SCROLL NEWS / DISCOVERY
Search headlines
Find the stories shaping the conversation.
Results for " predicts " 100 found 🔀 AI-powered Shuffle
Sign in to save shuffle history.
Germany’s Merz girds for a post-election fight for political survival
One conservative official predicts “all hell breaks loose” if the ruling CDU fails to win any seats in the regional election in Mecklenburg-Western Pomerania.
New method predicts where massive earthquakes will strike
UC Riverside scientists have developed a way to identify where Earth’s biggest earthquakes are most likely to occur.
Entertainment Industry Unites Behind U.S. Film Production Incentive
Entertainment industry groups launched the U.S. Film & TV Production Coalition to back a federal film production incentive. A new study predicts a big boost in jobs.
Buccaneers Hall of Famer Predicts Big Season From Rookie Bain
The Tampa Bay Buccaneers' pass rush should see major improvements thanks to the addition of Rueben Bain Jr., who was the team's No. 15 overall selection in the 2026 NFL Draft. See why Bucs legend DB Ronde Barber is so high on the first-year EDGE.
ESPN predicts outcome of Auburn's Week 5 matchup vs Tennessee
Will the Auburn Tigers pick up a much-needed SEC road win over Tennessee on Saturday? Here's what ESPN is saying.
ESPN analyst predicts Texas A&M vs. The Citadel
ESPN analyst Bill Connelly predicts that Texas A&M will defeat The Citadel 53-0 on Saturday.
ESPN FPI predicts Auburn's remaining games after Week 2 win
How many more games does ESPN FPI predicts Auburn to win after their Week 2 win versus Southern Miss?
Commanders draft national champion WR in latest 2027 mock draft
ESPN predicts the Washington Commanders will draft Indiana wide receiver Charlie Becker with the No. 10 overall pick in the 2027 NFL Draft.
Good science requires much more than evidence
Empiricists believe that scientific theories are true only if they make correct empirical predictions: if we can't verify these predictions, then there is no reason to expect that they tell us anything about reality. However, Professor of Philosophy at King's College David Kyle Johnson argues that this is mistaken. In response to Bas van Fraassen, who has written on this subject for the IAI, he argues that we have many reasons for believing a theory to be true, such as it being parsimonious and having large scope — qualities which have been used throughout the history of science as criteria for judging theories. In 2024, Princeton philosopher Bas van Fraassen wrote a short article for IAI News entitled “Science Does not Describe Reality: The Limits and Benefits of Explanation.” At the time, I was working on two projects (a science book and a Great Course) that, when completed, would (among other things) defend science’s ability to describe reality. I wanted to write a response to van Fraassen, but the projects demanded too much of my attention. But now that those projects are complete, I am in a position to explain the objections to van Fraassen’s thesis. Summarizing van Fraassen’s argumentIn a nutshell, his argument is this: What scientists do is make predictions based on theories and then invite others to test to see whether those predictions are borne out. “If theory T is right, result R should happen in experimental condition C.” Many think that running an experiment that shows that R happens in condition C is proof that theory T is true (and should be believed). But, van Fraassen argues, that R happened in condition C is only a reason to think that R happens in condition C. That’s the only observable, or empirical, result we can actually see, so that is all that we are really justified in believing. (The underlying theory that made that prediction still may or may not be true.) That is why he calls those who accept his position “empiricists.”Before I explain what is wrong with this position, let me explain what it gets right. What van Fraassen Gets RightFirst, a theory correctly predicting the results of an experiment is not, strictly speaking, “proof” that the theory is correct. The basic line of reasoning is this:If the theory T is true, result R should happen. Result R did happen. Thus, theory T is true.In other words:“If T then R. R. Thus, T.”This is a deductively invalid, fallacious mode of reasoning, called “affirming the consequent.” Those premises do not guarantee that the conclusion follows. “If I am in Pennsylvania, then I am in the USA. I am in the USA. Thus, I am in Pennsylvania.” Clearly that argument doesn’t guarantee its conclusion; neither does the above argument.However, science is not a deductive method of reasoning; it is an inductive one. And inductive arguments merely try to raise the probability of their conclusion, not “prove” them for certain. Thus, the fact that an experimental result doesn’t prove a theory is true is irrelevant; it’s not supposed to. That is not what scientific reasoning is about.___If the hypothesis that something exists clearly provides the best explanation for what we can observe, then the belief that it exists is what is rational.___Second, it’s also true that a single experimental result doesn’t provide enough justification to believe that a theory is true. Scientists do not come to a consensus about whether a theory is true by looking at a single experiment; they look at all the evidence to determine where the preponderance of evidence points. The mere fact that I am in the USA raises the probability that I am in Pennsylvania, but not nearly enough to conclude that I am in Pennsylvania. However, if I am also regularly encountering people wearing Eagles and Phillies hats and seeing advertisements for Founding Father tours, I probably am. In the same way, a theory successfully predicting the outcome of one experiment is not enough to conclude that it is true; but if a theory makes a wide variety of predictions that are all vindicated by multiple experiments, performed by many different groups and people, all actively trying to prove the theory false (which is what good scientific experiments are designed to do)—that is a good reason to conclude that the theory is true. In other words, if you repeatedly try to prove something false but can’t, that is a very good indication that it isn’t. The problem of underdeterminationIn reply, those whom van Fraassen calls “empiricists” often point to the “problem of underdetermination.” For any given theory that makes successful predictions about what happens in experimental conditions, there are countless other possible theories that make exactly the same predictions—and since they do, the experimental evidence in question is just as much reason to accept any of those theories as it is to accept our original one. Thus, the argument goes, we have no way of knowing whether our original theory is true, or one of the other “predictively equivalent” theories is true. SUGGESTED VIEWING How Occam's Razor changed the world With Johnjoe McFadden As an example of underdetermination, philosophers often offer this example. The evidence of fossils seems to provide reason to conclude that the universe is quite old; ancient fossils are what the “ancient universe” hypothesis predicts. But it’s also possible that God created an ancient-looking universe 200 years ago, complete with a “fake” fossil record for us to find, to make it look old. The existence of fossils is what that theory predicts too. So, the argument goes, the fossil record can’t be used as evidence that the universe is ancient; the fossil record supports the “200-year-old ancient-looking universe” theory just as much. The problem of underdetermination points out that, for any theory with supporting evidence, we