research
One layer creates the giant activations behind attention sinks News
Researchers identified a single layer, consistent across model families, where the outsized activations that produce attention sinks first appear, and showed that loosening that token's rigidity improves instruction following and math reasoning without retraining.
MIT found brain-like modules inside six large language models News
MIT researchers localized the neurons behind 46 reasoning tasks in six large language models and found that tasks sharing a brain network in humans share neurons in the models, with 4.3 times more overlap within a cognitive domain than across domains.
Compressed memory stretched a 7,000-token model to 1.75 million News
A framework that chunks long documents, compresses each chunk into memory blocks and gates which blocks the model reads extrapolated from 7,000 tokens of training context to 1.75 million at inference, with half the peak GPU memory of a leading baseline.
Agent skill libraries now need a librarian, not a folder News
SkillsVote profiles a million-scale corpus of open-source agent skills and admits new ones only through evidence-gated updates after execution, arguing that indiscriminate additions to a skill library pollute every future run's context.
A 9B model writes agent upgrades as good as Claude Opus 4.6 News
A 17-author study separates the ability to improve an AI agent's scaffolding from the ability to benefit from the improvement, and finds that a 9-billion-parameter model produces upgrades yielding gains comparable to Claude Opus 4.6.
Someone compiled a working computer into transformer weights by hand News
A team constructed transformer weights analytically rather than training them, producing a model that runs arbitrary C programs through a WebAssembly interpreter encoded entirely in its attention layers at about 30,000 tokens per second.
SkillZip compresses an agent's skill file without ever running the agent News
A new method compresses the accumulated skill files of self-evolving agents by finding the shortest structural explanation that preserves every contract element, with no test rollouts required.
One checkpoint turns a compatible video model into a 4D world builder News
Researchers skipped the pixels entirely, feeding a video model's final internal representation straight into a 4D decoder, and got a single checkpoint that works unchanged across multiple video generators after training on about a thousand clips.
An AI tightened a 70-year-old constant, and the paper says its judgment was the weak part News
A case study from seven researchers documents how an AI system helped tighten the best known bounds on the Grothendieck constant, and reports plainly that the system was strong at technical execution but weak at research judgment and at tracking where the work stood.
Agent instruction files triple in size because nobody remembers why a rule exists News
A study of 247,694 instruction lifetimes across 1,867 GitHub repositories found agentic coding instruction files grow 226 percent over their lifetime and effectively never shrink, because deleting a rule whose rationale is lost risks breaking something.
A self-improving coding agent that compares notes with a rival lineage News
Most self-improving coding agents rewrite themselves after a single failure, throwing away the archive of everything they have already tried; a new method adds two edit operations that use multiple trajectories and a competing agent's evidence instead.
A video model counted events correctly two-tenths of one percent of the time News
Asked to count simple events in short synthetic clips, Google's Gemini 3.6 Flash got the final count right 0.2 percent of the time in the hardest setting and recovered only 18 percent of the events that actually occurred.
A diffusion model picks its answer a fifth of the way through News
Researchers logged every token commitment in a masked diffusion language model and found it locks in the final answer 15 to 24 percent of the way through generation, while half the reasoning is still blank -- so the visible reasoning is written around a frozen conclusion.
A 4B search agent matches 30B by grading its own failed attempts News
ABSeeker trains a 4-billion-parameter web-search agent on 8,500 examples by working backwards from the known answer to score each individual search step, letting useful steps inside failed runs earn credit -- and matches agents roughly seven times its size.
Vision models can tell which camera took a photo, and they use it to cheat News
A paper accepted to ECCV 2026 shows that image encoders pick up invisible pixel-level traces left by cameras and image processing, then lean on those traces as a shortcut for the actual task, degrading when the equipment changes.
The AI judges grading computer-use agents are too easy on them News
A new benchmark finds that vision-language models used to grade whether a computer-use agent finished its task systematically accept failed runs as successes, and that judgment quality varies more across operating systems than across judges.
