is Huawei’s answer to its lack of EUV lithography. Without EUV, planar density cannot keep pace with TSMC, Intel, or Samsung. If transistor density cannot shrink further, Huawei’s alternative is to shorten wires, reduce buffering, and stack logic vertically. “LogicFolding”, Huawei’s implementation of this new scaling idea, is, in practice, an aggressive 3D stacking approach. AMD V-Cache places SRAM above or below a CPU die. AMD’s MI350X places active interposer dies (AIDs) underneath accelerator and compute dies (XCDs), with the AIDs handling cache, IO interfaces, the network-on-chip (NoC) and embedded metal-insulator-metal (MIM) capacitors. With LogicFolding, parts of the same logic block are split across multiple active dies bonded face-to-face at ultra-fine pitch. This allows Huawei to shorten some critical paths and reduce buffer overhead, not merely add cache capacity or offload the IO and interconnect. Shortening wires is where the higher clocks come from. A large share of a modern core’s delay and energy budget goes into driving long interconnects and the repeater buffers along them. LogicFolding distributes a block’s critical-path gates across multiple stacked tiers bonded at a very fine pitch, so the bond interface behaves like an additional metal layer and the longest paths get shorter. That is how Huawei expects to recover frequency and efficiency it cannot get from the process alone. Huawei’s roadmap shows its intent. Prime core frequency is targeted to rise from 2.75 GHz in the Kirin 9030 to roughly 5 GHz by 2031, far beyond what planar scaling alone could deliver. Prime cores with 3.1 and 3.39 GHz clocks are being tested in its labs, although their power consumption is unknown. Beyond that, chips are in the design, simulation or pathfinding phase, meaning the frequencies are targets. However, the direction matters more: LogicFolding also helps with performance, not just density. The catch is that Huawei’s density claim is not directly comparable to foundry densities. A stacked design can report more transistors per package footprint by adding active layers, even if each patterned die remains well behind TSMC or Intel in front-end density. This is how Huawei can claim to reach foundry 14A-equivalent density by 2031. This is not a like-for-like foundry comparison, with Huawei using stacked logic and measuring density per package footprint. On a normalized Bohr-density basis, SMIC N+3 is ~114 MTr/mm², 38% less than Intel 18A’s HD library. Huawei’s 3D roadmap closes the gap by stacking active logic, reaching 215 MTr/mm² by 2030. In 2031, the roadmap density jumps to 295 MTr/mm², implying either a third active layer, partial EUV insertion or aggressive planar DUV scaling. Huawei’s methodology makes other foundries look much denser as well. Applying it to AMD’s MI450X with an N2 top die on an N3P base die yields a theoretical density of 460.2 MTr/mm² in 2026, compared with Huawei’s 295 MTr/mm² in 2031. This Kirin 9030 does not use LogicFolding, remaining in a conventional mobile SoC package. Instead, it forms the baseline for how far Huawei and SMIC can push planar scaling. Future teardowns of Kirin and Ascend chips will show both planar logic density and Huawei’s hybrid bonding solutions. Export controls changed China’s optimization problem rather than ending it. EUV restrictions raised the cost and complexity of leading-edge manufacturing without freezing it. SMIC reaches N6-class logic density through DUV immersion, SAQP and DTCO, while Huawei shifts more of the burden onto architecture, packaging and system-level integration. Future nodes will be tougher. N+3 still had room to tighten local metals and reduce cell height and CGP. Further scaling without EUV leaves fewer levers. More aggressive multi-patterning adds masks and overlay error. SMIC can keep pushing DUV, but each step will get more expensive and less forgiving. The design side is just as critical. Huawei had domestic EDA tools and flows before the Kirin 9030, with the Kirin 9000s, 9010 and 9020 making that clear. Huawei was able to ship multiple consumer SoCs on SMIC N+2 and N+3 while cut off from the Western EDA stack. US export controls restricted EDA tools for advanced chips in 2022 but did not target tools for more mature chips. In 2025, the US government briefly placed much broader restrictions on EDA software from Synopsys, Cadence, and others, before lifting them less than two months later as part of a trade deal tied to rare earths. Huawei has been unable to access those tools because it remains on a US trade blacklist. That forced Huawei, SMIC and Chinese academic institutions to build their own tools and flows. Researchers at Peking University recently announced a prototype EDA tool for Huawei’s LogicFolding architecture, which requires a new flow to handle the multilayer layout and floorplan. This is not the same as replacing the full Synopsys or Cadence stack, but it shows where domestic EDA is headed: toward tighter co-optimization between architecture, process and packaging. These advances are also diffusing into the Chinese ecosystem. SMIC is licensing its N+2 and N+3 processes to HLMC/Hua Hong at the government’s direction rather than by choice. If the same process learning feeds into Ascend accelerators for AI training and inference, the choke point shifts from one named fab to an ecosystem. Alibaba’s T-Head silicon arm and Cambricon, a Chinese AI chip designer that is expected to supply ByteDance, could also be major beneficiaries. Sanctions aimed at SMIC alone become less effective once the manufacturing knowledge has spread to other fabs and design houses. China is not closing the gap with Intel, Samsung and TSMC. The teardown shows the opposite in several places: no EUV, no backside power, higher process complexity, and visible trade-offs. But China is still advancing. If domestic chips become good enough for phones, inference, networking and security-sensitive workloads, they can matter strategically without matching TSMC at the leading edge. Behind the paywall, we show what else STEEL can do, with material and process flow analysis of SMIC N+3, and analysis of the Kirin 9030 package. We’re diving deep into the most advanced datacenter and AI hardware hitting the market. To learn more about what’s in the pipeline, access the full Kirin 9030 and SMIC N+3 analysis or to commission a custom teardown, contact sales@semianalysis.com.
New Research Finds You Probably Need to Eat Way More Carbs than You Think Before a Major Workout
Scientists dial in the dose to maximize stored carbohydrates before a race. It’s a lot.
We can create the future of science right now
Paul Litvak is the founder and Executive Director of the Robyn Dawes Institute and a Visiting Scholar at UC Berkeley. He has a PhD in Behavioral Decision Research from Carnegie Mellon where he studied emotions and the sunk cost bias. Over 15 years in industry, he solved a wide range of challenging technical problems. At Meta he created models optimizing the ad review process. At Google he ran experiments to measure social influence. At Airbnb, he was a product manager leading machine learning teams optimizing search ranking and price suggestions. He also co-founded and led product at Rhythmic Health, a venture-backed biosensing startup, creating an accurate low cost system using a color changing strip and a smart phone to measure salivary lactate. The bottleneck At this point, it is uncontroversial to say that science needs to stop using the PDF article as the unit of knowledge and currency. The unbearable slowness of scientific publishing, the profit motive and margins: I’m not saying anything new. The PDF also sucks because it’s hard to extract structured information, which makes it hard to do evidence synthesis. As a result, we do much less evidence synthesis than is needed. And evidence synthesis ultimately undergirds most policy and medical decision-making. I can see second-by-second real-time odds for any sporting or newsworthy event, but a school board can’t see the best evidence on whether their 8th graders should be taught algebra. As a society, we don’t treat this as an important problem. Again, not controversial. Not only is the problem well understood, but the solution has already been laid out. What we need is AI-assisted living evidence synthesis - (1) an open knowledge graph of atomic claims, (2) claims linked to evidence, (3) assessment and synthesis of each piece of evidence, and (4) continuous updating with new data. What few realize (yet) is that the technical capacity to build this vision for a significant portion of science already exists. Not only that, scientists and startup teams are already building many of these components. I know this because I’ve been surveying the space and talking to many of the builders. There are some missing pieces: for example, evaluations of how well some of the components work. But at this point, most of what’s missing is a fully end-to-end working integration of all of these parts. In the rest of this essay, I’m going to lay out all the parts of a working living evidence layer for science and who is working on them, and propose concrete next steps for building this system. What’s now possible The diagram above outlines the components of a living evidence synthesis platform, including some of the teams working on each component1. Scientific PDFs are processed into claims with associated evidence. The evidence is subjected to a forensic audit, methodological evaluation and robustness and reproducibility checks. Finally it’s given a weight in a continuously updating synthesis. What follows is a description and status of each component and a few of the teams working on them. Document understanding The first thing you need to be able to do is turn an article into structured data. Mostly that means parsing PDFs. There are often multicolumn layouts that confuse non-specialized PDF to text processing libraries. For scientific papers, there is the added complexity of parsing formulas and tables and figures. This is a really hot area - there are startups offering APIs, and it seems like a new open source package gets posted to Github every few weeks. What follows isn’t exhaustive. A package called GROBID was the state of the art for a while, they didn’t update their package for nearly two years until very recently. In the meantime reducto.ai released an AI powered PDF extraction API, PaddleOCR became popular, IBM released a model called Docling, and both Mistral and Gemini created models and libraries. I also know of at least one other well-funded psychology research group working on a paper parser. By contrast, there are few open evals in this space, with no extensive evals for complex table comprehension in particular. Nonetheless, I’m confident this will be a solved problem soon, given the combination of LLM advances and developer interest. Hypothesis level extraction There has been increasing interest in comprehending the extracted text of papers and linking information to evidence for each hypothesis. A lot of work has already been done. Trialstreamer (Marshall et al. 2020) and RobotReviewer LIVE (Marshall et al. 2023) demonstrated automated extraction of trial population, intervention, and outcome at scale on clinical RCTs. PaperQA2 (Skarlinski et al. 2024) and Ai2 ScholarQA (2024) extended this to retrieval-augmented question answering with citation grounding. Elicit, Consensus, and SciSpace operationalized claim-level extraction for end users. OpenEval (Booeshaghi et al. 2026) is the most recent and most ambitious: 1.96 million atomic claims extracted from 16,087 eLife manuscripts using Claude Sonnet 4.5, grouped into ~299,000 results, with LLM evaluations showing 81% agreement with human peer review on a 2,487-paper subset. None of these solutions link claims to test-statistics, as you would need to evaluate randomized controlled trials. This is why I built the evidence.guide API - to extract hypotheses and associated test statistics from behavioral science papers. The best public eval of this kind of extraction I’m aware of comes from the recent SCORE project - they had humans code