SCROLL NEWS / DISCOVERY
Search headlines
Find the stories shaping the conversation.
Results for " datasets " 48 found 🔀 AI-powered Shuffle
Sign in to save shuffle history.
How we build genomic sovereignty in the Arab world
Genomic medicine is being built from datasets that don't represent the Arab region, although it has the talent, data and infrastructure to close the gap.
SDSC Helps Improve AI-Ready Geospatial Data Through GeoCroissant
Oct. 1, 2026 — Researchers working with Earth science data often face a basic but time-consuming challenge: Determining what datasets exist, what they contain and whether they can be used together. Doug Fils, with the University of California San Diego Halıcıoğlu School of Data Science and Computing’s San Diego Supercomputer Center (SDSC), is helping advance a […]
The Math Singularity Is Here, and Science May Be Next
OpenAI has released an extraordinary body of mathematical work produced by an unreleased frontier model: 722 manuscripts spanning 372 research families across fields including number theory, complexity theory, and mathematical physics. Each result required, on average, the equivalent of roughly three hours of ChatGPT Pro reasoning. A New Era of Discovery We may be witnessing the beginnings of a “math singularity”: a point at which machine-generated discovery accelerates beyond humanity’s ability to independently reproduce, verify, or fully absorb it. If frontier models can generate hundreds of meaningful results in the time it takes a human researcher to investigate one, the constraint will no longer be producing new mathematics. It will be our capacity to validate, interpret, and integrate it. This does not make mathematicians obsolete. It changes their role. They may increasingly serve as guides, critics, and meaning-makers. Increasingly, they will decide which questions matter, determining which results can be trusted, and explaining why they are important. Mathematical insight may become abundant, while human judgment becomes more valuable than ever. Mathematics May Be Only the Beginning Every scientific discipline could eventually experience a similar acceleration. AI systems could generate and test hypotheses, connect findings across specialized fields, and explore possibilities at a scale no human research community could match. In medicine, imagine models identifying overlooked disease mechanisms, designing drugs and clinical trials, personalizing treatments, or discovering cures in months rather than decades. In materials science, AI could help create better batteries and cleaner industrial processes. In climate science, it could accelerate advances in energy, forecasting, and carbon removal. In biology, it could uncover relationships hidden across enormous genomic and molecular datasets. The result would not simply be faster research. It could expand the frontier of what is scientifically possible. The New Bottleneck Scientific discovery, however, is not the same as scientific progress. A medical breakthrough must still survive laboratory testing, clinical trials, regulatory scrutiny, manufacturing constraints, and real-world deployment. As machine-generated ideas multiply, the bottleneck may shift from producing breakthroughs to validating them safely and translating them into practice. We may soon have more hypotheses than laboratories can test, more drugs than clinical trials can evaluate, and more mathematical results than experts can review. That imbalance will require better systems for automated verification, reproducible research, prioritization, peer review, and collaboration between AI and domain experts. The importance of OpenAI’s work is not merely the number of manuscripts produced. It offers a glimpse of a world in which intelligence can explore thousands of intellectual paths in parallel. If ideas cease to be scarce, humanity’s defining challenge will be deciding which are true, which matter, and how to turn them into progress. The math singularity is not the end of human inquiry. It is the beginning of a new kind—one in which machines expand the frontier of knowledge and humans decide where it should lead. Subscribe Share Leave a comment
How low should blood pressure go? Science has the answer
New research suggests that aiming for a lower blood pressure target may deliver bigger heart health benefits than previously thought. Using large datasets and simulation models, scientists found that keeping systolic blood pressure below 120 mm Hg could reduce the risk of heart attack, stroke, and heart failure more than higher targets.
Scientists Stunned: Colossal Eggs Discovered 3,500 Meters Beneath an Active Volcano
Beneath an active seafloor volcano off Vancouver Island, researchers have documented a congregation of giant eggs at roughly 3,500 meters, a discovery that reframes how life can flourish under extreme conditions. The eggs belong to the Pacific white skate, an elusive deep‑sea species adapted to cold, crushing pressure and perpetual darkness. What seemed an unlikely nursery is actually a geothermally warmed refuge, where subtle heat helps embryos survive a marathon incubation. The finding links volcanic energy to a quiet engine of biodiversity, turning a hostile landscape into an incubator. A living volcano, a hidden nursery The seamount long assumed dormant now exhales mineral‑rich, shimmering plumes, proof of hydrothermal life at depth. Expedition cameras swept across black basalt and pale bacterial mats, revealing vents that pulse with warmth. That warmth, though modest by human standards, alters the physics of development for cold‑adapted embryos. In a realm where metabolism creeps, even a few degrees can mean faster growth and higher survival. The volcano is not only active; it is ecologically formative, drafting a blueprint for a thriving oasis. Giant eggs and a skate built for the abyss The Pacific white skate, Bathyraja spinosissima, is built for the abyss: broad wings, armored spines, and a life history that trades speed for endurance. Females lay outsized, rectangular “mermaid’s purses,” often 46–50 centimeters long, packed with rich yolk to fuel years of growth. In the frigid deep, embryos can require around four years to hatch, a timeline that magnifies every energetic advantage. The scale is striking—large eggs, large adults, and a textbook case of deep‑sea gigantism. The volcano’s gentle heat may literally shave months or years from the wait. Why heat matters at the bottom of the sea Hydrothermal circulation delivers low‑level warmth without the turbulence that would scatter fragile egg‑cases. Slightly elevated temperatures accelerate enzyme kinetics, nudging development along while keeping embryos within a safe range. Mineral‑laden fluids likely shape local microbiomes, which, in turn, can influence biofilms on egg surfaces. The environment acts like a slow, steady incubator, countering the deep sea’s tyrannies of cold and scarcity. For a species with long maturation and few offspring, every incremental boost compounds across generations. An ecosystem sculpted by fire and water Seafloor volcanoes create steep environmental gradients, stitching together hot vents, cool eddies, and mineral gardens into mosaic habitats. Microbes harvest chemical energy, seeding food webs that support crustaceans, fishes, and skates. Cameras documented egg‑cases nestled among rocky ledges, near communities nourished by vent‑driven chemistry. These islands of productivity behave like deep‑sea oases, sustaining life far beyond the vent orifice’s flickering light. The nursery’s success underscores how geological forces engineer biological opportunity. “Finding a cradle of life tucked beneath a breathing volcano reminds us that the planet’s harshest places can be profoundly generous.” What the discovery could change The nursery reframes conservation priorities in regions where volcanism quietly underwrites reproduction. Protecting hydrothermal corridors could safeguard entire lifecycles, not just charismatic adults spotted during occasional surveys. It also sharpens questions about how climate‑driven shifts in ocean circulation might influence vent activity, temperature fields, and embryo survival. With methodical mapping, time‑series cameras, and environmental DNA, scientists can track whether nurseries remain stable or drift with geologic rhythms. Key implications include: Enhanced understanding of deep‑sea reproduction at hydrothermal sites. New conservation targets centered on geothermal nurseries. Comparative studies with other volcanic regions, from the Galápagos to rift zones. Biotechnological insights from organisms adapted to long, cold development. Better models of how seafloor heat shapes regional biodiversity. Methods behind the revelation Remotely operated vehicles threaded through chimney fields, gathering video, temperatures, and mineral signatures. High‑definition imagery captured the distinctive, horned egg cases, their edges dusted with fine volcanic silt. Geochemical profiles mapped how heat and chemicals disperse, outlining the “thermal lanes” where eggs most cluster. Combined, these datasets reveal a nursery organized by geothermal topography, with egg‑cases concentrated in pockets of gentle, sustained warmth. A fragile boon in a changing ocean Deep‑sea environments feel remote, yet they intersect with global pressures: mining proposals, cable routes, and shifting oxygen levels. The nursery’s resilience hinges on stable heat flow, intact substrates, and minimal disturbance from noise or sediment. Safeguarding such places is not just about protecting novelty; it preserves a slow, ancient strategy that lets large, long‑lived animals endure in the coldest reaches of the Pacific. Where geology and biology meet, the planet’s deep time still writes living stories. This volcano‑warmed nursery adds a vital chapter to that story, revealing how subtle thermal gifts can tip the scales toward life in the abyss. With each dive, researchers trace the contours of an unseen world, where fire feeds water, and water—against all odds—feeds the future.
US government is pushing to gain unprecedented access to your medical records – as data protections are weakening
Large datasets can be useful research tools, but current government efforts to gather medical records depend on policy and technology that falls short of protecting people’s privacy.
Trump 2.0 has deleted or altered nearly 400 US datasets, endangering public health, education and more
Health data appears to be most affected by administration’s attempt to wipe findings that don’t align with its priorities, tracker shows
NVIDIA Opens 250+ AI Models to EU Science [2026]
NVIDIA joins EU RAISE, giving researchers 250+ open models, 250+ datasets, and 1,400+ projects via Hugging Face and GitHub.
Where do citizen science data come from?
Citizen science projects are playing an increasingly important role in ecological and conservation research worldwide. In these programs, members of the public support researchers by submitting photographs, audio recordings, observations, or even by contributing to data analysis. In Hungary, no fewer than 17 such initiatives combined their datasets in a joint study. The authors aimed to understand which socio-economic and environmental factors influence participation patterns in Hungary. By comparing the spatial distribution of citizen science data with an independent administrative dataset (municipality level data on income, education, demography, density and protected areas), the researchers identified several general trends as well as numerous project-specific relationships. For example, they found more per capita observations from municipalities with higher proportions of protected natural areas and lower population densities. Furthermore, excluding Budapest, the capital, both the proportion of people with diplomas and the proportion of elderly people showed a positive correlation with participation activity. In contrast, no consistent general pattern emerged regarding wealth or the proportion of children, although several project-specific relationships were observed. The study not only reveals how citizen science data are distributed across Hungary, but also provides robust support for the view that various biases must be taken into account when analyzing such data. The authors hope that their findings will help researchers in other countries design more effective projects and produce more accurate analyses by accounting for some of these or similar biases.