could invent another completely different theory that predicts exactly the same evidence, and thus undermine that evidence’s ability to support the original theory.But this “argument from underdetermination” betrays a fundamental misunderstanding of what scientific reasoning is and how it works. To explain, let me use another rather strange (but very real and commonly discussed) philosophical thought experiment called “The Grue Problem.” The “Grue” ProblemSuppose I want to know what color emeralds are, so I observe a few, see that they all appear green, and form the theory that “all emeralds are green.” I thus predict that the next 100 emeralds I find will also appear to be green. And let’s say that happens, and so I thus conclude that my theory that all emeralds are green is true. Now, one might rightly point out that I don’t know for sure that all emeralds are green—I have not seen them all. Maybe some I haven’t found yet are blue. Fair enough. After all, we used to think that all swans were white, until we found some black ones. But let’s say that I go on a quest and collect every emerald in the universe—and, for the sake of the argument, let’s say I can do this, and can even know that I have every one of them. So, I collect them all, look at them all, and verify that they all appear green. Can’t I now rightly conclude that all emeralds are green?Inspired by the problem of underdetermination, a friend might argue that I cannot.Sure, they all appear green now—but there could be another kind of color called “grue.” Something is grue if it is green up to a certain point in time, but then spontaneously turns blue. And there could be an infinite number of “shades” of grue. Something that is one shade of grue will turn blue at noon today, but something that is another shade of grue will turn blue at noon tomorrow, etc. So yes, the emeralds appear green now; and since they haven’t turned blue yet, they are not some “past” shade of grue. But they might be some “future” shade of grue (e.g., they may turn blue at noon tomorrow). And since all future shades of grue are consistent with “being green up to this point in time,” the evidence you have is completely consistent with the notion that all emeralds are some future shade of grue, rather than green. Indeed, that they would appear green until now is what the “some future shade of grue” hypothesis predicts. So, we have just as much reason to conclude that all emeralds are some shade of grue as we do to conclude that they are green.You’re probably thinking “It’s not equally possible that the emeralds are grue; they are obviously green.” You are right! But explaining why your instinctive reaction to my friend’s argument is correct will help inform why van Fraassen’s “empiricist” is wrong and science actually does describe reality.The main mistake my friend makes is in thinking that the “empirical evidence” (of all the emeralds appearing green) is (or even should be) all I am using to draw the conclusion that all emeralds are green. It is not. I am also considering known facts about how colors typically work; things don’t usually change color, and when they do, they don’t do so suddenly, without cause, and all at once. So, even though the simple evidence of the emeralds all appearing green up to this point is consistent with them being some future shade of grue—that is even what the grue hypothesis “predicts”—it is still exceedingly unlikely that they are. The hypothesis that they are green is much more likely. Understanding science refutes the problem of underdetermination The misunderstanding about science that the “empiricist” pointing to “the problem of underdetermination” makes is similar to the one that my friend made. Science is not just about considering experimental evidence; in fact, it’s not even just about considering empirical evidence. As Ernan McMullin and many other philosophers of science have pointed out, the inference that makes science is “inference to the best explanation” (sometimes called “abduction” or “retroduction” by McMullin). While, depending on the circumstances, scientists use many different kinds of reasoning, science as a whole is, simply put, the process of considering multiple competing explanations and comparing them to certain criteria—explanatory criteria (the kinds of things explanations must be and do, by definition, to be a good explanation)—to see which explanation is the best.And these criteria do not deal only with empirical results. Yes, successful predictions raise the probability of a hypothesis, but it’s also the case that the more new assumptions a theory makes, the less likely it is to be true; every new assumption presents an opportunity for the theory to be false. (The fewer assumptions a theory makes, the simpler or “more parsimonious” it is said to be.) Likewise, the more existing evidence a theory contradicts, the less likely it is, because in order for a theory to be true, every piece of evidence it contradicts has to be somehow faulty. (The fewer established theories a new theory contradicts, the more “conservative” it is said to be.) And, obviously, the more something explains, the better explanation it is. (This is called having “scope.”) If a theory raises unanswerable questions, it’s not expanding our understanding, and thus is not wide-scoping. Scientists consider all these things when deciding which theory to accept, because the theory that adheres best to these explanatory criteria is most likely to be true.___the mere fact that a theory’s predictive success is why it “sticks around” or “becomes accepted” in the scientific community is not a reason to be skeptical about its truth.___The reason the grue hypothesis is no good is because it invents an entirely new kind of color—not just a new shade, but an entirely different category of color (a different way colors work)—and thus is not parsimonious. It raises unanswerable questions about how and why emeralds would just spontaneously change from green to blue (and thus does not have wide scope). And it contradicts what is already well established about how color works (and thus is unconservative).The theories that the problem of underdetermination points to do the same thing. Yes, the theory that God created an “ancient-looking universe” (complete with an ancient-looking fossil record) two hundred years ago is also consistent with the empirical evidence (of a fossil record) that presently exists; but that theory is monumentally unparsimonious (it requires many extra assumptions about infinite entities) and not wide-scoping (it raises unanswerable questions about why God would do such a thing). So, I can rationally conclude that it is false.Indeed, the history of science includes many examples of theories that were rejected because they lacked parsimony and/or scope, even before their falsity (and the truth of their competing theory) was confirmed by observation. As I point out in my book, geocentrism was rejected by the scientific community because it lacked parsimony, almost 200 years before the observation of parallax confirmed that the Earth moves around the sun. And the continuum theory of matter was completely rejected by 1860, because it couldn’t explain chemical reactions or organize matter into a periodic table, even though the experiments that eventually confirmed atomic theory weren’t even proposed until the 1870s. Solving the best of a bad lot problemIn response, the “empiricists” are likely to object, this time pointing to the “best of a bad lot” problem. “We have no reason to conclude that an explanation that has been shown to be better than others is true because we have no guarantee that the true