thousands of psychology papers to extract their claims by hand. It would be extremely helpful to the world if all scientific PDFs were available as structured open data. I’ve been working to make this happen, both directly at Berkeley and through coordination with large entities I can’t yet speak of; as hard as it is to do, I think it’s possible2. Forensic audit A lot of work has been done on forensic audit, but some gaps remain. Of course, for biology papers that rely on images for evidence, there are a variety of tools (notably Proofig and ImageTwin) to spot anomalies. These are still well short of what sleuths like Elizabeth Bik can do on her own, but these tools are constant companions among fraud analysts. There’s someone working on auditing Excel files for anomalies, and a number of teams are automating numerical checks like GRIM and SPRITE, including Scrutiny project, the INSPECT-SR team as well as statcheck. The regcheck team is building a way to use AI to compare preregistrations to analyses in papers, to ensure there aren’t significant deviations. Nonetheless, there are many other kinds of anomalies to screen for, both public and less publicly known. And there are no formal evals for anomaly detection that I’m aware of. Still, there’s a lot to draw from in this space and I’m pretty certain we will be able to scan papers for most kinds of obvious anomalies in the near future. Methodological review This area has been white hot, though I fear for many of the startups in this space, because this capability may become commoditized. There are at least six different AI peer review companies, including Refine.ink, Reviewer3, ReviewerZero.ai, Q.E.D. Science, Paper Wizard, and Isitcredible. Coarse (a pun on refine) was also recently created as an open source alternative. These systems provide qualitative feedback on the content of papers, spotting methodological weaknesses and mathematical errors. They seem to work pretty well, and many academics report bitterly that they exceed the average quality of typical peer reviewers. But there are few evals here either. What evals exist so far involve using LLM-as-judge (circularity problems abound) or comparing against human reviews of questionable quality. What you’d ideally want is an eval that measures capturing known errors in papers3. Reproducibility and robustness Another active area has been using AI agents to automate computational reproducibility4 and robustness5 checks in papers that report numerical results. For more recent papers where data and code are available, AI agents can see whether they can re-run the analyses and produce the numbers reported in the published paper. In addition to a handful of individual academics who have been experimenting using Claude Code for this, the Institute for Replication is a leading group working on building an end to end system. The evals related to this problem are the most mature, with CORE-Bench (Siegel, Kapoor, Narayanan 2024) and PaperBench (OpenAI 2025) available to benchmark agents on this task. There is also work on getting AI agents to test alternative ways of analyzing the data to ensure the results are robust to small analytic design choices. Synthesis This is the most underdeveloped area where significant investment is required. Although some automated evidence synthesis systems exist — for example, otto-sr is building an AI agent to write systematic reviews — none of these incorporate the full range of paper level signals to weight evidence appropriately. Nor is there anything like an eval or a gold standard for a good systematic review. Arguably Cochrane reviews are the closest we have to gold standard human systematic reviews, though I’ve heard academics in the know complain about their uneven quality. A key question for a synthesis platform is how to weight anomalies and methodological issues in assessing the quality of a piece of evidence. This is an unsolved problem and one I’m very keen to work on. Continuous updating There are many pieces of basic infrastructure available for monitoring for new research and initiating updates. OpenAlex is the current open citation graph. Retraction Watch integrated into Crossref in October 2023. Scite tracks how citations support, contrast, or mention prior claims. The Living Evidence Network demonstrated continuous-update workflows in clinical guidelines. Engineering this is a relatively straightforward task. When you look over this technical architecture and all the progress being made, it’s hard not to be optimistic that a living guide to scientific evidence will be built. The stakeholders are ready The social infrastructure for this is starting to coalesce — it’s not just a pie in the sky academic exercise to imagine this coming into existence. Institutions like the Center for Open Science, the Institute for Replication, the INSPECT-SR, the Living Evidence Network and more are all working on scaling work to improve research quality. Funders are also aligned. The Sloan Foundation has funded living evidence work through COS. Coefficient Giving supports the Institute for Replication and COS. The Astera Institute and the Institute for Progress have shown interest in this space. NIH has established an Office for Replication and Reproducibility. Although there are (very unfortunately) serious headwinds in science funding generally, there is an active group of funders interested in metascience. A brief word about what I’ve been doing at RDI. First, as a Visiting Scholar at Berkeley I’ve been actively figuring out how a non-profit and a public university can conduct and make public the results of large-scale academic article data mining. With some of the money I raised from donors, I commissioned a legal analysis of recent case law and publisher text data mining (TDM) agreements in order to understand whether a massive open data mining of academic articles is possible (with caveats, it is). I’ve also been working to bring together stakeholders in this space, and identify gaps. I’ve also been doing some software development in this space, with more to come. A pilot proposal The assumption undergirding all of this is that an AI, given all this information, would make the right judgment about a scientific claim with lots of conflicting evidence, weighing all the factors appropriately. That’s the hypothesis we need to test. Randomized control trial research is the best place to focus on first. RCTs are used to make many of the important decisions in society - from medical trials to public policy changes. And they use a relatively uniform set of inferential statistics with lots of known and available diagnostics. Behavioral science experiments, within the broader realm of RCTs, should be first used as a testbed whose results can be generalized. Because behavioral science is at the vanguard of open science practices, replications abound (there are thousands of them) to serve as ground truth training data. Key Hypothesis Therefore the pilot would test, in behavioral RCTs that have been replicated, whether the quality of evidence for a claim can be used to accurately predict whether that claim will replicate. Secondary Hypotheses Compared to claims that replicated, non-replicated claims demonstrate a greater share of forensic anomalies in their source literatures. Hypothesis level claim and statistic extraction is accurate enough to scale living evidence without onerous human review costs. Replication prediction is more accurate than prediction markets6 or journal prestige. If all the different quality signals we gather do accurately predict which studies will replicate, then we can use that model to score evidence to power the living evidence layer. Why this is informative regardless of outcome If the pilot succeeds, the architecture extends to medical RCTs (where Living Evidence already operates and integration is mostly about claim representation), then to slices of basic biology with stable replication structure. If it fails, the field learns which quality signals are load-bearing and which ones metascience has oversold. Either result is a contribution to knowing what the literature supports. Implications for funders Because this burgeoning ecosystem of builders already exists, a well-informed philanthropic or government funder could play a crucial catalyzing role in bringing this future about. They could play at least three roles: creating open structured datasets, publishing open benchmarks and incentivizing the solving of key technical challenges. First, open archives of papers that the government maintains, like PubMed, could be turned into structured data amenable to large scale metascience and claim aggregation. I know the US government already has an interest in doing this, though some key open questions remain unanswered. How do you determine the best models and systems for accurately extracting information from papers? How can you establish a robust way to allow researchers to flag errors and correct them? And finally, how do you create a legal regime in working with publishers to maximize the scope of available papers? For the latter, a university or a private philanthropy may be better positioned to make structured data publicly available under journal subscription terms or fair use. Philanthropists or government funders could also coordinate to create or commission benchmarks that evaluate whether important problems have been solved. For example, an open benchmark for claim extraction from a range of different scientific article types would be extremely helpful. Ensuring the underlying data are accurate is vital for creating these evaluations. I’ve discussed opportunistically using various human-created datasets for this purpose, but a consistent problem is that errors in human data make it difficult for them to serve as a gold standard. Finally, with structured data and benchmarks available, the government or private philanthropy could use them to incentivize groups to develop machine learning systems that meet these benchmarks. Prizes are one potentially valuable tool for this. For example, you could establish a prize for a system that accurately updates a living meta-analysis for a small set of claims. Prizes are particularly useful as signals of problem importance, and can help create vibrant ecosystems of public and private research—see, for example the role the government played in kickstarting the current work on nuclear fusion. Conclusion The drawbacks of the current scientific publishing system are known. Scientists agree, metascientists agree, philanthropists agree: the published PDF plus citation graph isn’t the right substrate for maintaining a representation of the evidence base in science. The pieces needed to build the alternative either already exist or are rapidly taking shape. The community is forming around exactly this problem, with concrete partnerships and shared infrastructure. A pilot should start on behavioral science RCTs because that’s the slice of empirical science most amenable to legibility, where replication ground truth is richest, and where the failure modes are best documented. What’s been missing is the galvanizing mission to assemble these pieces into something that works. That’s what I’m proposing to build. *** 1 I have a broader field map that I’ll release publicly soon. This is me, building in public! 2 If this is something that you are excited about, please reach out and talk to me. 3 More on this very soon too! 4 This tests whether, given the code and the data, you can get the same statistics as reported in the published paper. 5 This tests whether the results are the same as a paper’s given alternative analytical decisions in conducting the analysis (like outlier omission). Closely related is the idea of a “multiverse” where you come up with many different ways of answering the same underlying research question with the same data, and test whether the results hold in all those alternative methods. There’s been work on the latter as well. 6 Some of the replications, e.g. those from the SCORE project, had paired the experiments with forecasts from prediction markets. So we get to look at this for free.