OpenAI Foundation Gives $125M to Public Data for AI-Driven Health Program
The Public Data for Health program will assist nonprofits and universities in developing, preserving, and sharing high-quality scientific datasets.
Large-scale neuroimaging datasets often lack information specific to women’s health, constraining AI’s analysis potential
Addressing this gap will require collecting widespread data on pregnancy, menopause and other life events women experience—and could bring us closer to the “holy grail” of linking brain and behavior.
Scientists warn fake research is spreading faster than real science
A sweeping new study from Northwestern University reveals that scientific fraud is no longer just the work of a few rogue researchers—it has evolved into a global, organized enterprise. By analyzing massive datasets of publications, retractions, and editorial records, researchers uncovered networks involving “paper mills,” brokers, and compromised journals that systematically produce and sell fake research, authorship slots, and citations.
Voices: Before we declare religion harmful to women’s health, let’s look at the evidence
“Across religious traditions, the pattern is clear,” write three researchers in an op-ed. “In nearly every case — and across two distinct datasets — greater religious attendance is associated with better health.”
WellSpan Health Announces Comprehensive Migration of Technology Portfolio to AWS
Add to FavoritesWellSpan Health, the leading health care system serving Central Pennsylvania and Northern Maryland, today announced plans for a comprehensive cloud migration of its full technology infrastructure. Together with Amazon Web Services (AWS) as its preferred cloud provider, WellSpan will transition its applications, data and environments from traditional data centers to AWS. This migration will establish the foundation for delivering next-generation, AI-enabled health care solutions that reduce administrative burden and improve outcomes for patients across the region and beyond. Supercharging Innovation, Efficiency and Patient Outcomes WellSpan's migration of systems and clinical applications, which serve 1.2 million patients across nine hospitals and 250+ facilities, is expected to be complete by April 2026, one of the fastest full-scale migrations in the health care industry. This transformation enables WellSpan to unify disparate data systems for care teams and patients alike, achieving operational efficiencies and care improvements that fragmented systems are unable to deliver. “Our comprehensive migration to the cloud accelerates our ability to deploy new clinical capabilities that transform patient care while advancing our overall innovation strategy,” said Roxanna Gapstur, Ph.D., R.N., president and CEO, WellSpan Health. “When a physician can access complete patient imaging in seconds instead of hours, we're not just modernizing technology—we're improving patient outcomes. Our new cloud infrastructure also enables us to quickly implement solutions that will free our clinicians from thousands of hours of administrative work annually so they can focus on what they do best: caring for our communities.” The transition allows WellSpan to move away from a data center-based model and enables the rapid deployment and scale of AI technology. WellSpan’s strategic partner, General Catalyst’s Health Assurance Transformation Company (HATCo), facilitated the collaboration with AWS to support the migration. WellSpan’s transformation journey benefits from HATCo’s specialized technology solutions, as well as General Catalyst and AWS’s significant partnership to improve patient outcomes, which was announced in January 2025. “Our decision to go all-in on AWS will enable insights and innovations impossible in traditional siloed environments,” said Kasey Paulus, executive vice president and chief operating officer, WellSpan Health. “This is the largest health care technology transformation in our region's history, and it positions WellSpan to be a premier partner for technology leaders across the country.” “WellSpan is building a foundation with AWS that will improve care and outcomes for patients across their health system by bringing data together into standardized formats that are optimized for AI applications,” said Dan Sheeran, vice president and general manager, Healthcare and Life Science, AWS. “To do this, it is essential to have the most secure and compliant cloud when handling patient data, along with AWS’s built-in AI governance and responsible AI frameworks that follow the industry’s highest standards to protect patient information. We look forward to continuing to work alongside WellSpan as they unlock the cloud and AI’s full potential to deliver exceptional care to their patients.” Key Benefits of the AWS Collaboration Full Portfolio Migration: WellSpan will transition its applications and data from on-premises data centers to AWS, including full IT environments and specialized applications for functions such as nursing, cardiology, pharmacy and the company’s supply chain systems. This represents approximately 7.5 petabytes of clinical and non-clinical data and more than 300 applications—the most comprehensive health system cloud migration in the Mid-Atlantic region. Approved Architectures: AWS will deliver secure, WellSpan-specific architectures tailored to each build, ensuring compliance and reliability across the health system. Unified, AI-Ready Data: Migrating to AWS enables WellSpan to harmonize data across separate systems, transforming unstructured clinical notes and multi-modal datasets into standardized, AI-ready formats. Accelerated Results and Improved Outcomes: Automated data normalization and processing will enable the deployment of applications that reduce turnaround times for items such as analyzing medical imaging or patient records from days to minutes, helping clinicians get patient results faster and improving overall health outcomes. Reduced Administrative Burden: The transition to AWS will enable swift deployment of AI-powered solutions that reclaim thousands of hours for WellSpan’s clinical and corporate teams, allowing them to focus on patient care rather than paperwork. Elastic Infrastructure: Cloud architecture will enable WellSpan to scale computing resources dynamically, supporting seasonal demand fluctuations without infrastructure investments, as well as the rapid deployment of new services across multiple locations simultaneously. Most Secure, Compliant Cloud: AWS offers the highest network availability of any cloud provider, with more than 166 HIPAA-eligible services and certifications for global compliance standards, including HITRUST CSF. Built-in AI governance and responsible AI frameworks ensure that WellSpan’s data is protected under the industry’s most rigorous standards. Tags:News releaseInnovation and partnerships
How should European scientists respond to the US AI kill switch?
Earlier in June, Europe’s worst fears of technological dependence were confirmed when the White House ordered the AI firm Anthropic to stop non-US citizens using its leading models. This kill-switch moment crystalised anxieties in Brussels that the continent, lagging the US on the creation of leading-edge AI models, is now a technological of vassal of Washington. The shock has specific implications for European scientists and universities. Some governments are now worried that with AI being integrated into labs, Donald Trump now has his finger hovering over an off switch for European science. “Having restrictions on this tool, it can have an impact on our own research,” said France’s research minister, Philippe Baptiste, at the VivaTech conference in Paris on June 18.Baptiste, himself a former academic, said that colleagues in mathematics had recently warned him they were now cut off from leading-edge US models, risking their work. “This model is not available anymore now because of the US restriction, and for the first time this had an impact on mathematics,” a field where large language models (LLMs) are making a significant contribution, said Baptiste. So, should European scientists avoid relying on US models in the future? Or would this be a self-harming overreaction? And how is AI being used in science, anyway? What happened?On June 12, it emerged that Washington, citing cybersecurity concerns, had ordered Anthropic to stop non-US citizens using two models: Claude Mythos 5, and Claude Fable 5, a version of Mythos but with guardrails to prevent abuse. Anthropic said Washington’s concerns were overblown. But forced to follow the order, yet with no way to check who was using the models, the company suddenly shut down access for everyone, US citizens included. There’s still no agreement on how to restore broad access. But last week, Bloomberg reported that select US companies, and the US government, still had access to Mythos despite the ban, but the EU’s cybersecurity agency Enisa had been shut out. Class division of researchersIn the longer term, access outside the US should be restored, said Marcel Salathé, co-director of the AI centre at the Federal Polytechnic School of Lausanne (EPFL), because a permanent restriction would be financially ruinous to US AI firms. “Even Trump will eventually budge to economic pressure,” he said.However, US AI firms might end up operating rather like banks, doing due diligence on users of their leading-edge AI models, he went on.European scientists are unlikely to fall foul of such checks, he said. But having a Chinese postgraduate in their lab, or doing a type of research the US disapproved of, could potentially result in access being denied. “It will create a sort of class division of researchers,” he said. More than frontier modelsHowever, successfully deploying AI in science is about far more than just having access to leading LLMs such as Mythos.Instead, many science-specific tools, most famously Alphafold, which predicts protein folding structures, require “far less” computing power to train than frontier models, well within the capabilities of European supercomputers, according to a recent analysis by the Arq Foundation, a Brussel-based AI think tank. Instead, tools such as Alphafold are possible because certain fields, like protein folding, weather prediction or astronomy, have large, clean, standardised datasets that can be used to train predictive models. Rather than compute, “data is the central constraint for AI for science,” the report says. For example, a less powerful physics model trained on CERN’s data “will be more valuable than the best model without certain data,” said Bengüsu Özcan, author of the report. European advantage?Europe could therefore steal a march by using its scientific data and infrastructure to create models US digital giants cannot. “They might actually match or go even beyond what these big [US] companies are doing,” said Özcan.Indeed, some European start-ups hope to create automated testing labs to create the kind of standardised data that could speed up the discovery of new materials. Materials scientists say there’s currently too little data to train models. AI scientistsHowever, even if the bottleneck for AI in science is often data, rather than compute, being at the mercy of the US for LLM access could still become a problem for European scientists. A new class of “AI scientists” based on leading LLMs is also emerging, Özcan’s report says.Unlike domain-specific tools like Alphafold, these try to automate day-to-day scientific work, by providing an overarching tool that coordinates specific models to synthesise existing literature, generate hypotheses or run analysis of data. The problem for Europe, however, is that these tools are sometimes based on US leading models. Google Deepmind’s AI Co-Scientist, for example, is built on Google’s Gemini LLM. Systematic failureThese scientific automation tools are still in their early stages, and still exhibit “systematic failures on scientific tasks,” Özcan’s report says. One of the problems is that the messy, day-to-day process of scientific discovery is largely not documented, meaning LLMs don’t have a good source of data to learn how scientists actually work. Meanwhile, some tasks, such as literature synthesis, don’t typically require access to leading edge models.However, the performance of AI scientific assistants could improve as companies such as Google pour resources into developing them. Rely on US“Unfortunately, for scientific assistants, there is not much we can do right now except rely on US frontier AI,” said Antoine Bosselut, a natural language processing expert based at EPFL. Chinese-built open models, which can run on local computers, lessening the risk of a shutdown, are touted as one alternative to US dependence. There’s also Mistral, once seen as Europe’s main AI hope, but the Paris-based company has seen its recent models fall behind in performance. But the Chinese alternatives are generally not as capable as US frontier AI models, said Bosselut, unless paired with fiddly, custom tools that allows them to act as independent agents. “Most scientists do not have the time,” he said. Don’t jump ship yetAlthough the kill switch moment is seen a warning shot for European scientists, it doesn’t mean labs should abandon US AI tools immediately. “I do not think the right response is simply to abandon US-based models and systems under the banner of sovereignty or independence,” said Mattias Björnmalm, secretary general of the Cesaer university association. “That would be an overreaction.”Related articlesEuropean sovereignty in AI requires ‘ugly trade-offs,’ say experts Robotic lab start-up makes case for €500M EU-funded materials testing facilityAI in science: is it useful? Instead, European researchers need to ask themselves whether they’re becoming dependent on technology that could be turned off in Washington, said Özcan. “They might not see this today, but they should really start asking this question,” she said. European researchers need a plan B, said Salathé. “I would recommend to every researcher, very proactively explore alternatives.”Own toolsIn addition, Björnmalm and others want European research institutions to develop their own scientific AI tools.Instead of signing AI deals with US digital giants, European universities should be using this money to create their own models, which may end up being superior for research, said Sabine Leonelli, an expert on the use of AI in science at the Technical University of Munich. “Universities in Europe seem to be happy to sign commercial contracts with US companies on the assumption [their AI tools] are better,” she said. But US frontier LLMs are typically opaque, and have been developed without independent scientific input, she said. “This is not good enough for science.”Last year, Swiss researchers released Apertus, an LLM designed to be completely transparent, from its development methods to its training data. The idea is that this makes scientific tools built on top of the model far more reproducible than with closed, commercial alternatives.“Europe needs more of this: serious capacity-building that gives researchers, universities and companies credible options,” said Björnmalm.