theory was among those that we were considering.” If you have a “lot” (a collection) of poor explanations, even the best one is not true.The responses to this problem, however, are numerous. Peter Lipton argues that the set of hypotheses that scientists consider are not just randomly selected and thus unlikely to contain a theory that is true; they use existing knowledge to home in on what is most likely to be true. A doctor trying to explain abdominal pain is not going to bother wondering whether demons are causing it; they are going to stick with things like appendicitis and kidney stones, because they are most likely. Further, despite its name, inference to the best explanation is not just the process of comparing hypotheses and accepting the best one; we also want one that is a good explanation in its own right. If it’s not, we are going to keep looking until we find an explanation that is. And while that process does not guarantee that we will come across the exact truth, it’s very likely that the theory that emerges as the best will be at least approximately true, and thus we will be justified in believing it. And lastly, as I point out in my book, although many theories not based in scientific evidence (like geocentrism and continuum theory) have been completely overturned, science almost always progresses by the refinement of scientifically established theories—not their replacement. (Although our sun is not the center of the universe, heliocentrism was our first step in understanding the Earth’s true place in the universe.) So even if a theory we accept today may not be completely right, accepting it is a step on the journey toward the complete truth. The example of atomic theoryNow, it’s worth pointing out that most empiricists are not skeptical of all scientific theories; they are only skeptical of the ones that hypothesize the existence of “unobservables.” It’s not like they think that round Earth, heliocentrism, and germ theory are just useful (but untrue) “instruments” for making predictions about how to most efficiently navigate the seas, predict planetary motion, and treat diseases. They admit that the Earth actually is round, that the planets revolve around the sun, and that germs exist and cause disease. These facts and entities can be observed; the theories that suggest they exist are true. It’s only when theories start hypothesizing the existence of unobservables—like electrons, or quantum fields—that can’t be observed directly, that we should start doubting the theories’ truth. And to be fair to van Fraassen, in this published work, he doesn’t really say that we should doubt the truth of such theories. Only that scientific reasoning does not rationally force us to accept them as true. Skepticism, or agnosticism, he argues, is warranted. SUGGESTED READING Science is based in metaphor By Andrew Reynolds But the history of atomic theory throws a monkey wrench into that argument. Atoms used to be unobservable; again, scientists first accepted the theory that they exist because it had such grandiose explanatory power. But then atoms were seen with electron microscopes. They thus moved into the category of the observable. But this didn’t substantially change anything about what scientists were justified in believing; it certainly didn’t change any “atomic skeptics’” minds. It just confirmed what everyone already knew! If the hypothesis that something exists clearly provides the best explanation for what we can observe, then the belief that it exists is what is rational. The fact that the thing itself is not technically observable does not make doubting its existence rational.What’s more, the atoms were seen with an electron microscope. But electrons are themselves unobservable. How can an electron microscope reveal an atom if electrons are not real? This leads us to what is famously known as Hilary Putnam’s “miracle argument.” It’s (not) a miraclePhilosopher Hilary Putnam’s miracle argument is somewhat famous. Basically, he argued that a scientific theory being able to repeatedly make successful predictions—even about unobservables—without being at least approximately true would require a “miracle.” In other words, it would be such an improbable occurrence that it could not rationally be believed—the better explanation for why scientific theories about unobservables make the right predictions is because the unobservable entities and structures they hypothesize exist.Van Fraassen has addressed this argument. Indeed, in his original article, where he references Alison Gopnik’s point about orgasms not being “necessary for procreation” even though the desire for orgasm “drives the creation of progeny,” he is pointing to his own argument about how organisms are not naturally selected by evolution to get true beliefs; to the extent that belief-forming processes are selected for, the ones that are selected are the ones that are useful for survival—not the ones that produce true beliefs. In the same way, van Fraassen argues, scientific theories are accepted, not because they are true, but because they aren’t falsified; they make successful predictions; that’s what makes them “stick around.” So, the fact that they make successful predictions (and thus became accepted) is not a reason to think they are true.This argument is similar to C. S. Lewis’s “argument from reason,” which he used to argue against evolution: “The truth of evolution undercuts our ability to know that it is true.” Lewis’ argument famously failed; Elizabeth Anscombe, by Lewis’s own admission, demolished Lewis in a debate on the topic, and I have explained elsewhere why modified versions of the argument fail as well. In short, it’s because it fails to recognize that a belief-forming process can make an organism more likely to survive because it reliably produces true beliefs. It is therefore not the case that beliefs generated by naturally selected belief-forming processes are automatically false or unjustified.In the same way, the reason a theory “sticks around” could be because it’s not falsified, because it “makes correct predictions.” But the reason it makes correct predictions is most likely because it is true. (In pointing this out, I am echoing the arguments of Musgrave, Lipton, Psillos, and Kitcher, who refuted van Fraassen’s argument and analogy long before his 2024 article.) It would be a huge coincidence that it could do so without being true. And that is the point of Putnam’s miracle argument. Thus, contrary to van Fraassen’s suggestion, the mere fact that a theory’s predictive success is why it “sticks around” or “becomes accepted” in the scientific community is not a reason to be skeptical about its truth. ConclusionLet me close by saying something about the fact that empiricists don’t doubt the truth of all scientific theories, just the ones that hypothesize unobservables. In today’s society, this theory is dangerous to espouse publicly. The nuance between being skeptical about all of science, and just the highly technical theories about unobservables, is guaranteed to be lost to the average reader. And there is so much unjustified skepticism and doubt about the findings of science and experts that a headline like “Science Does Not Describe Reality” could easily fuel the anti-science movement, worsen climate change denial and cause the next pandemic.This is why I believe it was important to still respond to van Fraassen’s article, a full two years after he wrote it. Not only, as I have argued, does it represent a philosophically inaccurate view of how scientific reasoning works and what scientific reasoning can do, but it is a dangerous view to boot. It should thus be rejected.