GE HealthCare gets FDA OK for radiation treatment AI tool
GE HealthCare is putting its MIM imaging software to work to equip oncologists with more personalized planning advice to maximize, and minimize, radiation treatment.
BOSS Summit Miami Helps Owners Maximize Business Exit Value
BOSS Summit Miami brought business owners together to learn exit planning, tax strategies, wealth preservation, and M&A.
Piecework’s jigsaw puzzles are making ‘grandma hobbies’ cool
From maximalist pizzas to retro beaches, Piecework brings canvases with a coffee-table book flair to gen Z’s homes
Trinity School of Medical Laboratory Science achieves maximum accreditation
STEUBENVILLE — Trinity Health System’s School of Medical Laboratory Science Program has been awarded the maximum 10-year accreditation by the National Accrediting Agency for Clinical Laboratory Sciences. As the premier agency for ensuring the highest standards of education in clinical laboratory sciences, NAACLS grants this prestigious 10-year status only to programs that demonstrate exceptional educational […]
School of Medical Laboratory Science achieves maximum accreditation
STEUBENVILLE — Trinity Health System’s School of Medical Laboratory Science Program has been awarded the maximum 10-year accreditation by the National Accrediting Agency for Clinical Laboratory Sciences. As the premier agency for ensuring the highest standards of education in clinical laboratory sciences, NAACLS grants this prestigious 10-year status only to programs that demonstrate exceptional educational […]
NC State Health Plan board approves Medicare Advantage plan benefit changes despite retiree concerns
Retirees on the state's Medicare Advantage will face higher out-of-pocket maximums next year, as well as higher copays for many medical services.
Official | Allan Saint-Maximin leaves RC Lens
Allan Saint-Maximin’s (29) exit from RC Lens has been confirmed. The Frenchman is expected to make a move to the MLS.Having joined Lens last February following the termination of his contract with M...
FIFA's World Cup Ticket Strategy Is Extreme Version of Wider Approach
FIFA's World Cup ticket strategy has drawn criticism. It's a more extreme version of a profit-maximization approach that is becoming common in sports.
Microsoft is Rolling Out Windows 11 Performance Boosting Update
If readers recall, Microsoft was preparing a "Low Latency Profile" feature as part of the Windows K2 effort, which boosts the system's CPU to its maximum frequency for 1-3 seconds to ensure smoother app launches. Today, Microsoft is gradually rolling out this feature as part of the KB5089573 update ...
A battery-powered Starlink Mini is likely on the way
Maximum portability from Starlink’s tiniest dish.
Ubiquiti patches three max severity UniFi OS vulnerabilities
Ubiquiti has released security updates to patch three maximum severity vulnerabilities in Unify OS that can be exploited by remote attackers without privileges.
SBA To Double To $10 Million Maximum Loans For Some Small Businesses
The change, taking effect on July 4th, will benefit businesses that need expensive equipment or real estate.
Xreal's Project Aura smartglasses are a maximalist take on Android XR
At Google I/O 2026, we took a closer look at Xreal's Project Aura, one of the first Android XR smart glasses.
West Texas A&M Partners with PlayFly Sports on Sponsorship and Ticketing
West Texas A&M today announced a five-year partnership with Playfly Sports, the sports industry's leading revenue maximization company for its ticketing and sponsorship sales across all sports.
Whoa! Netflix is Sending Narnia Heading to Theaters...What This Means for the Future of Film
(Welcome to the “Most Important Story of the Week”, my bi-weekly strategy column analyzing the most important (but often not buzziest) news story of the last two weeks. I’m the Entertainment Strategy Guy, a former streaming executive who now analyzes business strategy in the entertainment industry. Please subscribe.) Here’s a funny thing about the debate over whether streamers should release their movies in theaters first: Both sides think the other side represents the “conventional wisdom”. If you’ve been reading me for any amount of time, or even just scanned my front page, you know I think movies make more money by having a robust theatrical window compared to going straight-to-streaming. I’ve published thousands of words on this topic. And for years, I felt like I was standing up to a wave of opposition. See, while the “put your films in theaters to make money” take was obvious at any point before 2019, post-streaming revolution, most of the trendy thinkers took the opposite point of view. Big names argued that Netflix putting their film straight-to-streaming (“S-to-S” from here on out) actually made more money. Netflix was “disrupting” theaters, just as they disrupted Blockbuster Video, and since innovation is always good, this too was a great idea. This thinking then leaked into the mainstream business outlets like the Wall Street Journal and Bloomberg. I read many articles whose take boiled down to “Netflix makes tons of money from S-to-S releases” and “theaters were in structural decline” anyways. Covid-19 shutting down theaters and the 2023 strikes shutting down film production seemingly justified this point of view, as the box office struggled to return to 2017-2019 heights. Yet I’ve also read folks arguing the opposite point: the conventional wisdom says to put films in theaters! From some writers’ points of view, “everyone” actually agrees/agreed with my “send your films to theaters” take—especially the stodgy traditional studios—and folks still didn’t understand Netflix’s genius. I feel like, for the last two years, many people have argued this point as if it’s conventional wisdom. You know what? I may have been wrong. At this point, the “S-to-S strategy makes money” might be the minority viewpoint. Especially with the news this month that Netflix will put at least one film—Greta Gerwig’s Narnia film The Magician’s Nephew—in theaters in 2027 for a full 49-day run is big news. So that’s the story of the week, but I’ll also explain why they’re making this move now and why Netflix may not completely embrace theaters, along with other topics, like how I don’t buy Disney’s “super app” strategy, the latest good news out of the unions, my take on the latest attacks against free speech, and a whole lot more. Let’s get started. Subscribe Most Important Story of the Week - Netflix Goes to Theaters! Listen, I promise not to victory lap too hard today. As I just wrote, I’ve argued against sending films straight-to-streaming for—checks calendar—eight years now. Since 2018. Yikes. That’s a long time. And for years, others painted my thinking as “conventional” in that I was standing up for the status quo against trendy, cool disruption. And that’s not cool! But math is math, and I never could make the S-to-S numbers work. The only thing that ever enabled that model was Wall Street’s brief exuberance for it. It seems like the heads of Netflix now agree with me. (There were always rumors that former head of film for Netflix Scott Stuber argued internally for theatrical releases as well.) As they announced: “The Magician’s Nephew_—the first film in a Narnia series of films—will open on 12-Feb-2027, with a 49 day theatrical exclusive run.”_ I’m a little surprised at how little coverage this big change actually got. Credit to Bloomberg for leading with it and Matt Belloni’s The Town for highlighting it with the provocatively titled podcast “Hell Hath Frozen Over”. But after that? Instead of getting as much coverage as, say, the FCC’s latest egregious crackdown on free speech, this news story got crickets. But it’s massively more important for the future business of entertainment. So let’s start with the obvious question… Why Put Your Movies in Theaters? So…why should Netflix put (some!) of their movies in theaters? Well, I’ve already written that in a very, very, very long article from 2022: [ ](https://entertainment.substack.com/p/the-data-is-in-theatrical-films-massively) [ The Data Is In: Theatrical Films Massively Outperform Straight-To-Streaming Films ](https://entertainment.substack.com/p/the-data-is-in-theatrical-films-massively) Entertainment Strategy Guy · April 5, 2023 [ Read full story ](https://entertainment.substack.com/p/the-data-is-in-theatrical-films-massively) To summarize, theatrical runs make much more money, especially blockbusters, because they earn more money per viewing,1 and that helps boost home entertainment sales. Theatrical films also perform as well or better than the streaming-only films. Now, partially this is driven by increased marketing for said films, but said marketing ultimately boosts the awareness of the movies and drives improved streaming and library performance as well. Again, I’ve made this case after reviewing the economic models and streaming data for years. Just last week, Emily Horgan also made the case for kids films! Yes, it is probably conventional wisdom, but that’s also because it’s true! I’d also argue we’ve seen a lot of data in the various studios’ behavior. As I’ve noted before, the more a studio needs to make money, the likelier they are to put films in theaters. As such, Disney, Universal, Warner Bros., Paramount and Sony all send their major films to theaters. Even if they once made a big S-to-S push (*cough* Disney *cough*), they’ve since relented and returned to theaters. The tech companies don’t need to make money, per se, so it took them years to embrace theaters. Apple doesn’t need to make money because it doesn’t disclose their (likely) Quibi-sized losses each year. (They fear bad press more than losing money.) Amazon announced two years ago that they planned to make Amazon-MGM Studios/Prime Video profitable on its own, and guess what? Now they’re putting films in theaters. That leaves Netflix. They’re in between the “no need to make money at all” and the “desperately need to make money to offset linear TV declines”, so they had less pressure to put some films in theaters. Plus, for years, they were the sexy disruptor, and Wall Street rewarded them for that. (Honestly, they still do.) Yet, even Netflix likely saw the money they were leaving on the table (I’ve estimated it at an easy billion dollars before) so they changed their course. I heard one pundit argue that, because Netflix already makes so much money, they don’t need to make more money, which sort of goes against everything I’ve ever learned about business and economics. Again, to be super-nuanced, sending “films to theaters” does not mean EVERY film. TV movies have been a thing for decades. Lifetime films, Hallmark holiday films, Disney Channel Original Movies, and so on, don’t need theatrical runs. But films above a certain budget level (say $25 to 50 million) almost certainly need theaters to maximize their revenue. The Scenarios (with a Dose of Skepticism) Now time for some skepticism: it won’t surprise me at all if Netflix re-changes course. Specifically, Netflix announced that their Narnia film will go to theaters. And I’d bet anything we see KPop Demon Hunters 2: The Huntering2 also goes to theaters, cause it’s the surest hit outside of a Disney animated sequel right now. But how much does Netflix truly commit to theaters? They’ve already tried to emphasize that this is a one-off move. I see three broad scenarios:
China Politics
China's Foreign Minister Wang Yi arrives alongside foreign diplomats for the presentation ceremony of the ministry's Outstanding Diplomat Medal in Beijing, China, Monday, May 18, 2026. (Maxim Shemetov/Pool Photo via
How New Hampshire Motor Speedway Delivers Maximum Entertainment for Maximum Value
Experience more than just racing at New Hampshire Motor Speedway’s NASCAR weekend. Beyond exhilarating races at "The Magic Mile," enjoy live music, the Fan Zone, camping and more. NHMS focuses on affordability and fan value, offering a lively, all-day motorsports celebration packed with action and entertainment for the whole family.
Lamborghini Fenomeno Roadster (2027)
Lamborghini Fenomeno Roadster features a completely new aerodynamic package compared to the Coupe, designed to ensure maximum performance even without the roof.
Trump demands Hakeem Jeffries be charged with inciting violence with 'maximum warfare' rhetoric
President Trump accused Hakeem Jeffries of inciting violence, tying the Democrat's "maximum warfare" rhetoric to the recent assassination scare.
"This Is Fine" Dog Fights For Therapy In Official Metroidvania
In This Is Fine: Maximum Cope, help Question Hound overcome his anxieties in this nightmare-inspired video game debut
‘The reality, for better or worse’: Columbia comp sci students and faculty grapple with AI’s disruption of the field
‘The reality, for better or worse’: Columbia comp sci students and faculty grapple with AI’s disruption of the field ‘The reality, for better or worse’: Columbia comp sci students and faculty grapple with AI’s disruption of the field A field that once seemed like a direct path to stable, lucrative work is becoming less certain under AI, students and faculty told Spectator. By Ria Vasishtha and Arjun Menon May 3, 2026 Shanying Liu / Deputy Illustrations Editor Asia Genawi, SEAS ’29, first encountered artificial intelligence during her sophomore year at her high school in Indiana. Back in 2023, policies on AI varied widely—even within the same district—and some schools had no guidance at all. As Genawi began to see the effects of AI trickle into her classroom, she turned to guidance from her home state’s Department of Education, which she said she found “outdated and inaccessible.” Genawi spent the rest of her high school career dedicated to the issue, including writing a 15-page memo exploring varying AI policies across Indiana schools, which she sent to every Indiana state senator and representative whose contact information she was able to find. She later helped draft a bill that would have required public high schools to establish and share their own AI policies with students in an accessible format. But with so many other bills already awaiting referral, hers never made it to committee. When Genawi arrived at Columbia to study computer science, she found that the gaps she had spent years trying to address in Indiana had followed her there. Many students and professors were also grappling with what AI meant for the field. Across Columbia’s undergraduate schools, computer science has become the most popular major—accounting for around 11 percent of degrees awarded in Columbia College, 32 percent in the School of Engineering and Applied Science, and 12 percent in the School of General Studies in 2024. While the number of computer science majors in SEAS grew from 166 to 173 during the 2024-25 academic year, the total number across the three schools fell around 4.5 percent from 393 to 375, driven by declines at Columbia College and General Studies. At Barnard, it has grown to become the second most popular major as of 2025, up from third the year prior. The number of degrees awarded have grown fivefold since 2016. Faculty members and students in the department told Spectator that the surge in the major’s popularity has been driven in large part by its apparent promise of a secure, well-paid career path. However, they noted that the recent rise of generative AI has compromised the sense of stability that once defined the field, even as interest in technical skills among students persists. According to data from the Federal Reserve Bank of New York in 2024, out of 74 majors tracked, computer science had the fifth highest unemployment rate among majors at 7 percent, while computer engineering had the second highest at 7.8 percent. “Anecdotally, people are moving away from CS as a major,” Daniel Bauer, a senior lecturer in the computer science department, said. “They’re still taking some of our classes, right, because they think they need that, but they’re instead majoring in other things.” Bauer described a field that had shifted from what he called a “nerdy outlier discipline” to one that “everyone needs,” a sign that technical skills have become a baseline expectation across industries with the emergence of AI. “A lot of people thought, ‘Oh, it’s a guaranteed path to a stable income. You get a six-figure job right out of your undergrad,’” Bauer said. Now, companies may see AI as a way to reduce demand for entry-level programmers, especially in what he calls “code monkey jobs,” roles where a large number of employees are “just writing out code nine-to-five.” Chris Murphy, a senior lecturer in the computer science department, described the current moment as the convergence of two forces: the downturn of the job market and generative AI coming onto the scene. “Things kind of started to really pick up in 2014, 16,” he said, recalling the years after the weaker 2010 market. Murphy added that this was also a period where “students were getting more experience outside of class than they were inside of class” through internships, jobs, and side projects. Murphy suggested that the earlier momentum helped make computer science feel like both an academic program and a reliable pathway into work. That feeling is harder to sustain now as the industry reconfigures, which is a shift that students are seeing firsthand. Rebecca Yu, SEAS ’27, said that during a recent internship in Big Tech, she saw that “every single tech company was racing to integrate” large language models. She added that her manager said her intern class was the first to complete their assigned projects, which she attributed to the increased efficiency that comes with AI. Frank Liu, SEAS ’29, said that with AI lowering the barrier to coding, companies now expect everyone to write code. “A lot of companies outright say if you’re not using coding in your job, you might get fired for not being productive enough,” Liu said. Yu similarly expressed that students increasingly have to accept AI being a part of industry workflow, adding that many also “make use of it to maximize their productivity.” Richard Li, CC ’28, described a “franticness” among his peers as the pace of change accelerates. He explained how “some new tool” is released regularly, noting that tools rise and fall in prominence within weeks. He has come to believe that he doesn’t need to “match that speed all the time” and keep up with every development. Li has learned that AI tools allow him to “take a breath” and “get back into the loop much easier than before” when he needs to. While some companies specifically target Columbia when hiring, sending representatives to campus career fairs and recruiting large numbers from Columbia, a computer science degree cannot be treated as a static credential, Yu described. “You can’t just get your degree and stop learning after that. You have to always be learning,” Yu said. “That is something you basically have to expect for your job.” AI has not only affected students’ decisions to major in computer science, but also reshaped how the subject is taught. Across departments, faculty members no longer treat work done outside the classroom as a reliable measure of understanding, shifting instead toward in-person exams, quizzes, and attendance checks. While homework once made up as much as 60 to 70 percent of a student’s grade in computer science classes, Murphy said that breakdown has since flipped, with exams and quizzes now accounting for the majority. “The assumption—especially among pessimists—is that they’re using it to do their homework,” Murphy said. “They take the homework assignment. They put it into generative AI. Out comes the solution. Submit solution. The end.” Murphy said that faculty members ranged in their responses from fully prohibiting AI use to actively teaching students how to effectively work with AI. He said he shifted his approach this year after his teaching assistants called his belief