Mantis Biotech is making 'digital twins' of humans to help solve medicine's data availability problem
Mantis takes disparate sources of data to make synthetic datasets that can be used to build so-called "digital twins" of the human body, representing anatomy, physiology and behavior.
As AI Transforms Science and Research, Indiana University Cybersecurity Experts Protect Against Hidden Risks
AI is now commonly used to analyze large datasets, identify patterns, generate ideas and accelerate discoveries. But what isn’t as well understood is the risks versus rewards of the AI explosion.
MIT researchers are mapping extreme weather risks — and building tools to act on them
MIT researchers are translating climate science into actionable tools for extreme weather preparedness. The Climate Grand Challenges project has produced papers, digital tools, and datasets to help planners and utilities assess flood, hurricane, wildfire, and heat risks.
New Health Data Unlocks Progress
AI-powered wearable health trackers are transforming healthcare by using massive datasets to predict risks, personalize insights, and empower individuals.
Scientists detect a sharp acceleration in global warming
Global warming appears to have shifted into a faster gear. After removing the effects of El Niño, volcanic eruptions, and solar cycles, researchers found that the planet has warmed at roughly 0.35°C per decade over the past ten years, compared with just under 0.2°C per decade from 1970 to 2015. The acceleration, which becomes visible around 2013 to 2015, appeared across five major global temperature datasets with more than 98% statistical certainty.
People in Brazil are living past 110 and scientists want to know why
Scientists are uncovering why Brazil may be one of the most important yet underused resources for studying extreme longevity. Its highly diverse population harbors millions of genetic variants missing from standard datasets, including rare changes linked to immune strength and cellular maintenance. Brazilian supercentenarians often remain mentally sharp, survive serious infections, and come from families where multiple members live past 100. Together, they reveal aging not as inevitable decline, but as a form of biological resilience.
Multimodal embeddings at scale: AI data lake for media and entertainment workloads
This post shows you how to build a scalable multimodal video search system that enables natural language search across large video datasets using Amazon Nova models and Amazon OpenSearch Service. You will learn how to move beyond manual tagging and keyword-based searches to enable semantic search that captures the full richness of video content.
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.
Scientific Papers Can Now Become AI Agents
There are millions of scientific research papers that include tons of information. Of course, not all that information is valuable, but buried across those papers are experiments, datasets, methods and software that could help researchers answer new questions. The problem is that no scientist can possibly read and make use of more than a small […]
Discover, explore and analyse business statistics
Have you ever wished you could see how a certain economic activity is performing without drowning in datasets or spreadsheets? Meet the Business sector profiles: your one-stop gateway for fast insights, easy to grasp comparisons and deep dives into the EU’s business economy. This tool puts everything in one place offering...
Within days of the January 2025 inauguration, more than 3,000 taxpayer-funded federal data sets from the CDC, the Justice Department and the Office of Personnel Management were pulled offline, and Harvard public health researchers held marathon sessions scraping the pages before they vanished
Within days of the January 2025 inauguration, thousands of federal datasets began disappearing from public websites. Researchers scrambled to mirror the pages before they vanished — and much of what was captured cannot fully replace what was lost.
The fun and flexibility of data science
Taylor Bolt spent his Ph.D. and postdoc digging through brain imaging data for clues to cognition. In industry, the datasets are different but the joy of answering questions with data remains.
Inaugural ASU–Science Prize winners use AI to help farmers, trafficking victims
From continent-spanning satellite imagery to distributed criminal networks hiding online, some of our most intractable challenges are buried within mountains of data. Two researchers are using artificial intelligence to make sense of such complexity and turn overwhelming datasets into practical solutions.
Viewpoint: Unlocking Europe’s wildfire data
Europe is already investing in the science, technology and infrastructure needed to understand, prevent and fight wildfires. Copernicus provides satellite-based Earth observation and environmental information, while the European Forest Fire Information System provides information on fire danger and forest fires across Europe. Weather and climate systems add forecasts of heat, drought, wind and humidity; historical fire databases show where fires have occurred; and researchers generate increasingly detailed knowledge about fuels, fire behaviour, forecasting, modelling and risk management. Meanwhile, new European research projects are adding digital twins and other tools to this growing ecosystem.However, these data do not necessarily sit together, nor are they necessarily described, structured or governed in ways that allow them to be used together. Much of the information is public, but some is commercially held. Some is sensitive. Some is subject to legal, security or operational restrictions. Even when data can be legally accessed, different systems may use different formats, terminology and rules.Europe therefore faces an unusual paradox: we have vast amounts of relevant information to respond to the wildfire challenge, but not the means to bring it together. For example, a wildfire researcher or emergency service might need to ask whether a particular combination of vegetation, drought, wind and topography has historically been associated with rapid fire spread. The relevant information may exist in several institutions, using different systems and subject to different conditions of access.The challenge is to conduct an authorised analysis that interacts with these sources, combines the results and respects the rules imposed by each data custodian.An important part of the answer can be found in the Fair principles, the idea that data should be findable, accessible, interoperable and reusable. Developed more than a decade ago, these principles have become a central reference point for modern data stewardship, making it easier for people and machines to find data, understand what it means and determine whether it can be reused.Readability by machines is essential, since AI systems will be key to meeting the data challenge. Bring analysis to the dataThe proposition is not to give an AI system unrestricted access to sensitive databases. Quite the opposite. The goal is to make it possible for approved algorithms to go to the data, rather than requiring sensitive data or large datasets to be copied to wherever the analysis happens.Imagine an analysis asking a series of questions of different data stations. At one station it might examine weather and climate data; at another, historical fire records; at another, experimental measurements of fire behaviour. The algorithm does not need to download everything. It can operate within the environment permitted by each data custodian and return only the results that the rules allow to leave.This is sometimes described as data visiting, and related approaches already exist in areas such as federated analysis and federated learning.Sensitive data should therefore not be treated as a free-for-all for machines. This requires appropriate governance at multiple levels. Access needs to be authenticated and authorised. The purpose of an analysis should be defined. The code or model being run should be controlled. Data custodians should be able to specify what can be queried and what results can leave their environment. Activities should be auditable and human oversight should remain where the consequences require it.In other words, the objective is not to trust AI with the data. It is to design systems in which AI never has more access than it needs.As open as possible, as closed as necessaryFair data should also not be confused with open data. Some wildfire data should be openly available. Other information may legitimately need to remain restricted because it concerns critical infrastructure, security, commercial interests, privacy or operational capabilities. The principle should be: as open as possible, as closed as necessary.The important point is that restricted data should not automatically become unusable data. If machines can discover what information exists, understand its meaning, establish where it came from and determine the conditions under which it can be used, authorised analyses can potentially be carried out without the underlying data ever leaving the environment in which it is protected.Restricted data can still be Fair and machine-actionable.This is also an important question of data sovereignty. European institutions should not have to surrender control of their data to a handful of technology platforms in order to benefit from AI. Nor does Europe need to start again by building one giant repository for every piece of wildfire information. The better approach is to connect what already exists.This is the principle behind emerging approaches such as Tryangle, developed through the Leiden Initiative for Fair and Equitable Science. Rather than creating another centralised data warehouse, the approach seeks to provide an interoperable layer through which approved algorithms can visit distributed data sources while respecting the rules of their custodians.In this way, a data station can remain where it is. Its custodians can retain control. Existing databases and applications do not have to be abandoned. What changes is their ability to communicate with other systems through common, machine-readable descriptions, access rules and interoperability mechanisms.But approaches like this need data that meets the Fair principles. Only then can an interoperable ecosystem safely begin to turn fragmented information into collective intelligence.Data infrastructure is wildfire infrastructureThe implications of this approach are entirely practical. Firefighters could gain better intelligence about local and evolving risks. Land managers could test prevention strategies using more comprehensive information about fuels, weather and past fires. Researchers could compare evidence across ecosystems without repeatedly moving huge datasets between institutions. Smaller regions and resource-constrained teams could access analytical capabilities without having to build large data infrastructures themselves.Europe is already investing heavily in satellites, databases, research programmes, forecasting systems and scientific expertise. The question is whether we will connect those investments well enough to turn them into actionable intelligence.That requires policymakers and funders to treat data infrastructure as seriously as physical infrastructure. Data generated or supported through European public funding should be Fair, verifiable and machine actionable by design, wherever legally and ethically possible.Related articlesData Corner: Europe’s burning need for wildfire mitigation researchClimate monitoring must remain a global enterprise, experts sayIn practice, that means funding not only the production of data, but also the infrastructure that makes data usable across systems: metadata, provenance, interoperability, access mechanisms and governance. Where data must remain restricted, we should ensure that it can still be securely discovered, verified and used for authorised analysis.Initiatives such as the European Open Science Cloud offer an existing framework through which these principles can be put into practice. What is needed now is to accelerate implementation and make them a standard part of how Europe funds and builds research and innovation systems.We do not need another generation of disconnected databases. The data is already there. The opportunity is to make it work together, securely, responsibly and quickly enough to help prevent the next fire, rather than simply document the last one.Jean-Claude Burgelman is professor emeritus at the Free University of Brussels (VUB) and editor-in-chief at Frontiers Policy Lab. For more than 20 years he held leading roles in research and innovation policy at the European Commission.