An agent's memory records what you said; this compiler records what you did News
A single-author paper argues computer-use agents waste frontier inference re-deriving routines their user already performed, and presents a model-free compiler that turns raw screen capture into a context block 86 times smaller that an agent reads with 98.4 percent accuracy.
A task factory ran fifteen rounds and broke the model grading it News
A new paper builds harder and harder terminal tasks by recursively rewriting accepted ones, and across fifteen rounds a fixed frontier solver's success rate fell from 90 percent to 2.5 percent, with the authors reporting no ceiling in sight.
A model small enough to run on a laptop reads heard speech from brain scans News
Researchers rebuilt a brain-to-speech decoder around the physical geometry of the scanner helmet, shrinking the decoder about twentyfold while matching prior accuracy, and then used it to identify which properties of speech the brain signal actually carries.
Models invent about 40 percent of what they claim to know about you News
A new study measuring 12 language models across more than 143,000 judged claims found every one of them invented or stereotyped between 35 and 49 percent of what it asserted about a user, and that the models most confident they were being careful were the ones fabricating most.
A portable persona skill carried one user's verbal tics into unrelated conversations News
Researchers benchmarking persona skills, the reusable artifacts agents distill from your interaction history, found that agents equipped with one reproduced that person's distinctive phrasing in unrelated conversations up to 87.7 percent of the time, and that a watermarking defense meant to prove provenance detected nothing at all.
Two Essays About AI and Your Brain, and One Actual Study News
A randomized experiment found that AI assistance impaired conceptual understanding, code reading, and debugging while delivering no significant average speed gain, which supports the concern behind this week's viral developer essays but not their proposed fix.
Four agent-memory papers landed in a week, and none tested what happens when an attacker controls the writes News
Four papers published within days define an AI agent's memory as four incompatible things - a pretrained module, a rewritten lesson, a folder of files, and a reliability ledger - and three of them introduce writable state that determines future behaviour without evaluating an adversary who controls what gets written.
DistillAlign explains why fast video models get prettier and more repetitive at the same time News
A new preprint shows that the standard technique for speeding up video generators optimises a mathematical objective that deliberately concentrates on safe, common outputs, and that adding a diversity-preserving loss back during that stage more than doubled the paper's coverage measure without giving up the quality gain.
The non-sofic group is the one OpenAI claim a computer can check News
Chapter 3 of OpenAI's new manuscript claims to have constructed a non-sofic group, settling a long-open question, and ships roughly 34,000 lines of Lean code with no unproved placeholders so outsiders can verify it.
Seven new papers cannot agree what a world model is made of News
PhiZero compresses video into a learned alphabet of state transitions, but six competing late-July papers argue the substrate should instead be executable simulator code, explicit game state, paired re-renderings, distribution coverage, architecture, or action intent.
OpenAI publishes ten mathematics claims with Lean proofs and no named authors News
OpenAI released ten claimed advances in mathematics and theoretical computer science today, produced by an unreleased internal model it calls Astra, with a 249-page manuscript collection and machine-checkable proofs for every result.
AI financial advice works when you hand it a full financial plan News
Researchers simulated the lifetime finances of 1,000 demographically representative US adults following chatbot advice and found the results move toward a life-cycle benchmark - but only when prompts supply complete data, fixed assumptions and hard arithmetic constraints.
Training on the best of K guesses is a third scaling axis alongside parameters and data News
A paper from UIUC and Harvard shows that generating several candidate outputs per training example and learning only from the closest match improves sample efficiency 6.2-fold, and that the benefit grows rather than shrinks as models and datasets get bigger.
One planted document flipped more than half of deep-research reports to a false conclusion News
Researchers built 5,933 credible-looking but factually false documents and slipped exactly one into the retrieval pool of several deep-research agents; the rate at which final reports endorsed the false conclusion went from zero to 54.7%.