that students weren’t using the technology “naïve,” and now permits students to use AI on assignments. His surveys of students in his introductory classes have told a more complicated story than the one so-called pessimists offer: Around 60 to 70 percent of students who reported using AI said they used it to have concepts explained to them, to clarify assignment instructions, to generate test cases, or to create practice questions. “Those are all ways that you might have had a tutor in the past to do that, and now you have this AI essentially as your tutor,” he said. Bauer has taken a different approach, teaching students how to use AI productively without sacrificing the core principles of the discipline. “People who are trained to actually use AI tools productively, while being in charge of designing the overall project, will be very much in demand in the next couple of years,” he said. Over the past year, a working group of computer science faculty has been redesigning the introductory and intermediate programming sequence to integrate generative AI as both a subject and a tool, representatives from SEAS wrote in a statement to Spectator. The effort is “coordinated” across the department, “not course-by-course,” with a systematic assessment of how AI integration is affecting student outcomes underway to shape the next round of changes. The goal, the department explained, is for students to be able to “read, verify, and reason about code, whether they or a model wrote it.” Tony Dear, senior lecturer in the computer science department, is piloting individualized AI feedback in a discrete mathematics course, while a machine learning course is also testing a Socratic AI dialogue system designed to “guide inquiry rather than supply answers,” the statement reads. Beyond individual courses, SEAS funded seven cross-departmental faculty projects in generative AI this past academic year, spanning biomedical engineering, chemical engineering, industrial engineering, and computer science, according to the statement. This phenomenon extends beyond the computer science department. SEAS’ new AI minor, which it announced in November 2025, reflects the University’s attempt to adapt. The minor is specifically designed for noncomputer science majors and aims to provide “essential AI knowledge” while emphasizing practical and responsible use across disciplines, according to the SEAS website. At the center of the minor is AI in Context, a course taught by a team of faculty from the departments of computer science, applied mathematics, philosophy, music, literature, and the Writing Center, the representatives told Spectator. Bauer described it as targeted at students who want to bring AI into fields like economics or biology, rather than as a path into software engineering itself. At the graduate level, SEAS announced a new Master of Science in artificial intelligence in February, with the inaugural cohort starting in the fall. While recognizing how AI has helped them navigate coursework, students also described its rise as isolating, even among members of the Application Development Initiative, one of Columbia’s largest computer science clubs. “As people start to ask AI for questions that they would normally ask students and members, it starts to isolate people” Li said. Alexa Kafka, BC ’27, echoed Li’s sentiment, pointing out that computer science has always been a collaborative discipline with code reviews and pull requests, even as AI makes independent work easier. Despite this, Kafka noted that she hasn’t been to office hours “in at least a year.” Bauer added that faculty are exploring long-term solutions to the assessment problem. “One of the things that we’re trying to do at the department or school level is to maybe move towards computerized testing facilities,” Bauer said, referring to ideas in the department to revitalize computer labs as an alternative to pen-and-paper exams. “We have to accept that this is the reality, for better or worse.” But more exams, students say, does not mean better learning, nor does it reflect how computer science is actually done. “Code is so variable and can be approached in so many directions, there’s never just going to be one correct answer. There can be multiple optimal solutions,” Kafka said. Students also expressed concern that the shift toward exam-heavy grading pulls focus away from the applied, project-based work that better reflects both discipline and industry expectations. ADI has tried to respond more quickly than the curriculum can. Through a vibe coding workshop this past semester, club members said they wanted to help students understand not just that AI exists, but how to use it deliberately. “Vibe coding” refers to the practice of prompting AI to write code for a task or project, and has become widespread enough that ADI felt the gap in formal instruction was worth addressing, according to Sophia Huang, BC ’29. The workshop introduced students to the concept of context engineering, which Shaina Sahu, SEAS ’27, described as not just writing a prompt, but optimizing “the rest of the relevant information that you would feed into the model.” Without that, students said that AI-generated code can fail at scale or introduce difficult-to-detect bugs. Genawi said the shift toward AI has changed what it means to be good at computer science. “I think the shift has kind of gone towards what can you create and what can you do, instead of, what do you know,” she said. Indeed, during ADI DevFest, its annual hackathon, Li said more teams are trying to complete their projects than in years past. While sign-ups have historically outpaced submissions, Li said this is beginning to change because teams “have a tool that completes the vision for them.” While many students see AI as an equalizer that lowers the barrier to entry for coding and technical work, Murphy said it also creates new equity concerns inside the classroom around who can afford stronger tools and who already knows how to use them. “The equity issues I worry about are students who have access to paid versions of things versus free versions of things,” he said. Although popular AI models like ChatGPT and Claude have free versions available, both have paid models that can reach over $100 per month, which output different qualities of code. Murphy also pointed to differences in prior exposure to AI literacy itself. “Students who go to high schools or K-12 education, where they’re taught AI literacy, prompt engineering, context engineering, they understand how to do this, as opposed to students who go to schools who don’t teach that,” he said. “Then they come in, and then they’re somehow expected to know how to use these tools.” For Genawi, those disparities begin long before students arrive on campus. “It’s very interesting to consider how lucky you were geographically,” she said. “To be born where you were is influencing your ability to get access to these spaces.” Students like Yu, though broadly apprehensive about the future of computer science, are not without hope. “I’m optimistic, because I think if you’re truly passionate about CS, there will definitely be a place in the field to do something you want to do,” she said. Senior Staff Writer Ria Vasishtha can be contacted at ria.vasishtha@columbiaspectator.com. Follow Spectator on X @ColumbiaSpec. Staff Writer Arjun Menon can be contacted at arjun.menon@columbiaspectator.com. Follow Spectator on X @ColumbiaSpec. Want to keep up with breaking news? Subscribe to our email newsletter and like Spectator on Facebook. Produced with Spectate by Emily Ta
2026 National Science Foundation Graduate Research Awards & Honorable Mentions
Congratulations to our current graduate and undergraduate students and alumni who have received prestigious 2026 National Science Foundation (NSF) Graduate Research Fellowships (GRFP) and Honorable Mentions. The National Science Foundation (NSF) Graduate Research Fellowship Program (GRFP) recognizes and supports outstanding graduate students in NSF-supported STEM disciplines who are pursuing research-based master’s and doctoral degrees at accredited U.S. institutions. The five-year fellowship includes three years of financial support including an annual stipend and a cost of education allowance to the institution. Left to right: Row 1 – 2026 NSF Graduate Research Fellowship recipients: Jasmin Bioteau, Kasey Chung, Daniel Cohn, Michael Eng, Tyler Martinez. Row 2 – Henry Shavel, Breanna Sprague, Abigail Wilson, ofia Ando, Daniel Kendall. Row 3 – Rebecca Lee, Lauren Van Auken, Tim Duong. NSF Honorable Mention recipients: Evan Baerg, Isaac Hendrix. Row 4 – Nathan Palmer, Paul Saucedo, Helen Elmer and Maximilian Floridia. 2026 NSF Graduate Research Fellowships (GRFP) were awarded to current graduate students Jasmin Bioteau (Kwon), Kasey Chung (Garg), Daniel Cohn (Kappel), Michael Eng (Doyle), Tyler Martinez (Barber), Henry Shavel (Alexandrova), Breanna Sprague (Garg), Abigail Wilson (Garg); undergraduate researchers Sofia Ando (Shafaat, headed to Princeton for graduate school), Daniel Kendall (Barber, headed to MIT for graduate school), Rebecca Lee (Chen, headed to Stanford for graduate school) and Lauren Van Auken (Garg, headed to the Caltech for graduate school). Former undergraduate researcher Tim Duong ’25 (Neuhauser/Caram, now a graduate student at Princeton) also received the fellowship. 2026 Honorable Mentions by NSF, also a significant national academic achievement, were accorded to current graduate students Evan Baerg, Isaac Hendrix (Tinoco Valencia), Nathan Palmer (Conway),and Paul Saucedo (Doyle). Also receiving Honorable Mentions are former undergraduate researchers Helen Elmer ’25 (Gelbart/Knobler, now a graduate student at the Stanford University School of Medicine), and Maximilian Floridia ‘25 (P. Weiss, now a graduate student at Caltech). Congratulations again to all the recipients! Penny Jennings, UCLA Department of Chemistry & Biochemistry, penjen@g.ucla.edu.