Health Data Utilities as Essential Public Infrastructure: A Cooperative Federalism Architecture
Federal interoperability initiatives like Trusted Exchange Framework and Common Agreement transport clinical records across boundaries but fail to aggregate and enrich state-governed datasets, including Medicaid claims, prescription monitoring records, and social service registries.
Political Entrepreneurs or Bandits? The “Criminal” Origins of Peripheral Rebellions -
Political Entrepreneurs or Bandits? The “Criminal” Origins of Peripheral Rebellions By Janet I. Lewis and Stephen Rangazas, George Washington University How and why do armed groups that become known as “rebels” initially use violence? New datasets [...]
Scientists Release Biggest 2D Map of the Universe
Key Takeaways The DESI Legacy Imaging Surveys combined more than 263,000 telescope exposures to make the largest 2D map of the universe in visible and near-infrared light. Astronomers can pair the Legacy Surveys map with their own observations to explore our universe and search for rare phenomena. The 2D map serves as the foundation for the Dark Energy Spectroscopic Instrument survey to measure the universe in 3D and investigate dark energy. Hold on to your telescopes: the DESI Legacy Imaging Surveys team has released the largest-ever 2D color map of the universe. The 5.6-trillion-pixel map contains nearly 4 billion celestial objects, primarily stars and galaxies. The data is available for all to use and publicly viewable through the Legacy Survey Sky Viewer. Astronomers and citizen scientists can explore the map or combine it with their own observations to better understand our universe. Researchers can search for rare phenomena like gravitational lenses, observe fleeting events like supernovae, and investigate two of physics’ biggest mysteries: dark matter, the invisible substance that accounts for most of the mass in our universe, and dark energy, the force driving our universe’s accelerating expansion. The new map builds on earlier versions from the DESI Legacy Imaging Surveys that have already proved invaluable. To date, more than 1,800 science papers that reference the Legacy Surveys data have been published. “It’s part of the fabric of astronomy research now,” said David Schlegel, a co-lead of the Legacy Surveys and scientist at the Department of Energy’s Lawrence Berkeley National Laboratory (Berkeley Lab). “When you’re working with astronomical objects today, you often start by pulling up the Legacy Imaging Viewer to see what you’re looking at.” Covering roughly 75% of the sky in visible and near-infrared light, the updated map provides a deep view of the extragalactic universe not blocked by the dust and stars of our own Milky Way. Researchers expect it will remain the most comprehensive 2D map of our universe for years to come. More than 160 scientists contributed to data collection for the project, and a team of 20 produced the final dataset released today. It was built by combining 263,407 telescope exposures from three ground-based sky surveys: the Dark Energy Camera Legacy Survey (DECaLS) at NSF Cerro Tololo Inter-American Observatory, the Mayall z-band Legacy Survey (MzLS) at NSF Kitt Peak National Observatory, and the Beijing-Arizona Sky Survey (BASS) at the University of Arizona’s Steward Observatory. That was supplemented by years of data from NASA’s Wide-field Infrared Survey Explorer (WISE) satellite mission and additional public data. “For our team, these data are fundamental to our investigation of the expansion history of the universe and the formation of our galaxy,” said Arjun Dey, co-lead of the Legacy Surveys and an astronomer at NSF NOIRLab. “But the skies belong to everyone, and this survey gives everyone the chance to explore the sky and marvel at its wonders.” Here be galaxies The DESI Legacy Imaging Surveys were originally conducted to prepare for the Dark Energy Spectroscopic Instrument (DESI) survey. The Legacy Surveys’ 2D map is essentially a deep photograph of the sky; it records where galaxies and stars appear and how bright they appear. This crucial step enables DESI to select objects and measure their light in different wavelengths to determine their distances, building the largest high-resolution 3D map ever made. Scientists study the way galaxies have clustered at different ages of the universe to track dark energy over time. In April 2026, DESI completed its original five-year survey ahead of schedule and with vastly more objects than expected. The early results have shown surprising hints that dark energy’s impact may be weakening over time — a paradigm shift that could potentially shape the predicted fate of our universe. DESI expects to publish improved results using their first five years of data in 2027 and is continuing observations into 2028. DESI was so efficient at observing galaxies, the Legacy Surveys map needed to expand. Early on, “it became clear we might run out of galaxies to look at and run out of sky, and we better start doing something about that,” said Schlegel, who also works on DESI. The new Legacy Surveys map has been used to select DESI targets since June 2026 and will guide the telescope’s operations over the coming years. Computing the cosmos Merging hundreds of thousands of images taken on 2,285 nights, each with unique atmospheric and telescope conditions, was a massive computational effort. It took about a year to develop the computer code and eight weeks to process all the images at the Perlmutter supercomputer at the National Energy Research Scientific Computing Center (NERSC) at Berkeley Lab. Beyond supporting DESI, the Legacy Surveys will be a foundational reference for the next generation of telescopes. As new observatories like the NSF-DOE Vera C. Rubin Observatory (jointly funded by NSF and DOE’s Office of Science) and NASA’s Nancy Grace Roman Space Telescope come online, researchers can compare their observations with one of the deepest and most comprehensive views of the sky ever assembled. The Legacy Surveys data will also help scientists train artificial intelligence tools to analyze petabytes of astronomical data and accelerate new discoveries. It will be among the datasets used in an astrophysics pilot project within the American Science Cloud, part of the DOE’s Genesis Mission. The DESI Legacy Imaging Surveys are supported by the U.S. Department of Energy’s Office of High Energy Physics; the National Energy Research Scientific Computing Center, a DOE Office of Science user facility; the U.S. National Science Foundation, Division of Astronomical Sciences; and the partner institutions. DESI is supported by the DOE Office of Science and NERSC. Additional support for DESI is provided by the NSF; the Science and Technology Facilities Council of the United Kingdom; the Gordon and Betty Moore Foundation; the Heising-Simons Foundation; the French Alternative Energies and Atomic Energy Commission (CEA); the Secretariat of Science, Humanities, Technology and Innovation (SECIHTI) of Mexico; the Ministry of Science and Innovation of Spain; and by the DESI member institutions. ### Lawrence Berkeley National Laboratory (Berkeley Lab) is committed to groundbreaking research focused on discovery science and solutions for abundant and reliable energy supplies. The lab’s expertise spans materials, chemistry, physics, biology, earth and environmental science, mathematics, and computing. Researchers from around the world rely on the lab’s world-class scientific facilities for their own pioneering research. Founded in 1931 on the belief that the biggest problems are best addressed by teams, Berkeley Lab and its scientists have been recognized with 17 Nobel Prizes. Berkeley Lab is a multiprogram national laboratory managed by the University of California for the U.S. Department of Energy’s Office of Science. DOE’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science.
1000 new African genomes added to global science
Wits-led AGenDA project sequenced 1,000+ African genomes, uncovering vast genetic diversity and novel variants. This helps correct Eurocentric bias in global datasets, improving disease prediction, treatment design, and ensuring African populations benefit from genomic medicine.