Projecting Florida's College Football Playoff seed, results
Here's what seed CBS Sports predicts for the Florida Gators in the College Football Playoff, as well as the games, matchups and results.
Abdul El-Sayed is certain he’ll win Michigan’s Senate race. Some key Democrats quietly aren’t so sure
Abdul El-Sayed predicts that he will win the Michigan Senate race by 7 points.
TikTok Chief Expects Major AI Science Breakthroughs In Three Years
TikTok CEO Shou Zi Chew predicts artificial intelligence will drive massive scientific breakthroughs in medicine and physics within the next three years.
Online holiday sales expected to top $275 billion in 2026
With the Christmas holidays approaching and the shopping season underway, Adobe Analytics predicts more consumers will shop online, especially during the aggressive promotional period between Thanksgiving and Cyber Monday. Adobe...
Bilodeau’s Bakery Returns for Palmetto Sports Daily’s Locks of the Week
WATCH: Palmetto Sports Daily crew predicts their “Locks of the Week” for college football games including a cupcake selection and splitting on LSU-Ole Miss.
ESPN predicts Auburn's remaining schedule after SEC loss to Tennessee
How did Auburn football's 2026 outlook change following its Week 5 loss to Tennessee?
Home health, hospice deal-making on track to soar in 2027, report predicts
Despite concerns that a federal Medicare moratorium is stifling home health and hospice growth, a new analysis predicts merger and acquisition activity in the home health and hospice sector could be poised for a strong rebound in 2027.
Oklahoma vs. UTEP: Sooners Wire Staff Predictions
The Oklahoma Sooners open the season against the UTEP Miners on Friday night. The Sooners Wire staff predicts the outcome for Week 1.
ESPN’s models project just 7 wins for the NY Giants
ESPN predicts yet another losing season for the Giants, but are they setting themselves up for a surprise?
Supercomputer predicts if Juventus will make the top four
Juventus failed to finish in the top four at the end of last season, and the Bianconeri cannot afford to suffer the same fate again this campaign. That is why the club worked so hard during the summer...
Supercomputer predicts title race after Brighton beat Arsenal
Arsenal remain among the favourites to win the Premier League title at the end of this season, and the Gunners have maintained a strong standing among fans and neutrals despite their defeat to Brighto...
49ers Predicted to Bring Back $88 Million Super Bowl Champion
The Athletic's Matt Barrows predicts the San Francisco 49ers will bring back an $88 million edge rusher. The post 49ers Predicted to Bring Back $88 Million Super Bowl Champion appeared first on HEAVY.
College Football 27 predicts Alabama vs. Kentucky in Week 2
Alabama and Kentucky meet this Saturday to open SEC conference play. Here is who the College Football 27 video game predicts to win.
Analyst predicts Vikings pass rusher to be among league leaders
This ESPN analyst believes one Vikings pass rusher could be among the league leaders in sacks in 2026.
Michigan vs. Iowa Opening Point Spread Revealed
Michigan scored 52 against UTEP, but the opening line against Iowa predicts a low-scoring Big Ten battle at Michigan Stadium.
ESPN FPI predicts Texas A&M vs. Missouri in Week 6
ESPN FPI only gives Missouri a three percent edge of Texas A&M before Saturday's Week 6 matchup.
College Football 27 predicts Alabama vs. Tennessee in Week 7
Alabama travels to Tennessee this Saturday for a huge rivalry matchup. Here is who the College Football 27 video game predicts to win.
Iowa Business Council predicts economically positive third quarter
A new survey from the Iowa Business Council reports a positive economic prediction for the next six months. The IBC’s Economic Outlook Survey (EOS) for the third quarter was 0.1 points higher than the report for the second quarter, now reaching 64.91 points. If the points are higher than 50, the survey is reported positive. […]
Inflation expected to push holiday sales above $1 trillion
The National Retail Federation (NRF) predicts retail sales in November and December will total $1.013 trillion, up 4.2% over 2025 sales in the same period. 2025 holiday sales rose 3.1%...
Capitol News Illinois | Optimistic data-center report met with criticism
A new study predicts data centers expected to be built in Illinois over the next decade would create thousands of jobs and millions of dollars in property-tax revenue.
Michigan research council predicts legal challenges if money-out-of-politics campaign succeeds
The initiative would restrict political contributions by big utilities and state government contractors. But the U.S. Supreme Court has held that campaign contributions by corporations are protected speech.
Simple math formula predicts when AI chatbots will go rogue
Most of us now carry in our pockets devices capable of running small AI chatbots. These chatbots have little safety oversight to ensure they don't provide information and answers with the potential to ...
How England are building an attack to trouble Australia
Former England captain Michael Vaughan predicts England's new pace quartet will trouble Australia - so what is making them effective against Pakistan?
Scientists are about to test Einstein’s gravity with exotic matter
Scientists have found a new way to create a controlled beam of muonium, an exotic atom containing a heavier cousin of the electron. The advance could allow researchers to test for the first time whether gravity acts on second-generation particles exactly as Einstein’s theory predicts. Any unexpected difference would be a major surprise and could potentially point toward new physics, including a hypothetical fifth force.
Scientists find a simple size rule that predicts magnetic ground states
Researchers found three distinct magnetic behaviors that appear at specific crystal sizes across several unusual materials.
'Black Swan' investor predicts AI boom ends with crash of a lifetime
Universa Investments' Mark Spitznagel told BI that he expects one more "euphoric rally" followed by a "bigger crash than we've seen in our lifetime."
Defender Ng predicts big wins to come for Cardiff
Cardiff City defender Perry Ng says he expects the club to "batter some teams" moving forward as the Bluebirds finally put an end to their winless run.