Disney cuts some tech employees' stock-based compensation
Disney is cutting some employees' maximum stock awards to 25% of their base salaries, from 35%.
Amazon's business outside of AWS has been 'very resilient', says Maxim Group's Tom Forte
Tom Forte, Maxim, joins 'Closing Bell Overtime' with reaction to Amazon quarterly results.
Florida legislature approves new congressional map intended to boost Republicans in midterms
The Florida Legislature approved a new congressional map intended to maximize Republicans’ advantage in the state as part of the national redistricting battle that President Donald Trump launched ahead of this year’s midterms.
Match Preview: Manchester United vs. Brentford
Manchester United could move a step closer to confirming UEFA Champions League football for next season, when we host Brentford on Monday evening. The Reds need a maximum of two wins from our final five fixtures to rubberstamp a top-five finish, but may require even fewer points by the time we kick off against the Bees. Keith […]
Confirmed 2. Bundesliga lineups: VfL Bochum vs. Greuther Fürth
Uwe Rösler makes four changes from the defeat to Dresden. Maximilian Wittek, Noah Loosli, Cajetan Lenz and Philipp Hofmann come in for Erhan Mašović, Leandro Morgalla, Mats Pannewig and Callum Mars...
France's Macron says he will leave politics after 2027 election
French President Emmanuel Macron says he will not remain in politics after leaving the Élysée Palace in 2027, once his maximum two terms are over.
Morehead State Engineering Tech earns top ATMAE accreditation
MSU celebrates its engineering programs, achieving the maximum ATMAE accreditation, reflecting high-quality education and strong job-placement outcomes.
‘Maximum Pleasure Guaranteed’: Everything We Know About Tatiana Maslany’s New Show
Tatiana Maslany, Jake Johnson, and Murray Bartlett star in an unpredictable, darkly comedic thriller.
U.S. Delivers Maternal Health Equipment Across Madagascar - U.S. Embassy in Madagascar
U.S. Delivers Maternal Health Equipment Across Madagascar U.S. Chargé d’Affaires Stephanie Arnold today provided critical maternal and newborn health equipment that will reach healthcare facilities across Madagascar, demonstrating how U.S. foreign assistance builds a safer, healthier world by helping recipient countries build self-reliant health systems in line with the America First Global Health Strategy. Under President Trump’s America First foreign assistance policy and the America First Global Health Policy, U.S. support serves as a tool of strategic engagement – not global charity. Every U.S. taxpayer dollar is subject to rigorous oversight to ensure measurable results, maximum efficiency, and accountability, with zero tolerance for fraud, waste, or abuse. Through the bilateral Global Health Memorandum of Understanding with Madagascar’s Ministry of Public Health, both governments agreed on shared priorities and responsibilities, including a strong focus on maternal and newborn health and Madagascar’s increasing co-investment in its own health system. Madagascar continues to face major challenges in maternal and newborn health, including high maternal mortality rates and significant gaps in access to quality delivery care. In rural regions, many women still give birth at home or in facilities without adequate infection-prevention supplies. Through the U.S.-funded Momentum Country and Global Leadership Project, the United States is providing essential equipment to healthcare facilities nationwide: calibrated reusable drapes for infection prevention in approximately 1,000 community health centers (CSBs), single-use calibrated drapes for 35 hospitals, and management tools for health workers to improve data collection and care coordination. “U.S. foreign assistance makes America safer by preventing instability and strengthening health systems that can detect and contain infectious diseases before they cross borders,” said Chargé d’Affaires Arnold during the handover ceremony. “Our support is temporary and targeted, designed to help Madagascar build greater self-reliance and reduce long-term dependence on U.S. taxpayer resources.” This U.S. support will help standardize safe delivery practices, strengthen infection prevention, and improve data management systems. Used effectively, these materials will contribute directly to reducing maternal and newborn deaths, especially in rural and hard-to-reach areas where access to quality delivery care remains limited. From major hospitals to remote health centers, U.S.-funded equipment is already delivering results. The nationwide scope of this support ensures that women and newborns across Madagascar benefit from improved infection prevention and safer delivery practices. This targeted assistance helps save lives today while supporting Madagascar’s transition toward a self-sufficient health system, ensuring that today’s U.S. support leads to Madagascar’s future independence from foreign assistance.
Why Democrats with 2028 hopes are calling Lina Khan – and what she’s telling them about remaking the economy
Lina Khan is getting constant calls from Democrats, many of them thinking about presidential runs, to sound out problems and workshop potential solutions. That’s a shift from when she was Federal Trade Commission chair and some party leaders avoided meetings or events with her. Khan hasn’t changed so much as the world around her has. Many Americans are furious about rising costs, affordability, technological changes, and soaring profits for the rich. And President Donald Trump has maximized the levers of federal power in ways that were once considered off-limits in both parties. What Khan has in mind for the next Democrat in the White House isn’t a 1,000-page blueprint like the Heritage Foundation’s Project 2025. It’s more of an overall strategy rooted in her specialty: excavating long-forgotten laws already on the books, then finding ways to apply them to 21st century companies.
Suze Orman: This Overlooked Health Insurance Detail Could Cost You Thousands in 2026
Suze Orman says your maximum out-of-pocket cost is the health insurance detail most people overlook -- and in 2026, that could cost you up to $10,600.
Ryan Reynolds on Business Ventures, Blake Lively and ‘Deadpool’
In part one of this Sunday Sitdown Live, Ryan Reynolds joins Willie Geist in front of a packed crowd to discuss the rapid rise of Wrexham A.F.C., his many business ventures like his Maximum Effort, the young audience he’s found at home for his “Deadpool” franchise and what’s next for the superhero. Reynolds also shares support for his wife Blake Lively and reveals how he is feeling about his upcoming 50th birthday milestone. (Sponsored by More to Parkinson’s and brought to you by Acadia Pharmaceuticals Inc. Venue sponsored by City Winery)
Russia Signals Pullback in Business Support as Budget Strains Mount
Russia’s government warned it will scale back new support for businesses as mounting fiscal pressures limit its ability to sustain the stimulus that helped drive recent economic growth, Economic Development Minister Maxim Reshetnikov said Friday.
White House chief of staff to meet with Anthropic CEO over its new Mythos AI model
The meeting comes after tensions have run hot between the Trump administration and the safety-conscious Anthropic, which has sought to put guardrails on the development of AI to minimize any potential risks and maximize its economic and national security benefits for the U.S.