Gemma 4 and what makes an open model succeed
Having written a lot of model release blog posts, there’s something much harder about reviewing open models when they drop relative to closed models, especially in 2026. In recent years, there were so few open models, so when Llama 3 was released most people were still doing research on Llama 2 and super happy to get an update. When Qwen 3 was released, the Llama 4 fiasco had just gone down, and a whole research community was emerging to study RL on Qwen 2.5 — it was a no brainer to upgrade. Today, when an open model releases, it’s competing with Qwen 3.5, Kimi K2.5, GLM 5, MiniMax M2.5, GPT-OSS, Arcee Large, Nemotron 3, Olmo 3, and others. The space is populated, but still feels full of hidden opportunity. The potential of open models feels like a dark matter, a potential we know is huge, but few clear recipes and examples for how to unlock it are out there. Agentic AI, OpenClaw, and everything brewing in that space is going to spur mass experimentation in open models to complement the likes of Claude and Codex, not replace them. Especially with open models, the benchmarks at release are an extremely incomplete story. In some ways this is exciting, as new open models have a much higher variance and ability to surprise, but it also points at some structural reasons that make building businesses and great AI experiences around open models harder than the closed alternatives. When a new Claude Opus or GPT drops, spending a few hours with them in my agentic workflows is genuinely a good vibe test. For open models, putting them through this test is a category error. Something else to be said about open models in the era of agents is that they get out of the debate of integration, harnesses, and tools and let us see close to the ground on what exactly is the ability of just a model. Of course, we can’t test some things like search abilities without some tool, but being able to measure exactly the pace of progress of the model alone is a welcome simplification to a systematically opaque AI space. Share The list of factors I’d use to assess a new open-weight model I’m considering investing in includes: Model performance (and size) — how this model performs on benchmarks I care about and how it compares to other models of a similar size. Country of origin — some businesses care deeply about provenance, and if a model was built in China or not. Model license — if a model needs legal approval for use, uptake will be slower at mid-sized and large companies. Tooling at release — many models release with half-broken, or at least substantially slower, implementations in popular software like vLLM, Transformers, SGLANG, etc due to pushing the envelope of architectures or tools. Model fine-tunability — how easy or hard it is to modify the given model to your use-case when you actually try and use it. The core problem is that some of these are immediately available at release, e.g. general performance, license, origin, etc. but others such as tooling take day(s) to week(s) to stabilize, and others are open research questions — with no group systematically monitoring fine-tunability. In the early era of open models, the days of Llama 2 or 3 and Qwen pre v3.5, the architectures were fairly simple and the models tended to work out of the box. Some of this was due to the extremely hard work of the Llama, Qwen, Mistral, etc. developer teams. Some is due to the new models being genuinely harder to work with. When it comes to something like Qwen 3.5 or Nemotron 3, with hybrid models (either gated delta net or mamba layers), the tooling is very rough at release. Things you would expect to “just work” often don’t. I’ve been following this area closely since we released Olmo Hybrid with a similar architecture, and Qwen 3.5 is just starting to work well in the various open-source tools that need to all play nice together for RL research. That’s 1.5 months after the release date! This is just to start really investing more into understanding the behavior of the models. Of course, others started working on these models sooner by investing more engineering resources or relying on partially closed software. The fully open and distributed ecosystem takes a long time to get going on some new models. All of this is lead-in for the most important question for open models — how easy is it to adapt to specific use-cases? This is a different problem for different model sizes. Large MoE open-weight models may be used by entities like Cursor who need complex capabilities in their domain, e.g. Composer 2 trained on Kimi K2.5. Other applications can be built on much smaller models, such as Chroma’s Context-1 model for agentic search, built on GPT-OSS 20B. The question of “which models are fine-tunable” is largely background knowledge known by engineers across the industry. There should be a thriving research area here to support the open ecosystem model. The first step is to understand characteristics of different base and post-trained models to understand what they look like. The second step is to tune pretraining recipes for open models so they’re more flexible. Interconnects AI is a reader-supported publication. Consider becoming a subscriber. Subscribe For The ATOM Project and other Interconnects endeavors, we’ve put in substantial effort to measuring adoption trends in the open ecosystem. Everything takes a long time to unfold after a model is first publicly available — and adaptability is why. What we know for sure now, when Qwen has been going from strength to strength with its releases, is that technical staff across the industry has gotten comfortable working with Qwen models. Countless research methods and datasets were made to work with Qwen. It’ll take patience for any other model family to get to this point — a patience I’m not sure many open model builders have. This takes us to Gemma 4, Google’s latest open models. Gemma 3 was released more than a year ago, in March of 2025, and is a bit underrated. Gemma 4 comes in 4 sizes for now, with a bigger, MoE model of over 100B total parameters rumored but not released yet. The models we have today come in sizes of ~5B dense, 8B dense, 26B total 4B active MoE, and 31B dense. I’m most excited that they’re finally adopting a standard Apache 2.0 open source license. This’ll massively boost adoption. The standard of better licenses for strong open-weight LLMs was set by mostly Chinese open model labs in the last 1-2 years, and now U.S. companies are following suit. I will personally be so happy if the horrible Llama licenses and Gemma terms of service were an ~18-month transient dynamic of the industry being nervous about releasing strong open models. The Gemma 4 scores look very solid, the small models have incredible benchmark scores (especially in general domains like LMArena) and the 31B model rivals the recent Qwen 3.5 27B, which is the leading member of that class. The ~30B size range is an important one, as it’s accessible both to researchers and to enterprises looking to deploy the model in real use-cases. Where the 7B model scale is the default for tinkering and research, a 30B model is the default for seeing if an open model can unlock substantial value in your specific workflow — a good mix of intelligence, low price, tractability for downstream training, etc. This takes us back to the above adoption criteria I mentioned for open models and the bigger question — do I think Gemma 4 will be an overwhelming success? Previous Gemma models have been plagued by tooling issues and poorer performance when being finetuned. Gemma 4’s success is going to be entirely determined by ease of use, to a point where a 5-10% swing on benchmarks wouldn’t matter at all. It’s strong enough, small enough, with the right license, and from the U.S., so many companies are going to slot it in. I’m cautiously optimistic that Gemma 4 is going to work better here. Winds are shifting for open models built in America. We saw GPT-OSS go through a bumpy launch to become an overwhelming success. There’s a collective energy around the likes of Reflection, Arcee, Nemotron, Gemma, Olmo, and peers that show substantial demand for building new stacks around open models. There’s capital to be spent on AI stacks across the economy by those who want more ownership of everything, including the model. After launching The ATOM Project 240 days ago, the conversation is shifting into the next stage. Summer of 2025 was a crisis moment where the U.S. AI scene realized it can’t wait and figure out open models after building AGI. The two markets will capture different areas and proceed in parallel. Now that more companies in the U.S. are releasing strong models, we need to improve the ecosystem so that these models are easy to use, understand, and build value around. It’s the hard work to build another inflection point in these adoption plots I’ve been updating consistently, but that’s the work to be done. Join me in it. More data coming soon! Here’s a sneak peek:
Linux Finally Catches Up to Windows with a Game-Changing Performance Feature
For years, the Linux kernel’s scheduler has been world‑class at balancing loads, yet it missed a crucial, cache‑aware instinct. In modern multi‑core systems, that gap can turn into measurable latency, especially when threads bounce between cores that don’t share the same cache. A new upstream feature, often called Cache Aware Scheduling, is now set to change that equation. A cache‑savvy scheduler arrives At the heart of any OS, the scheduler decides which thread runs where and for how long. In modern CPUs, small private caches (L1 and L2) sit beside each core, while a larger Last Level Cache (LLC, typically L3) is shared among groups of cores. When a task migrates across LLC boundaries, its warm data may vanish from the cache, forcing slower trips to main memory. Cache Aware Scheduling keeps related tasks close to their shared LLC, reducing destructive migrations. By respecting cache topology, the kernel minimizes cold‑start penalties and preserves locality that many workloads desperately need. The result is less thrashing, fewer memory stalls, and more consistent throughput. "Keep tasks close to their data, and the system will keep performance close to its peak." Why this narrows the Windows gap Windows has long leaned on topology‑aware, cache‑sensitive heuristics, especially since the Windows 10 era. That advantage helped Microsoft handle hybrid designs with P‑cores and E‑cores, plus complex cluster and NUMA layouts. With cache awareness integrated upstream, Linux brings parity to this vital dimension, without sacrificing its trademark flexibility. The Linux approach remains deeply configurable, reflecting the ecosystem’s breadth across servers, desktops, and embedded devices. It layers atop existing NUMA‑balancing and energy‑aware logic, refining placement rather than reinventing the wheel. Crucially, it aligns scheduling with the hardware’s real shape, not just with abstract CPU counts. Real‑world gains and who benefits Early tests on Intel Sapphire Rapids platforms point to gains in the 30–45% range for select workloads. Those wins appear in cache‑sensitive tasks like in‑memory analytics, high‑thread compilation, and microservices with tight working‑sets. Games and latency‑bound engines can also feel steadier frame‑times, especially when threads share hot assets. The benefits extend to AMD’s 3D V‑Cache parts, where keeping threads near enlarged cache slices prevents needless misses. Hybrid x86 designs with performance and efficiency cores further profit when cache and core roles are scheduled in concert. Even handhelds running SteamOS can squeeze more out of limited power, translating cache locality into smoother play. Key advantages include: Lower end‑to‑end latency Fewer expensive misses to RAM Better multi‑socket and NUMA behavior Improved energy efficiency More predictable QoS under mixed loads Stronger scaling on dense servers Caveats, tuning, and rollout No scheduler change is purely free, and trade‑offs still exist. Favoring locality can reduce cross‑cluster balance if limits are set too tight. Likewise, fairness and utilization must remain healthy, especially under heterogeneous, bursty workloads. Linux’s implementation is designed to be measured and tunable, not a blunt instrument. It cooperates with energy‑aware scheduling for mobile efficiency, and with NUMA‑balancers on big iron. Expect architecture‑specific refinements as vendors surface richer topology hints and cache‑sharing maps. Upstream integration has begun, with broad distribution adoption likely over the next cycles. Many users will encounter the change through standard kernel updates in 2025–2026, depending on distro cadence. Server‑class kernels may enable or tune it sooner for targeted fleets where cache locality is money in the bank. How to think about performance impact Cache Aware Scheduling amplifies gains when your workload is both CPU‑bound and cache‑sensitive. Think tight inner loops, hot code paths, and datasets that fit snugly in shared LLC. It’s less dramatic for I/O‑bound services or memory‑hungry tasks that blow past cache capacity. Still, even modest locality improvements often translate into smoother tail‑latency, which is where user experience and SLAs typically break. Developers can further help by pinning related threads, batching work, and aligning data to reduce cross‑cluster chatter. The scheduler’s new awareness works best when applications expose good hints. The bottom line By understanding and honoring cache topology, Linux removes a subtle yet costly bottleneck. The kernel’s smarter placement keeps hot data hot and work near where it belongs. That narrows a long‑standing gap with Windows, while preserving the openness and tunability Linux champions. For gamers, creators, and operators, the payoff is practical: more consistent frames, faster builds, and snappier services. For the ecosystem, it’s another step toward hardware‑savvy scheduling that turns transistor complexity into real‑world speed. And for Linux itself, it’s a timely upgrade that matches today’s silicon with tomorrow’s expectations.