ESPN predicts Memphis basketball’s floor and ceiling
On Tuesday, ESPN’s Myron Medcalf predicted the floor and ceiling for 13 mid-major programs, including the Tigers. Medcalf predicts a floor of 15 wins and a ceiling of 19 wins for Penny Hardaway’s team.
USA TODAY Sports predicts Florida vs Ole Miss outcome in Week 4
Here are the picks from USA TODAY Sports' college football experts for the Florida Gators vs Ole Miss Rebels' battle in the Swamp in Week 4.
Early childhood BMI trajectories predict later cardiometabolic health risks
Childhood obesity predicts later cardiometabolic disease, but single-time weight measures may miss dynamic growth information.
Byron Donalds predicts conservative Democrats will back him over David Jolly
'They're saying, "I don't want to be a part of this."'
ESPN predicts Texas A&M Week 4 road matchup vs. LSU
Texas A&M enters Saturday's primetime road test vs. LSU with a 37.2% to pull off the upset, per ESPN.
Engrams Embedding Entendre: Codesign for Efficient DRAM/SSD Offloading
Engram extends standard token embeddings with learned multi-token lookups. Recurring local patterns retrieve vectors directly, reducing the need to reconstruct them through attention and feed-forward layers. With Engram model architecture optimization, it allows for lower HBM capacity to be needed for models at the same quality. This does not mean there won’t be an insane demand for HBM but it just means that model architecture will continue to innovate around constraints. This model architecture design is naturally codesigned for parameter offloading: each token accesses a few embedding rows whose addresses depend on token IDs, not hidden states. The runtime can prefetch those rows from host DRAM while earlier layers compute, keeping the table outside HBM without transferring entire weight matrices. Our Memory model contains our latest estimates of quarter by quarter HBM, DRAM, & NAND supply and demand. Offloading frees HBM for model weights and KV cache, potentially supporting larger batches or more concurrent sessions. When DRAM becomes the next constraint, NVMe offers another tier. Recommendation systems already cache frequently or recently accessed embedding rows in faster memory while backing colder rows with SSDs. After NVIDIA roadmap had to change due to massively despec’ing Rubin Ultra from 1024GB to now ~200GB of HBM per chip, model architecture optimizations like emgram maybe helpful. Our DeepSeek-V4.1-Flash configuration uses roughly 189 GiB of memory for Engram. We replace it with a memory-mapped (mmap) file and measure serving performance with offload to SSD. Later on in our report, we will show our Engram offloading experiments along with the official InferenceX agentic inference serving results on Engram models like DeepSeekv4.1 Flash across all 6 NVIDIA GPU SKUs along with MI355X. Unsurprisingly, the CUDA Moat is still mogging MI355X on the ultra popular DeepSeekV4.1 Flash model. We also show how even on high capacity HBM SKUs, offloading emgrams to DRAM could result in even better performance for most of the pareto than keeping the emgram in HBM. Our benchmark has been widely reproduced, validated and/or supported by almost every major buyer of compute from Google Cloud to Microsoft Azure to Oracle, to Meta and many more. Furthermore, it has the support of the ML community including from vLLM, LMCache, SGLang, PyTorch, Huggingface and the support of major labs like OpenAI, MiniMax, ZAI, Qwen, Moonshot Kimi, etc. Star the InferenceX GitHub repository if you find the open-source benchmark and data useful!. InferenceX is the only inference benchmark in the world to have TPUv7, Jalapeño, Nvidia Rubin NVL72, AMD, and soon, SambaNova and Trainium. Due to how realistic AgentX scenario is to real world agentic inference workloads, AMD has committed to collaborating on MI455X UALoE72 too. DeepSeek did not release the original paper’s two trained Engram models. We replicated its setup on fineweb-edu using the released code and training hyperparameters, at an estimated 6E18 FLOPs per run. We observed the same U-shape scaling: Engram improved performance over pure MoE baselines. We also reproduced DeepSeek’s results, where earlier-layer representations with Engram resembled those of later layers. Like the original Engram paper, it is possible to probe Engram’s gate scores to see what n-grams DeepSeek-V4.1-Flash makes the most use of. Our gate scan finds names, code fragments, relational phrasing, and boilerplate. These examples prioritize interesting-ness over gate strength. One unexpected result was Wright : Ace Attorney. These examples suggest learned memory optimizes the training objective, not a judgment of which facts deserve storage. Licenses, bibliography fragments, API scaffolding, and website furniture can provide prediction shortcuts, so the value of additional Engram capacity may depend on what survives data preparation. This does not show that table capacity is “wasted”: the evaluation-corpus scan establishes neither training exposure nor the capacity occupied by each category. For offloading, strong gates do not identify cache-hot rows. Low gates do not automatically save reads either: computing the gate requires the retrieved key and undoes the performance gain of a fused kernel. Skipping reads would require a separate usefulness predictor before retrieval. In the original paper’s inference-time ablation, factual-knowledge benchmarks retained just 29–44% of their original performance, while reading comprehension retained 81–93%. This is due to the training–inference mismatch. The resulting degradation therefore measures this trained model’s dependence on Engram, not the performance difference between models trained with and without it. In our ablations, suppressing Engram worsens token likelihood across all evaluated domains, especially encyclopedia text and several code corpora. Surprisingly, GSM8K accuracy stays within measured run-to-run variation and removing Engram has no effect. Engram is not a detachable dictionary beside an unchanged MoE. Removing it changes downstream features and expert selection. We tested whether rerouting hurts or compensates by holding tokens fixed in a teacher-forced experiment on CRUXEval, a code-reasoning benchmark of small Python functions where the model predicts a function’s output from its code and an input, and scoring the reference answer. Removing Engram raised answer loss from 0.2848 to 0.3093 bits/token. Forcing the ablated model to use the original Engram-on expert choices made it worse still, at 0.3375 bits/token. Rerouting partially compensates