Tavernier wants maximum points from split as Rangers target title
Rangers captain James Tavernier has set his side a title-clinching target of five wins from five after the split. The Ibrox side, who were 13 points adrift of leaders Hearts when Danny Rohl took charge in October, are now just a point off the pace. Rangers begin their post-split games at home to Motherwell next weekend while Hearts are at Easter Road for an Edinburgh derby against Hibernian and third-place Celtic host Falkirk.
Sports quiz of the week: I Am Maximus, Marie-Louise Eta and Rory McIlroy
Did you follow the big stories in football, rugby, golf, baseball, basketball, boxing, snooker, cricket and racing?
Avoid these common small business tax mistakes
Block Advisors Strategy and Small Business Program Manager, Latasha Randle, chats with Host, Elizabeth Gore, about how self-employed entrepreneurs can simplify taxes, retirement, and healthcare planning. Randle shares practical advice on staying organized year-round, maximizing deductions, and building long-term financial security through proactive planning.
Jaylen Brown shares positive health update ahead of playoffs: 'I feel a lot different than last year'
"I'm excited to maximize my potential and take advantage of every opportunity that I have to win."
Maximum-security KY prison neared riot over guard planting contraband on inmates
Investigators say an officer who “discovered” contraband secretly planted it.
Worried about the midterms, Republicans mobilize rural Virginia voters against redistricting
Republican groups are mounting an intensifying, 11th-hour campaign to mobilize rural voters ahead of Virginia’s April 21 referendum to redraw its US House maps. At stake are as many as four US House seats Democrats want to redraw in their favor for the midterms. To maximize Democratic advantage, the map would put wealthy Washington, DC, suburbs in the same districts as farm communities 90 miles away, including in one misshapen district nicknamed the “lobster claw.”
How to Use Google Chrome’s New AI-Powered ‘Skills’
The premade Skills available through the Gemini sidebar in Chrome include ways to maximize protein in recipes or summarize YouTube videos.
In VfB masterclass: why Mittelstädt just let fly
In VfB Stuttgart’s impressive performance against Hamburger SV, Maximilian Mittelstädt was especially in the spotlight. The left-back not only delivered a flawless defensive display in the Swabians...
The Screen Time Maximalists Who Spend an Ungodly Amount of Time on Their Phones
As debates over social media addiction rage, people with extreme screen times tell WIRED they have no plans to cut back.
$10,000 bounty on the implementation of rollback netcode for Ultimate Marvel vs. Capcom 3 placed by Maximilian Dood
It's been nearly 15 years since Ultimate Marvel vs. Capcom 3 was released on November 15, 2011. Funnily enough, Ultimate Marvel vs. Capcom 3 was released about nine months after Marvel vs. Capcom 3 (February 15, 2011), which spawned ...
“AI polls” are fake polls
A few weeks after Donald Trump’s second presidential win, I took the train up from London (where I was living at the time) to Oxford to attend a conference on polls and forecasts of the 2024 election. Most of the attendees were pollsters or academics, but I also watched presentations from Aaru and Electric Twin, two companies that do what is interchangeably called synthetic sampling, silicon sampling, or synthetic audiences. Stripped of startup jargon, that means they use large language models (LLMs) to simulate responses to public opinion polls by having AI agents take on the role of survey respondents. I had already heard of Aaru thanks to some articles with eye-catching headlines like “No people, no problem: AI chatbots predict elections better than humans” in the months leading up to Election Day. The founders were making some big, some might even say far-fetched claims, such as: “within two years, we will simulate the entire globe — from the way crops are grown in Ukraine to how that impacts production of oil in Iraq, trade through the strait of Malacca, and elections for the mayor of Baltimore.” When Semafor asked Aaru’s cofounders — Cameron Fink and Ned Koh — about my boss, they said “we respect all those who came before us.” Nate (as he so often does) shared his thoughts on Twitter: Fink and Koh were relatively good-natured about this back-and-forth when we spoke at Oxford. They even offered to mail me one of the t-shirts featuring Nate’s quote they apparently had made. I never took them up on the offer, which I now somewhat regret. These synthetic sampling companies fell off my radar for a while, but they do still exist. In fact, Aaru recently received a $1 billion valuation. Is what they’re doing anywhere close to the most important frontier in AI development? Not by a longshot, especially when Anthropic just developed a model so adept at exploiting software vulnerabilities that it’s only being released to 40 companies. Still, silicon sampling is increasingly finding its way into public polling. Axios reported in March that “a majority of people trust their own doctors and nurses” based on findings from Aaru — without mentioning that the “people” in that sentence were actually LLMs. Around the same time, the Public Sentiment Institute “boosted” their online sample of 373 real survey respondents with 114 AI agents.1 (Spoiler alert: even the co-founder of Electric Twin doesn’t think that’s a particularly defensible approach.) Polling companies like Qualtrics and Ipsos are also developing synthetic data panels. So, what should we make of these … “polls”? Let’s get one thing out of the way: whatever they are, they’re not polls in the way that term is usually defined. Subscribe You can’t replace polls with AI On one hand, using LLMs to essentially make up fake survey respondents sort of sounds like the dumbest idea ever, one that will at best imperfectly replicate real polls while introducing all sorts of biases. On the other hand, with LLMs improving at a remarkable, perhaps even alarming rate, maybe that means I’m a dinosaur at the ripe old age of 24 because I still want to rely on polls that talk to actual people. I’m not going to argue that synthetic samples are completely useless. In fact, as I’ll return to later, there is evidence that some techniques can replicate topline survey results quickly and cheaply. But the marketing from certain companies can be slightly optimistic. “No traditional poll will exist by the time the next general election occurs,” said Fink in 2024. We’re just 206 days away from the midterms, and based on the fact that I still have to collect a bunch of polls every day, I’d say he should have run that prediction by a sample of AI agents before the interview.2 To see why synthetic samples can’t replace polls, here’s a quick primer on how they work. The simplest version of these models involves taking a LLM (like ChatGPT or Claude), giving it a demographic profile (e.g., a white, college-educated woman who lives in Utah and makes $70k a year), and then asking it to respond to a survey question. You repeat that process a few thousand times using different demographic profiles and end up with a sample of synthetic survey responses. The actual models used by private companies are more sophisticated than this, usually because they incorporate more hypothesized demographic characteristics for each agent and provide them with extra information. Aaru, for example, feeds agents a diet of news and information they’d be likely to consume, while Electric Twin incorporates their customers’ proprietary data about the audience they’re trying to replicate. The way Ben Warner, the co-founder of Electric Twin, explained it to me was “we have a large amount of data on […] for instance, 5,000 people. Can we make an accurate prediction of how they would respond to another question?” Still, it should be obvious why synthetic samples can’t replace polls. Polling is fundamentally a data collection process. We might use surveys to make predictions by feeding them into election forecasts, but the main purpose of a poll isn’t prediction, it’s gathering new data about what people think and how they feel. Silicon sampling, on the other hand, produces no new data. It’s simply a model: you input LLM training data, demographic prompts, and a bunch of other information, and it spits out a prediction for what a poll would say. We love models here, but models aren’t polls. That difference is an important philosophical sticking point for most pollsters I talk to. “I think politics should stay away from [synthetic sampling], because we’re trying to […] represent the voice of the people,” said Natalie Jackson, a vice president at GQR Insights. Democratic pollster John Hagner told me: “I think I’m just incredibly skeptical of this idea. I don’t think it’s research. At that point, you’re asking the machine to tell you what you already believe.” Hagner has seen some presentations of early synthetic sampling experiments, but so far, “if it’s being used in a campaign, people are keeping it incredibly quiet.”3 But Eli, I hear you saying, aren’t polls themselves increasingly governed by modeling decisions? Indeed they are: pollsters’ choices on which sampling method to use, how to define their likely voter models, and how to weight their samples can and do lead to dramatic differences in the results they publish. Aaru even referenced these limitations in the methodology statement included with that maternal mortality “poll” — although I’m using the term “methodology statement” loosely here, because it doesn’t really explain how the model works at all. We can ignore the (frankly preposterous) implication that synthetic sampling isn’t subject to a separate set of biases. The important point is that there’s still a meaningful difference between using weighting and other statistical techniques on actual polling data and using a model to predict what a poll would say. The latter is far closer to election forecasts or techniques like MRP — potentially useful models, but not a replacement for polls.4 To be fair, other synthetic sampling companies are perfectly happy with the distinction between polls and models. Warner compared polling and synthetic sampling to different tools in a toolbox. “The mistake I think we make is we think that these