Students found a star from the dawn of the universe drifting into the Milky Way
A group of undergraduate students stumbled into a cosmic time capsule—one of the oldest stars ever discovered—while combing through massive astronomy datasets. What began as a class project quickly turned into a breakthrough when they spotted an extraordinarily “pristine” star made almost entirely of hydrogen and helium, hinting it formed near the dawn of the universe.
Viewpoint: Europe needs a new scientific contract
The White House's new science strategy should be required reading in the Berlaymont and every European capital.Not because Europe should copy it. Parts of it should concern Europeans deeply. But because it asks the question Europe has so far avoided while steadily losing ground in the global race for technological leadership.Europe already possesses many of the world's finest universities, scientists and research-intensive companies. It produces world-leading research in fields ranging from quantum technologies and biotechnology to climate science. Yet time and again Europe generates the scientific breakthroughs while others build the companies, industries and strategic advantages around them, particularly in digital technologies. Increasingly, we find ourselves regulating technologies whose underlying capabilities were developed elsewhere.This is not primarily a scientific failure. It is an institutional one.The settlement that shaped modern science policy emerged after the Second World War. In Science, the Endless Frontier, Vannevar Bush argued that governments should invest in fundamental research while protecting scientists from political direction. Public investment was justified precisely because no government, university or company could predict where the next transformative discovery would emerge.That scientific contract served liberal democracies extraordinarily well. It underpinned decades of advances in medicine, computing, energy and communications, but the world for which it was designed no longer exists.Artificial intelligence is transforming not only technology but the production of knowledge itself. Advanced chips, computing power, data and scientific instruments have become strategic assets. Scientific knowledge now diffuses globally almost instantly. Competitive advantage increasingly lies not in making discoveries first, but in possessing the infrastructure, capital, engineering capability and institutions required to translate those ideas into products, industries and strategic advantage.The old boundaries between science, industrial policy, national security and geopolitics have largely disappeared. Dual-use technologies now underpin both economic power and military deterrence.The US has recognised this shift. In July, the White House published Science: A New Golden Age, a comprehensive attempt to rethink the relationship between science and national power. The report builds on Bush's legacy, but departs from his framework in important ways. Science is no longer treated simply as the source of future discoveries. It is presented as the foundation of economic resilience, technological leadership and national security. AI, advanced infrastructure and the capacity to build become as important as discovery itself.Yet the report also reveals a striking paradox. The same administration calling for “a new golden age” of US science has simultaneously proposed deep cuts to the federal research enterprise while expanding political influence over the allocation of public research funding. Europe should therefore distinguish between the report's diagnosis and its prescription. The diagnosis, that science has become a strategic asset, is compelling. But the prescription is self-defeating. Long-term technological leadership depends on strong universities, academic freedom and the institutional autonomy that makes scientific discovery possible. This distinction matters. Bush's achievement was never simply to make science serve national strength. It was to establish a division of labour in which governments invested in research without attempting to direct discovery itself. Scientific breakthroughs emerge precisely because researchers retain the freedom to pursue questions whose value cannot yet be demonstrated and whose conclusions cannot be anticipated.Neither China, nor the USAs science becomes more strategically important, the temptation to politicise it inevitably grows. But scientific freedom is not an obstacle to technological leadership, it is an essential precondition.China represents the opposite model. For decades it has integrated science, manufacturing, infrastructure and industrial policy into a single national strategy. Its ability to mobilise resources around critical technologies is formidable. But it does so within an authoritarian system where scientific priorities are increasingly inseparable from political power.Europe should imitate neither model. It should learn from the strategic ambition of the US without importing its growing politicisation of science. And it should learn from China's capacity to build without adopting its concentration of power.Instead, Europe needs a scientific contract of its own. Bush's scientific contract protected the freedom to discover because the defining challenge of the 20th century was producing new knowledge. Europe's next scientific contract must preserve that freedom while ensuring that scientific discoveries can be converted into technological, industrial and strategic capability. Because in a transactional world, Europe's freedom to shape its own future depends on its capacity to build it.Related articlesThe US ‘golden age’ science strategy: what Europeans need to knowEuropean researchers sound the alarm that Trump rule changes will hit cooperationNIH turmoil risks reciprocal EU-US access to health research fundingThis does not mean subordinating free science to competitiveness or industrial policy. On the contrary. Scientific autonomy remains the first principle of Europe's next scientific contract. Governments should not direct scientific discovery. Their responsibility is to ensure that Europe's scientific, industrial, financial and regulatory institutions work together rather than in isolation.Meeting that responsibility requires far more than research funding. European researchers need access to world-class computing infrastructure, advanced laboratories, shared datasets and scientific facilities. Innovative companies need patient capital, testing environments and a genuine single market in which they can scale without facing 27 different regulatory regimes. Public procurement must evolve from purchasing existing technologies to creating markets for new ones. Regulators must become partners in experimentation rather than guardians of yesterday's rules.A connecting architectureEurope does not need to invent the Defense Advanced Research Projects Agency (Darpa) model from scratch. Britain's Aria, Germany's Sprind and Wellcome Leap already show that Europe can build institutions capable of taking high-risk, mission-oriented bets while preserving democratic accountability. The challenge now is not to create more isolated initiatives, but to connect them to Europe's broader scientific, industrial and financial system so that breakthrough discoveries can scale.Above all, Europe must stop treating research, innovation, industrial policy, finance and security as separate policy domains. The challenge is no longer to create excellent institutions in each of these fields, but to make them work as one European innovation system. Europe's strength will increasingly depend not on excellence within institutions, but on excellence across them.Europe's ambition should be neither to imitate the US nor to emulate China. It should demonstrate that scientific freedom, democratic legitimacy and technological capability are not competing objectives, but mutually reinforcing strengths. In an age of geopolitical rivalry, Europe's greatest advantage may ultimately be its ability to attract the world's best talent by combining scientific freedom, democratic openness and technological ambition.Bush's scientific contract helped define the 20th century because it protected the freedom to discover. Europe's challenge is not to abandon that achievement but to extend it. The defining task of Europe's next scientific contract is to preserve the freedom to discover while building Europe's capacity to translate scientific discovery into technological power.Kaare Nielsen is founder of Pomus, a Danish think tank on science and innovation.