for the missing memory. Memory features and expert selection work together, rather than following a clean “memory stores facts; experts reason” division. On the same CRUXeval, removing Engram during either phase reduced accuracy and increased generated tokens; removing it throughout produced the largest changes. Keeping Engram for prefill leads to more correct answers than keeping for decode likely due to semantically richer KV cache transferred to decode workers, allowing it to mitigate some of the performance loss. Engram’s table is large, but each lookup is small. DeepSeek-V4.1-Flash requests 24 rows at each of two Engram layers, about 12.4 KiB per processed token position across the model, or 3.1 KiB per GPU when split across four GPUs. Currently as of Day 7 since Model Release, MI355X is still 2-4x worse performance per dollar compared to B200 even when normalized by Mi355X’s lower TCO. Our full total cost of ownership breakdown comes from our AI Cloud TCO Model along with monthly market surveys of over 100+ gpu clouds & gpu cloud customers. On the Day 0 release of DeepSeekv4.1 Flash, NVIDIA vLLM works out of the box with zero issues across all 6 SKUs: H100, H200, B200, B300, GB200, GB300! This was thanks to the amazing work by the NVIDIA & Interact teams! In comparison, AMD vLLM did not work on day 0 for DeepSeekv4.1 Flash.Source: SemiAnalysis InferenceX AMD’s vLLM documentation points to using vllm/vllm-openai-rocm:deepseekv41-flash-0909, but from hour 0 of the model release to hour 23, AMD has not publicly released the image. AMD claims, “SPEED IS THE MOAT,” yet it has still not released it by the 23rd hour. We wish that, going forward, the AMD team has a better process for hour 0 model releases. Eventually when they did publicly release for “day 0” image support, performance-wise, it is currently up to 14.8x worse perf per dollar than H200 and up to 42x worse perf per dollar than B200/B300. The power of the CUDA MOAT is NVIDIA’s collaboration with its massive 6 million-developer community ecosystem including most of the vLLM & SGLang & Tokenspeed maintainers which means that CUDA is optimized on day 0. Overall, AMD did make significant improvements but the performance per dollar is still currently 2-4x worse than B200. AgentX Engram DRAM Offloading Improving Performance The HBM and DRAM offload use the same GPU kernel to select and dequantize rows. With HBM, it reads GPU memory; with Unified Virtual Addressing (UVA), it reads pinned host memory directly. Both support full decode graphs. Moving the table into HBM accelerates only the sparse lookup, leaving decoder computation and communication unchanged, resulting in little overall benefit while consuming memory otherwise available to KV cache. Another benefit of Engram offloaded to DRAM which means you can reduce the communication overhead by using less HBM GPUs per replica. For example, when enabling Engram offloading on B300, we are able to switch from TP4 to now TP2 which improves the pareto curve by up to 1.6x. When iso-model quality, less HBM is required as the engrams could be offloaded to host DRAM. Thus HBM bandwidth matters way more than HBM capacity. For inference workloads where memory bandwidth matters the most, 4-hi HBM provides the best $/bandwidth and therefore lowest cost per token. If China continues to make more and more revolutionary model architecture innovations, soon it could potentially 0Hi HBM stacks. [ ](https://newsletter.semianalysis.com/p/long-live-the-short-king-why-4-hi) [ Long Live the Short King: Why 4-hi HBM Wins ](https://newsletter.semianalysis.com/p/long-live-the-short-king-why-4-hi) Myron Xie, Bryan Shan, and 3 others · Sep 13 [ Read full story ](https://newsletter.semianalysis.com/p/long-live-the-short-king-why-4-hi) Moreover, on B300 and week-0 stack, moving the Engram table back to HBM did not improve results and stay within run-to-run variance. This is a result of the work optimizing DRAM offload, such as async, and overlap. SSD Offloading On B200, we created a unoptimized vLLM fork and stored the Engram tables in memory-mapped files on local SSDs. File backing lets the OS reclaim table pages when other applications need RAM. Pages already cached in memory can be served without reading the SSD again. Furthermore, note that we were unable to turn on GDS The unoptimized SSD implementation changes how rows reach the GPU. It copies row IDs to the CPU, deduplicates them, gathers the requested rows into pinned buffers, copies those rows back to the GPU and dequantizes them. This work runs between segments of the GPU execution graph. Native UVA performs row selection and dequantization directly on the GPU, avoiding the CPU round trip. A warm filesystem cache removes physical SSD reads, but leaves the coordination, row gathering and transfers. This is why a file already cached in RAM can still perform worse than a pinned DRAM table. The comparison measures the whole serving path; it does not separate the time spent on each of these operations. B200 DRAM dominates both measured SSD serving curves in total tokens per dollar and P90 interactivity. Near 125 tokens/s/user, DRAM delivers 121 million total tokens per dollar versus 52 million for SSD. For production serving, SSD offloading is likely not worth the tradeoff. On the B200 configurations we measured, SSD offloading loses on both measures: every observed SSD point has a DRAM alternative that delivers higher P90 interactivity and more total tokens per dollar. The only points where Cheaper storage does not automatically produce a cheaper inference service. Moving Engram to SSD leaves the same four expensive GPUs and the rest of the server in place. Reclaiming RAM only creates an economic benefit if it enables a cheaper server configuration or additional useful capacity. The current unoptimized path provides neither benefit, and the filesystem cache still consumes RAM when table pages are resident. Next we will look at the mechanisms and specific implementation of ngrams from DeepSeekv4.1 Flash, LongCat, Qwen3.8 Flash Next.
Random ESPN simulation predicts dream 2026 Detroit Lions season
One simulation has the 2026 Detroit Lions winning the Super Bowl in an absolute dream fashion.
2027 NFL mock draft: 49ers finally address long-neglected position
USA Today Sports predicts the San Francisco 49ers will add to their secondary with the No. 19 pick in the 2027 NFL draft.
Gordon and Miller Answer Objections to Cosmic Fine-Tuning
Fine-tuning strongly favors theism because it predicts a life-permitting order in advance, while naturalism only retrofits explanations after the fact.