new tools should either work in exactly the same way or somehow replace these old tools,” he said. “Rather than thinking of it as, okay, so we’ve always had the hammer, we’ve always had the screwdriver, now we’ve got a saw. But don’t use a saw to try [to] do the job of a hammer.” A quick comment from Nate Eli didn’t ask me for a comment — rather rude of him, don’t you think? But since I’m editing this story, I figured I’d add a few quick thoughts rather than putting words in his mouth. Beyond the frequently misleading marketing, what bothers me about the AI “poll” hype is that as AI tools make statistical inference cheaper and/or better (note that these are not synonyms) that actually increases the comparative value of collecting original data. You might be able to train a model to make a reasonable estimate of what some hard-to-reach poll respondent would say — say, a young Black man who voted for Trump. (Such a person checks a number of boxes for a voter who is usually hard to reach in surveys.) Indeed, this is closely related to what models like the Silver Bulletin forecast already do. They essentially smooth out the kinks in noisy survey data by making inferences based on past voting patterns or national polls or surveys of other states. But you don’t actually know what these voters think unless you’re reaching them directly. If there’s a shift in opinion among this subgroup, you’re not going to detect it. So if I were running a campaign, I’d invest more in going the extra mile to find a representative sample of those voters. And then I’d hire some smart quants — or Claude? — to figure out the implications for campaign strategy based on proprietary data that my competitors didn’t have access to. -Nate Silver Are these models any good? If synthetic surveys are just a new type of model, the next obvious question is whether the models are at least accurate. The answer very much depends on who you ask. On one end of the spectrum, you have the maximalist argument that synthetic sampling is more accurate than actual polls. “It’s an incredibly challenging problem to go to someone and say ‘hey, we’re going to be more accurate at predicting human behavior than you, even when you talk to your customers directly’,” Koh recently told CNBC. In his view, synthetic sampling isn’t a saw to polling’s hammer, it’s “magic.” There’s certainly evidence that synthetic samples can replicate certain survey toplines. But if Aaru does have any examples of their approach outperforming the polls, they’re keeping those to themselves.5 Aaru’s 2024 election model, for example, had Kamala Harris leading in Michigan, Nevada, Pennsylvania, and Wisconsin on November 4th. And although they’ve since taken down their forecast page, they gave Harris a 50.5 percent chance of winning the race on November 2nd.6 After the election, Fink told Semafor he was happy enough with those results because they were “within margin of error,” a term that is completely meaningless when applied to a “sample” of AI agents. And of course, Aaru says their models have improved since 2024, so supposedly now they’d be more accurate than the polls? Still, their stronger argument is on cost: “We are significantly faster and cheaper than traditional polling, and still more accurate,” said Fink. The first two claims are undeniably true, but the third brings us to the opposite end of the spectrum. Both Jackson and Hagner are skeptical that these models are reliable for anything beyond replicating common survey toplines. “I just […] don’t think the machines are what we want when we’re looking for nuanced views. My example on this is people in Arizona and Nevada in 2024 who voted for Trump and voted for expanding abortion in their states on ballot initiatives,” said Jackson. Hagner identified another issue. Maybe the synthetic respondents, like sycophantic LLMs, are inhuman in one important way: they’re too nice. “The reports that have come through at the meetings that I’ve been at are that the early experiments on this, they cannot get respondents to be as racist or sexist or, frankly, as negative as human respondents,” he said. Academic research mostly agrees on this point. While there are some papers that show promising results when using LLMs to replicate polling data, most show that LLMs suffer from various quirks like producing too few “don’t know” responses and can seriously overpredict the favorability of politicians like Donald Trump and Kamala Harris. They also seem to struggle with too little variation between demographic subgroups, so the difference in predicted opinion between Democrats and Republicans, for example, is too small. When I asked Warner about these studies, his response to these papers was that just because academics can’t get synthetic sampling to work doesn’t mean that the technique doesn’t work in general. “Actually, the argument is, okay, yours does not [work]. That does not mean […] for this complex set of machinery, which uses a lot of investment, a lot of time, a lot of money, you can’t get it to work.” Cards on the table, I’m somewhat sympathetic to this argument because academics aren’t exactly great at making election forecasts. Usually, the people with skin in the game are the most accurate. Warner’s argument is that the approach Electric Twin takes — which includes, for example, making multiple predictions for each synthetic respondent using different models and prompts and subsequently averaging those to get a final prediction in a sort of ensemble forecast — produces better results than the simpler academic models. Warner shared a comparison between his method and the method from a recent academic paper with me, and Electric Twin was indeed able to get more accurate replication. But even still, he acknowledged that synthetic sampling “is not a crystal ball.” “If you asked me, do I think using other data sources will be more accurate than asking somebody who they will vote for, I would probably say no. But if you asked me ‘would your system be useful for our turnout modeling today?’ I would say yes.” For better or worse, it looks like the method is already getting more popular in the market research world. Most of the clients Aaru touts these days are businesses like EY and McDonald’s. And AI will probably begin to pop up in other parts of the political polling process. Pollsters are already using it to code open-ended survey responses, and some firms, like YouGov, are testing using LLMs to ask survey respondents questions. More worryingly, one danger to actual polls is that AI agents can be used to infiltrate online surveys. Most online polls use various checks to prevent that from happening, but there’s conflicting evidence on how effective those filters are and how prevalent AI agents currently are in online panels. If those agents ever become impossible to detect, it might spell the end of online polling, but the solution isn’t to replace all of your respondents with ChatGPT. Silver Bulletin is a reader-supported publication. To receive new posts and support our work, consider becoming a subscriber. Subscribe 1 That particular poll obviously doesn’t meet Silver Bulletin standards for aggregation. But we exclude all Public Sentiment Institute polls from our averages because we classify them as an amateur polling firm. 2 You could argue that Fink meant the next presidential election, but (a) I’m also confident we’ll still have real polls in 2028 and (b) in that case he should have asked an LLM to define “general election.” 3 Quick caveat: that’s reporting from a Democratic pollster. It’s possible that Republicans are more willing to use AI in political campaigns. 4 Indeed, Silver Bulletin does not include “polls” produced by MRP in our forecasts or averages, and we think it’s extremely misleading when their practitioners describe them in a way that suggests original data had been collected among a large number of states or Congressional districts. 5 A recent report from Aaru and EY did show two examples of a synthetic estimate being closer than a survey to a real-world benchmark — but I’d take those findings with a grain of salt because the report reads more like an ad and didn’t involve any sort of prediction being made ahead of time. 6 For comparison, our odds for Harris on the same day were 48.2 percent.
African frogs haven’t forgotten the ice ages. Scientists can tell by where they live.
Why are frogs diverse in some parts of Africa’s rainforests and less so in others? The patterns of cooling and glaciation during the last ice age would probably not have been your first answer or even your last-ditch guess, but it is, nonetheless, correct. “When the glaciers were at their maximum
Leading with Science and Evidence in a Time of Retreat
As global resources become more constrained, evidence-based tools that guide investments toward maximum impact are more important than ever. This World Health Day, the WHO urges the world to embrace evidence, support science-led solutions, and advance global cooperation to turn evidence into action.
My Blissful Week as a ‘Do Not Disturb’ Maximalist
I turned off my phone notifications for an entire week. It made me a less disturbed person, but the people closest to me got annoyed.
Iranian regime plays for a ‘maximalist advantage,’ says former political prisoner
Former political prisoner in Iran Kian Tajbakhsh discusses President Donald Trump’s stance on Iran on ‘The Will Cain Show.’
The Apple CarPlay Widgets Every iOS 26 User Needs to Enable Right Now
Maximize your dashboard with the best CarPlay widgets for iOS 26. From the new Ambient Music player to AI-powered voice bots and predictive traffic, here is what to install.
Spoiler Alert! Devils Rout Caps Behind Bratt, Hughes
Look at the Devils delivering maximum damage to a pair of Metropolitan Division rivals this week, although in very different ways. First, on Tuesday, they rolled over and played dead for a hapless Rangers team that got two points that won’t help them in their race to the bottom for Gavin McKenna. Forty-eight hours later […]
76ers' Edgecombe, Kings' Raynaud named March Rookies of the Month
76ers guard VJ Edgecombe and Kings center Maxime Raynaud have been named the NBA Eastern and Western Conference Rookies of the Month for March.
Apple AirPods Max 2 review
Maximum luxury, but should Apple have strayed from the formula more?