Five Berkeley Lab Scientists Receive DOE Early Career Research Awards
Five scientists from Lawrence Berkeley National Laboratory (Berkeley Lab) have received Early Career Research Program (ECRP) awards from the Department of Energy. The ECRP program, now in its 16th year, was established to bolster the nation’s scientific workforce by supporting exceptional researchers at the outset of their careers, when many scientists do their most formative work. Recipients use the funding to pursue projects that have the potential to solve scientific challenges and advance the country’s STEM capabilities. To be eligible for the program, a researcher must be an untenured, tenure-track assistant or associate professor at a U.S. academic institution, or a full-time employee at a DOE national laboratory or Office of Science user facility, who is within 10 years of having earned their doctorate. Awards to an institution of higher education will be approximately $875,000 over five years, and awards to a DOE national laboratory or Office of Science user facility will be approximately $2,750,000 over five years. This year’s Berkeley Lab awardees and their projects are listed below: Timon Heim: Better detectors for more powerful particle colliders Timon Heim is a staff scientist in the Physics Division. His ECRP project will develop a new type of detector that can track the particles generated by collisions in particle colliders with high spatial and temporal resolution. Current detectors in use at high-energy colliders, like the ATLAS detector at the Large Hadron Collider (LHC) at CERN, can track the trajectory and measure momentum of particles, but they can’t track a particle’s time of arrival with sufficient precision to discern from which of the many particle interactions they originate. Heim’s proposed technology — called a 4D tracking detector, because it captures data about particles’ positions in space and time — aims to solve this challenge, allowing researchers to glean more precise data from collision experiments to uncover new insights into fundamental forces and potentially discover new particles. The enhanced capabilities of 4D detectors will be a necessity for experiments at future colliders, such as the Muon Collider or Future Circular Collider. The project will develop custom microelectronics that process the signal of a pixelated silicon sensor that can register charged particles passing through it in an extreme radiation environment that these colliders produce during operation. The goal is to measure individual particles’ position with 50-micro-meter spatial and 50 pico-second (one trillionth of a second) temporal precision, for an incoming flux of around 1 billion particles per second per square centimeter. Harrison Lisabeth: Studying underground processes to unleash abundant energy Harrison Lisabeth is a research scientist in the Energy Geoscience Division focused on combining real-world experiments and AI to guide development of geothermal energy systems. His project will explore how deep underground rock formations – where geothermal wells are dug – are stressed by chemical reactions of the minerals and mechanical strain from movements in the Earth. A better understanding of subsurface stresses is essential to engineer safe and reliable geothermal wells that access the heat of deep rocks to produce electricity. It is estimated that the U.S. could generate ten times the amount of energy needed to power the grid using geothermal sources. Lisabeth plans to simulate how minerals grow, dissolve, and influence the behavior of surrounding rock in geothermal reservoirs by tracking chemical changes in rock samples under pressure in the lab. The project will also develop machine learning models to predict geologic strain based on field measurements and refine existing models that describe how common rock types deform under stress. Combining these approaches, Lisabeth hopes to develop a framework to predict stress to improve stability and longevity of geothermal wells. Daniel Carney: Pushing the quantum boundaries of particle measurement Daniel Carney is a staff scientist in the Physics Division who uses quantum information science to study particle physics and gravity. At very small scales, “quantum noise” — random fluctuations from the act of measurement — can drown out faint signals from rare processes. Carney’s ECRP project, “Fundamental Physics at the Quantum Limits of Measurement,” aims to build a new mechanical quantum sensor that can measure extremely small particle interactions below the limits of that quantum noise. The approach uses lasers to hold a tiny bead containing radioactive atoms; when one decays, the system measures the bead’s motion and the energy of any particles emitted. The experiment’s initial goal is to search for heavy sterile neutrinos, hypothetical particles predicted by some extensions of the Standard Model. The system could also help measure the mass of the neutrino, study the way particles interact through the electromagnetic and weak forces, and hunt for low-mass dark matter — all unsolved mysteries in particle physics. Callum Wilkinson: Using machine learning to demystify neutrinos Callum Wilkinson is a staff scientist in the Physics Division and part of the international Deep Underground Neutrino Experiment (DUNE) hosted by Fermi National Accelerator Laboratory. DUNE will create an intense beam of neutrinos that interact in detectors in Illinois and, after an 800-mile journey through the earth, in massive detectors in South Dakota. By precisely measuring neutrinos and how they change, researchers will get a better understanding of their fundamental properties and whether they are responsible for why there is so much more matter than antimatter in our universe. Wilkinson’s ECRP project will apply two machine learning techniques to improve the analysis of DUNE’s data and enhance the experiment’s precision. One approach will look for hidden patterns in how neutrinos interact and produce low-energy particles, while the other will help correct for distortions in what the detectors record. The methods will be tested on one of DUNE’s prototype detectors and could be used in other neutrino experiments, improving their measurements and helping scientists build better models. Aditi Krishnapriyan: Machine learning methods that scale as datasets grow Aditi Krishnapriyan is a faculty scientist in Berkeley Lab’s Applied Mathematics and Computational Research Division (AMCR) and assistant professor at UC Berkeley. Her project, awarded through UC Berkeley, will develop machine learning methods that scale with dataset size for modeling complex, multi-scale phenomena. The resulting tools can be applied in a wide range of research, including the study of new electrolytes for batteries or the examination of molecular interactions. ### Lawrence Berkeley National Laboratory (Berkeley Lab) is committed to groundbreaking research focused on discovery science and solutions for abundant and reliable energy supplies. The lab’s expertise spans materials, chemistry, physics, biology, earth and environmental science, mathematics, and computing. Researchers from around the world rely on the lab’s world-class scientific facilities for their own pioneering research. Founded in 1931 on the belief that the biggest problems are best addressed by teams, Berkeley Lab and its scientists have been recognized with 17 Nobel Prizes. Berkeley Lab is a multiprogram national laboratory managed by the University of California for the U.S. Department of Energy’s Office of Science. DOE’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science.
Observational Comparative Research in Cardiovascular and Brain Health and Disease
Resources for observational comparative research have rapidly expanded to include large, granular healthcare datasets and advanced analytical methods.
How AI Is Creating A New Age In Healthcare
Deploying AI to sift through massive biological datasets to identify patterns humans may not see is helping to spread medical solutions more widely and at lower cost.
Science on Target: Q3 Research We Conduct
The Target ALS Global Natural History Study (GNHS) is entering a new phase of discovery. As the only international, longitudinal study generating deeply characterized clinical, genomic, and biofluid data to accelerate ALS biomarker discovery and therapeutic development, the GNHS is now beginning to deliver large scale datasets and resources needed to support new insights into […]
August Monthly Review: ChatGPT In Epic
Death, taxes, another frontier lab healthcare launch inspiring truly insane LinkedIn takes. As much as I want to stay away, I repeatedly am the crewman unplugging his ears whenever we pass these sirens. So it’s not the first and probably not the last time we dial up the typewriter rather than tie ourselves to the mast. The full blow-by-blow timeline: Health API Guy is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber. Subscribe The End of the Standalone PHR (Jan 8): ChatGPT Health arrives for consumers, at the time powered by b.well. I was bullish on the approach in that the best way for the PHR to finally solve distribution was to meet consumers were they already are (i.e. When Horizontal Meets Healthcare (Jan 9): OpenAI for Healthcare puts out the shingle for enterprises, with SharePoint ingestion standing in for the EHR integration the product actually needed. Another One: Anthropic’s Healthcare Debut (Jan 16): Claude's version, on HealthEx rails instead of b.well, with Agent Skills for FHIR development as the one true vertical-specific investment. ChatGPT for Clinicians: The Trap Sprung (Apr 24): OpenAI filled in the missing GTM middle with a free product for verified clinicians, effectively peeling the enterprise wrapper off ChatGPT for Healthcare and taking the PLG fight directly to OpenEvidence. OpenAI’s Second Attempt at Health (Jul 31): a relaunch of their patient specific app, now branded Health in ChatGPT, where the privacy partition came out, the connectors thinned, and b.well disappeared in favor of first-party FHIR work against a smaller network Beyond a Shingle So what was announced now? Today, we’re introducing a new electronic health record integration that brings authorized patient context from Epic into ChatGPT for Healthcare, along with the Healthcare Public Data plugin for direct, structured access to official healthcare datasets like PubMed, DailyMed, and CMS Coverage. Together, these capabilities bring ChatGPT closer to the systems and sources healthcare teams trust, while supporting the controls and compliance healthcare work requires. The core portfolio is unchanged from prior announcements we delineated and discussed (aside from the relaunch of patient-facing). Since trade publications still seem to be confusing them by using the wrong names: Enterprise sales motion: ChatGPT for Healthcare Protecting their core product / PLG: ChatGPT for Clinicians Consumer/patient-facing product: Health in ChatGPT So really, this is a feature release (specifically for ChatGPT for Healthcare when it comes to EHR integration). What’s cool about it, though, is that we are moving beyond just putting out a shingle. Everything in the January enterprise launch was horizontalizable: Horizontal tech companies thus generally start verticalization with their “putting out a shingle phase”. Telling the world you’re open for business as a horizontal tech company reliably attracts early inbound interest from buyers who are already trying to force-fit horizontal tools into domain-specific workflows and are eager for any signal that the vendor intends to support their use case more directly. Nothing was really vertical specific, which is exactly what you'd expect from the ultimate hypergrowth horizontal company dipping their toes into specific industries. But eventually you have to stop changing the sign on the door and start changing the product. Can you guess where the horizontal product was going to run out of road? This is the core dilemma writ large: verticalization requires differentiation, while scale economics of consumer and horizontal push toward unification. I’ve buried the lede so deep here it will be painful to some readers, but what should be shocking to no one is that my perspective is that EHR integration is the only path that meaningfully resolves this tension for ChatGPT for Healthcare. Integration with Microsoft Sharepoint is fine, but it is categorically insufficient for any product that hopes to influence clinical decision-making. SMART Money Well here we are, eight months later! Real vertical investment has begun, which surprised a few people. One friend in the industry messaged me: “Absolutely no way Epic gave ChatGPT API access.” Looking forensically at the announcements, the integration is notably read-only across appointment notes, laboratory results, medications, and specialist documentation. So while I’m not a betting man, I'd wager the house this is a SMART on FHIR launch: That sounds a lot like USCDI! The recommended path for the workflow they outline would be SMART on FHIR Their Health AI lead’s post is evocative of both standalone and EHR launch SMART, which another termed “ChartGPT” and “EHR Plugin” The implementation burden for SMART on FHIR is the lowest of the available integrative paths This is the team who literally did patient-facing SMART on FHIR with Epic last month Most importantly, if you’re