The Athletic predicts upset in Florida's SEC opener against Auburn
Florida enters Saturday as a road favorite, but The Athletic predicts Auburn will hand the Gators their first loss of the season.
ESPN predicts FSU's remaining schedule after Louisville loss
ESPN’s updated Matchup Predictor lowered Florida State’s win probability in each of its final six games following the Seminoles’ loss to Louisville.
Blue Ribbon Yearbook predicts Memphis will win the American
Blue Ribbon Yearbook released their college basketball preseason magazine for the 2026-27 season, and they predict Memphis will win the American and make the NCAA Tournament.
The Athletic Predicts Red Wings Bottom Out This Season
The Athletic predicts Detroit bottoms out and picks top five in the 2027 draft. #RedWings
CBS Sports predicts Falcons 'light up' Bears defense in Week 6
CBS Sports predicts winner of Sunday's Falcons-Bears matchup in Atlanta
Bucs Predicted to Replace QB Baker Mayfield With ACC Superstar
Sports Illustrated predicts Bucs will replace QB Baker Mayfield with 2027 1st-round pick. The post Bucs Predicted to Replace QB Baker Mayfield With ACC Superstar appeared first on HEAVY.
Mavs Trade of Kyrie Irving? 'Makes Sense,' Predicts NBA Analyst
Kyrie Irving was named the Mavericks player who should be traded as Dallas balances his return with Cooper Flagg's long-term timeline.
Sunderland v Brighton: Our Team Predicts a Tactical Battle
Brighton arrive at the Stadium of Light in excellent form, but can Sunderland build on their recent performances and claim a crucial home win?
Millions Will Lose Medicaid Because They Aren't Able To Get Enough Work Hours, Study Predicts
By Dennis Thompson HealthDay ReporterWEDNESDAY, Sept. 2, 2026 (HealthDay News) — About 1 in every 5 eligible adults in Medicaid expansion states might lose their benefits due to new work requirements that take effect in 2027, a new study says. These folks likely will be unable to fulfill the minimum ...
Why CBS Sports Predicts Omarion Hampton Will Win NFL Rushing Title in 2026
The second-year running back should thrive in Mike McDaniel's system this season
ESPN predicts the winner of Georgia football at Alabama
ESPN predicts who will win the game between the No. 2 Georgia Bulldogs and the No. 7 Alabama Crimson Tide
ESPN analyst makes bold prediction about Titans' free agent signee
ESPN analyst predicts Tennessee Titans cornerback Alontae Taylor will struggle despite his lucrative new contract.
USA TODAY Sports predicts Florida vs Missouri outcome in Week 5
Here are the picks from USA TODAY Sports' college football experts for the Florida Gators vs Missouri Tigers battle at Faurot Field in Week 5.
CBS Sports analyst predicts 'Giants will make the Super Bowl' in 2026
A CBS Sports analyst boldly predicts the New York Giants will make the Super Bowl in 2026.
Vikings Trade Proposal Predicts Price For Eagles’ Saquon Barkley
Howie Roseman is ready to deal if the right offer is on the table for somebody on the Philadelphia Eagles‘ roster. The General Manager proved that much when he decided to shockingly move the veteran offensive lineman Cam Jurgens to the Baltimore Ravens earlier this week. Could it soon be time for the Eagles to […] The post Vikings Trade Proposal Predicts Price For Eagles’ Saquon Barkley appeared first on HEAVY.
Phoenix Suns owner Mat Ishbia predicts team will win opening round playoff series — and the NBA season hasn't even started
Phoenix Suns owner Mat Ishbia predicts first-round playoff win, says UWM stock decline won't affect team ownership.
ESPN predicts outcome of Auburn's Week 3 matchup vs Florida
Will Auburn football win its SEC opener on Saturday?
Analyst predicts when the Browns may bench Deshaun Watson
Could the Browns bench Deshaun Watson at some point despite early season success.
Cowboys’ $113 Million Edge Rusher Predicted for Career Breakthrough
The Athletic's Jon Machota predicts $113 million Dallas Cowboys edge rusher on verge of career breakthrough. The post Cowboys’ $113 Million Edge Rusher Predicted for Career Breakthrough appeared first on HEAVY.
CBS Sports predicts Bucs will fall short against Bengals in Week 1
CBS Sports’ Pete Prisco expects a close Week 1 battle but has the Bengals handing the Bucs a 29-28 loss in Cincinnati.
David Pollack predicts winner of Oklahoma-Michigan
David Pollack predicts Oklahoma will beat Michigan 24-17 after questioning the Wolverines' Week 1 performance.
Giants Urged to Trade Draft Pick for Dolphins Edge Rusher
Bleacher Report predicts the New York Giants could trade draft pick for Dolphins edge rusher. The post Giants Urged to Trade Draft Pick for Dolphins Edge Rusher appeared first on HEAVY.
Cowboys insider predicts veteran not on team will make roster
Corner could be a spot where the Cowboys look to add someone new
Julie Chen Moonves weighs in on Devens, Drew, and Taylor’s ‘Big Brother’ finale chances
Big Brother host Julie Chen Moonves weighs in on Devens, Drew, and Taylor’s finale chances and predicts who she thinks will win America’s Favorite Houseguest.
‘Crystal Ball’ – Sports Media Analyst Predicts Fox Will Lose Greg Olsen, Prompting a Drew Brees Promotion and ESPN Star Getting Poached
With the NFL season well underway, it appears the broadcasting world could see a couple of shakeups. With Tony Romo’s exit from CBS, attention is also turning to other player-turned analysts, including Greg Olsen’s situation at Fox, which is expected to create a ripple effect in sports broadcasting.
ESPN predicts huge honor for Falcons defender
Is Xavier Watts a future All-Pro? One ESPN analyst thinks so
Gary Neville predicts the outcome of Arsenal v Chelsea
Gary Neville has predicted how he expects the Chelsea-Arsenal game to end as two unbeaten clubs prepare to meet in the Premier League this weekend.Arsenal secured a hard-fought 1-0 victory against Ast...