a horizontal company, you generally still haven’t taken the full plunge and thus (overly) value reusability. SMART on FHIR is the logical compromise: healthcare-specific enough to matter, but standardized enough to reuse. Nobody Asked Judy Another friend asked “I thought they had to go to each health system, or did they go direct to Epic and now health systems opt in?” Given the trust paradigms in healthcare, this functionally cannot be Epic hoovering up all the data for OpenAI and shoving ChatGPT in their customers’ faces (or they would revolt). However, I do believe the aforementioned terrible coverage by TechCrunch and other outlets is probably responsible for that question, given the 325 million reference. When using FHIR as a business associate to an Epic customer, there is no EHR gatekeeping to speak of, as that isn’t how fhir.epic.com works: You register an app You pick your APIs You test against the sandbox APIs You list as “Ready for Production” Hospitals pick your app You test and go live with them Bluntly, nobody had to say yes or no in Verona (nor were they given the chance). Vendor Services (their next developer tier up that I doubt OpenAI is using quite yet in this initial release) certainly has contractual paperwork that some resent, but even there, the twin pressures of mounting antitrust and information blocking make outright gatekeeping increasingly fraught. My friend’s sentiment is, to me, representative of a broader industry neurosis, perhaps a sort of scar tissue of prior eras: people assume the gate is still there and never actually try the door. There are certainly other ways Epic (and any EHR) can put its thumb on the scale, but they are not so dumb as to stand in front of a federally mandated API and play bouncer at this exact moment in time. So I think they can and probably should go deeper, as this release is at best parity and at worst behind vertical-specific competition. OpenEvidence did a basic SMART launch with Sutter Health in February, allowing for more convenient evidence search by providers. They then add patient-context (the equivalent SMART on FHIR flow to what we see here) when Cedars-Sinai went enterprise-wide in May. UpToDate has four distinct applications across deeper workflows like patient engagement listed in Epic Showroom and announced a partnership with Epic to power Art at UGM Abridge and other ambient scribes have invested deeply into the deepest bidirectional clinical copilot workflow with integrations well beyond SMART on FHIR. In that light, the “Who is OpenAI primarily targeting here?” is clear. This release brings them into striking distance of OpenEvidence, but not the others quite yet. They are the only one of the three you can reach without building something truly Epic-specific. When will they go further? Competition is a great motivator to overcome the horizontal demons and start building things that can’t be reused: not across industries, not across EHRs, maybe not past the customer you built them for. That’s the price of actually verticalizing. The question is if and when OpenAI will be willing to pay it. Month in Review Here is the monthly review. As a reminder, this is a regular round-up of the month’s posts and other content to surface things you may have missed across regulation, litigation, interoperability, and beyond. AI assistance is used in these bullet summaries so I can focus on articles. Articles Published: None this month Video Content: The Information Exchange: Standards-based Thruple Edition (Aug 14): Back to school for us too, with Brad reporting in from the CMS Health Tech Ecosystem’s one year anniversary in DC. We get into the January CMS-0057 deadline, the fall rulemaking reading list, and why enrollment still keeps most apps off FHIR. The Information Exchange: Epic Dúnadan Edition (Aug 25): A UGM roundup with Ryan Brickner joining for the first time to check our takes against what the building actually thinks. Rangers, Ergo sitting on top of EHI, and a down-market lineup that needs some cuts. Regulatory: Stacked Deck, Bad Hand (Aug 05): ONC’s website refresh took the entire HIT Policy Committee record with it, so I rebuilt the archive and went looking for the regulatory capture story everyone assumes is buried in there. Stacking the deck and winning the hand turn out to be very different things. The Sixth Generation of Patient Access (Aug 19): A quick catalog of the five generations of patient access we’ve layered on since HIPAA, and the two candidates now competing to be the sixth. One is planned. The other is the market routing around the plan entirely. Court cases: Three Cases Walk Into a Docket (Aug 06): Two surprise settlements cleared the board in a single week, and then Judge Maddox dropped ninety-four pages on Vyne v. Henry Schein. The sleeper is a DMCA holding that makes direct-to-database a considerably riskier business model. Epic v. Health Gorilla: Into the MDL (Aug 10): Nine class actions are headed to Miami, and the Panel signaled it wants to bring the case that spawned them along too. That would leave Epic arguing the requests were obviously fraudulent in one courtroom and unknowable in the other. Amazon v. Perplexity: Agents Are Legalized! (Aug 11): The Ninth Circuit vacated the injunction against Comet, holding that the user is the one accessing the servers rather than the company that built the agent. A real win for agentic access, and a much narrower one than the headline suggests. Veeva v. Epic: Come At Me, Bro (Aug 26): Epic’s response brief wants the dismissal affirmed and, unusually, wants the opinion published as precedent. Buried in it is Epic’s own description of what its non-competes actually prohibit, which current and former employees should read closely. EHRs: The Contract Epic Would Never Sign Today (Aug 04): A 1999 SEC exhibit catches Epic licensing nearly its entire product line, Tapestry included, to the company that became TriZetto. The marketing services menu attached to it is the part that will make you blink. Forecast from Verona (Aug 07): The Epic Almanac makes one argument six different ways: the AI is only as good as the networks behind it. Emmie headlines, Art’s context stack is the most ambitious part, and Penny finally gets an autonomous coding date on the calendar. Works With Epic MyChart (Aug 17): Epic’s first new Showroom category since the death of Workshop certifies hardware instead of software, badge on the box and all. Made for iPhone, but for blood pressure cuffs, and rough news for anyone selling the RPM stack sitting in between. UGM Hot Takes 2026 (Aug 20): Everyone else covered the AI announcements, so I went after Savvy deleting the payment gateway, Rangers as Boost with the timer removed, and a down-market lineup with too many entries. Plus the Health Grid tidbits I cannot help myself on. MyChart Central Grows Up (Aug 24): Device data and Emmie turn Epic’s identity hub into a full consumer platform, which is both a logical answer to ChatGPT and a bit of a mistake. It also happens to be the best scraping target Epic has ever shipped. Industry Analysis: The Wrong Yardstick (Aug 12): Vertical software keeps getting judged against Superhuman and Notion, which is the wrong bar entirely. Your user is comparing your product to a whiteboard and forty phone calls before lunch. Primitives vs. Abstractions (Aug 27): Developers want building blocks, systems of record prefer to hand out business logic, and both sides have a real case. There’s no test from the outside that separates the engineering reason from the competitive one, which is why this keeps ending up in court. Cross-industry Comparisons: The Two Kinds of Platform Power (Aug 18): Attention power and record power are different problems that keep getting handed the same regulatory toolkit. Congress’s latest swing at Big Tech shows both what’s possible and where it falls apart. The Dogs of (Platform) War (Aug 21): A CourtListener alert turned up a property management fight with healthcare’s exact shape and none of the Cures Act. Shell prospects, ghost accounts, a notetaker bot that came back to haunt someone, and a lawyer arguing both sides of the same theory in two states. Fractals All the Way Down (Aug 31): Yardi dominates property management right up until you zoom into one segment, where AppFolio owns it outright and Yardi doesn’t appear at all. Where you draw the line around a B2B software market is about to decide an awful lot of cases. Other News: If Judy Had the Courage (Aug 13): A short eulogy for the era when software was allowed to look insane: WinAMP, bold colors, and everything the grey chatbot era has taken from us. External Media: Portland Monthly Health Tech Meetup: Erin O’Brien channeled my own feelings - maybe it’s just because it’s August and beautiful, but the monthly edition of the PDX Health community meetup was a ripper. Make sure to reach out if interested to join for the next one. STAT’s Coverage of Epic This Month: Brittany Trang of STAT did a fantastic job of breaking the FTC investigation that’s been lurking, as well as UGM related coverage, so I had to give her her wish of a meme. Fall Conferences: I’m pumped to be kicking off my fall conference season in NYC at Nabla Accelerate next month, which Chrissy provided the link to apply for. Here’s the rest of my schedule in case you want to meet up: Commonwell (Redwood Shores, CA): Oct 13-14 Open@Epic (Madison, WI): Oct 21-22 eHealthExchange (Austin, TX): Oct 27 Sequoia Project (Austin, TX): Oct 28-29 HLTH (Vegas): Nov 15-18 RSNA (Chicago): Nov 29-Dec 3 Posts I Liked: HTI-6 Should Unbundle API Certification: Josh Mandel makes the case that HTI-6 should split (g)(10) into separate authorization and data certifications, so a PACS or a genomics platform can certify only the role it actually performs. Imaging is the urgent example, but the structure solves a much bigger problem. On implementing ePrior Auth with Epic: Scott Rossignol’s field notes from a live ePrior Auth build on Epic, including the two app registrations nobody warns you about and the CPT mapping problem waiting at the end. CMS-0057 implementations are going to be such a beast. Healthcare point solutions are starting to look a lot like streaming services: Spencer Dorn runs the cable-to-streaming arc against health IT and lands squarely on the bundling half of the cycle. Bundling can be good! It can also be bad! Life is nuanced. Health API Guy is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber. Subscribe
Four researchers win data harmonization competition
The U.S. National Science Foundation National Synthesis Center for Emergence in the Molecular and Cellular Sciences (NCEMS) at Penn State has awarded four researchers from across the world for their efforts in a competition to build machine learning-based approaches for streamlining publicly available datasets for reuse.
How squid survived Earth’s biggest extinction and took over the oceans
Scientists have finally cracked a long-standing mystery about squid and cuttlefish evolution by analyzing newly sequenced genomes alongside global datasets. The research reveals that these bizarre, intelligent creatures likely originated deep in the ocean over 100 million years ago, surviving mass extinction events by retreating into oxygen-rich deep-sea refuges. For millions of years, their evolution barely changed—until a dramatic post-extinction boom sparked rapid diversification as they moved into new shallow-water habitats.
USRA Contributes Planetary Science Expertise to NASA-IBM Lunar Foundation Model
Open-source artificial intelligence model combines diverse lunar datasets to support scientific analysis of the MoonThe NASA-IBM model reproduces patterns of lunar ice prospectivity (scaled from blue to yellow), shown at four locations (left) near the Moon’s pole. Top row: reference ice prospectivity map of Mons Mouton near the lunar south pol...
Energy Department wants to create science-specific AI models
The agency is collecting information from interested organizations that could provide foundational open-weight systems, scientific datasets for pretraining or fine-tuning expertise.
Driving large-scale hearing and balance research through data science investments, and an update on NIDCD’s January Advisory Council meeting
Enabling access to large-scale multimodal databases is essential to advancing science across all institutes at NIH. As both datasets and analytical methods become more complex, proper data infrastructure can enable multidisciplinary research efforts at a scale greater than that of individual studies alone. In line with NIH-wide strategic plans, NIDCD is making investments in data science that address long-standing research gaps, provide important resources for the research community, and preserve and protect participant privacy and data security.
Scientists were wrong about lifespan. Your genes matter way more than we thought
For years, scientists believed our lifespan was mostly shaped by environment and chance, with genetics playing only a minor role. But a new study from the Weizmann Institute flips that idea on its head, revealing that genes may actually account for about half of the differences in how long people live. By analyzing massive twin datasets—including twins raised apart—and using innovative simulations to filter out deaths from accidents and other external causes, researchers uncovered a hidden genetic influence that had been masked for decades.