College GameDay predicts Georgia-Alabama winner
ESPN's College GameDay predicted the winner of the SEC championship rematch between the Georgia Bulldogs and Alabama Crimson Tide.
The Pants Predicts: Iowa Hawkeyes vs Northern Illinois
The Iowa Hawkeyes are a big favorite over NIU in their season opener. Our staff predicts the outcome vs the spread.
Broadband access predicts health tracker use, UWM research suggests
As Fitbits, Apple watches and Oura Rings grow ubiquitous, UWM research suggests digital inequity doesn't allow them to reach those who could benefit most.
NBA 2K27 predicts the season: LeBron's shocking start with 76ers, a new MVP and an NBA Finals rematch
Ducks Wire staff predicts Oregon Ducks vs. Nebraska Cornhuskers
Ducks Wire staff predicts Oregon vs. Nebraska, including final score picks, players of the game, and key matchup predictions.
Browns vs. Jets Week 5 picks: A 50/50 split
The DBN staff predicts the Week 5 games, plus the national media make their predictions for Browns vs. Jets.
NFL Expert Predicts Patriots Regression After Super Bowl Run
Sporting News predicts the Patriots will surrender the AFC East and fall short of another Super Bowl appearance despite winning a playoff game.
NFL analyst predicts large role for unexpected 49ers rookie
The 49ers surprisingly took a running back in the third round of the draft, and at least one analyst expects him to have a productive rookie year.
Eagles Get Important Zach Ertz Prediction Before Commanders Game
The Philadelphia Eagles are heading into Week 1 of the 2026 NFL season without making any changes to the tight end room so far. While there’s been some smoke surrounding a reunion with the former Washington Commanders pass-catcher Zach Ertz, the team hasn’t gone in that direction at this time. One Eagles insider predicts that […] The post Eagles Get Important Zach Ertz Prediction Before Commanders Game appeared first on HEAVY.
Peter Schiff predicts economic 'downturn,' higher oil prices: 'I don't think it's over'
Peter Schiff indicated during an interview that he expects oil prices to head higher regardless of whether the Iran war ends, and thinks that there will be an economic "downturn."
Seahawks Sam Darnold Injury Update: Sports Doc Predicts ‘Realistic’ Return Date
A sports doctor predicted a “realistic” return date for Seahawks quarterback Sam Darnold as he works his way back from injury. The post Seahawks Sam Darnold Injury Update: Sports Doc Predicts ‘Realistic’ Return Date appeared first on HEAVY.
North East Ambulance Service predicts mental health spike
North East Ambulance Service says mental health care needs will rise over the next few years.
CBS Sports predicts disappointing 2026 season for Bucs
CBS Sports' John Breech has the Buccaneers finishing 7-10 and losing the NFC South to the Saints in his 2026 NFL predictions.
Analyst gives bold prediction for Titans in 2026
CBS Sports predicts the Tennessee Titans will go from worst to first in the AFC South during the 2026 NFL season.
Alim McNeill Lands Bold Pro Bowl Prediction From The Athletic
The Athletic predicts Alim McNeill makes his first Pro Bowl for the Lions in 2026. #Lions
Gary Neville predicts Chelsea result against “really aggressive” Arsenal
Chelsea Arsenal is this weekend’s big game – and pundit Gary Neville has already had his say on the result of the match.If you enjoy Talk Chelsea coverage and want to see more of it, add us as a p...
ESPN predicts Jaguars miss playoffs in 2026 NFL season
One ESPN analyst is predicting that the Jaguars miss the playoffs. Here is why.
ABI predicts employer and financial stability for fourth quarter
The Iowa Association of Business and Industry reported that surveyed Iowa employers are looking towards 2026’s fourth quarter with a “neutral outlook”, approaching a stable economy and financially healthy end of the year. Each quarter, ABI surveys its board of directors in order to track business sentiment, identifying common concerns and inform upcoming priorities for […]
Supercomputer predicts Chelsea vs Leeds United in Carabao Cup
Xabi Alonso looking straight with concerned Daniel Farke ahead of Stamford Bridge background
New survey predicts trouble for Trump ahead of midterms
With just over five weeks until the midterm elections, CNN’s John King explains the key indicators he’s tracking.
The fitness test that predicts your risk of death
How quickly your heart rate returns to normal after exercise can reveal a surprising amount about your underlying health.
As shortfall looms, Pillen installs allies on board meant to keep politics out of budget forecast
Decades ago, lawmakers created a board that predicts Nebraska's tax receipts in an effort to keep politics out. Last week, Gov. Jim Pillen appointed two political allies to serve on it
Cowboys Prediction Puts Undrafted Preseason Star on 53-Man Roster
ESPN's Todd Archer predicts the Dallas Cowboys will keep UDFA rookie star on final 53-man roster. The post Cowboys Prediction Puts Undrafted Preseason Star on 53-Man Roster appeared first on HEAVY.
'GMFB' predicts OPOY winner for 2026
"GMFB" predicts who will win Offensive Player of the Year for the 2026 season.
As shortfall looms, Pillen installs allies on board meant to keep politics out of budget forecast
In an unprecedented move, Gov. Jim Pillen has reshaped the state board that predicts Nebraska’s tax revenues for budget writers, replacing three members with appointees that include two former Republi...
Want To Know If Someone Likes You? Pay Attention To How They Sway
Oddly, it predicts interest in a long-term relationship, but not short-term relationships.
Ed Yardeni predicts Fed interest rate hikes amid persistent inflation
Yardeni Research President Ed Yardeni joins Liz Peak to discuss Federal Reserve policy. Yardeni predicts a tightening cycle with multiple interest rate hikes over the next year to bring persistent inflation down to the target.
USA TODAY Sports predicts another rough season for Colorado football
USA TODAY Sports predicts Colorado to have another disappointing year, a finish that could raise questions about Deion Sanders' future.