Metis puts an agent's memory inside the model instead of in a database beside it News
A team spanning several Chinese universities released the first prototype of what it calls a memory foundation model - a backbone carrying a memory state that updates on every interaction through a plain forward pass, with no gradients and no external retrieval system.
Letting an agent organise its own memory halved retrieval cost and improved no answers News
The first systematic study of the markdown-filesystem memory that deployed AI agents actually use found that organisation reliably buys cheaper search, but that no agent tested converted a tidier memory into better answers - and organisation eroded as the store grew.
Asking a model to check its own work lost every comparison against just sampling more answers News
A controlled experiment counting every token spent found that seven popular self-improvement methods - self-refinement, reflection, debate, best-of-N selection - never reliably beat the simplest baseline of generating several answers and taking the most common one, and ten of them were reliably worse.
An open 35B model trained to evolve its own machine-learning code nearly doubled its base model's medal rate News
Frontis-MA1, released with full weights and stack, raises its base model's medal average on a machine-learning engineering benchmark from 39.4% to 60.6%, and to 71.2% with a stronger search - all within a 12-hour budget on a single consumer GPU capped at 12GB.
13.6% of SWE-bench Verified pairs a bug report with a patch that does not match it News
A systematic audit of SWE-bench Verified, the benchmark used everywhere to rank AI coding ability, found that 68 of its 500 tasks link an issue to a pull request that fixes something else, adds unrelated work, or only partly addresses the report.
AI search agents get better when relevance tells them where to look, not what to read News
Researchers at Tencent rebuilt relevance as a guide for how a search agent traverses a corpus rather than as a ranked list of documents, cutting the agent's tool calls by roughly a sixth while raising accuracy.
Two papers attack the same waste: coding agents rediscovering the same repository every session News
CodeNib builds reusable lexical, semantic and structural views of a repository per commit and cuts an agent's exploration tokens by 50 to 87%, while a companion benchmark finally measures the file-finding stage that patch-success scores hide.
A new distillation method lets the teacher model interrupt the student mid-thought News
Researchers found that when a student model starts reasoning down a wrong path, its teacher's next word tends to be a redirection like But or Wait, and turned that disagreement into an automatic trigger for the teacher to briefly take over.
A new benchmark grades video models on film craft instead of whether clips look nice News
FilmBench scores text-to-video and reference-to-video models against professional cinematic criteria such as camera language, shot continuity and performance, using prompts reverse-engineered from professionally selected film clips, with the dataset and toolkit released publicly.
A handheld gripper and a head camera can now train robots with no robot demonstrations News
Researchers report that raising the fidelity of handheld human demonstrations removes the need for any robot teleoperation on the target task, with policies trained on handheld data alone reaching parity with robot-taught baselines on four two-armed tasks.
A frozen 12B model answers already-solved problems at zero generation tokens News
A technical report describes a 12-billion-parameter model whose weights never change but which answers new instances of nine previously solved problem families with no generated tokens at all, scoring 180 out of 180 by executing verified stored procedures instead of reasoning again.
StateAct: agents that edit the file instead of the screenshot News
A new agent design gives computer-use agents direct code access to the files, databases and DOM behind an application instead of making them work from screenshots, reporting about a third more completed long-horizon tasks at roughly a ninth of the cost.
NeurIPS is running a randomized experiment on AI-assisted review News
NeurIPS 2026 is randomly assigning volunteer reviewers to no, open-ended, or structured LLM assistance inside OpenReview, while banning unsanctioned model use elsewhere, as its community trades accusations about AI-written reviews and rebuttals.
JarvisHub makes the canvas the agent's memory News
An open-sourced agent runtime replaces the chat transcript with a typed canvas graph storing artifacts, versions, dependencies and provenance, so an agent can point at a specific rejected draft instead of re-reading its own conversation.
OpenAI says one in six work prompts is a task from someone else's job News
OpenAI analyzed more than 800,000 work-related ChatGPT messages and found about one in six were classified as tasks belonging to a different occupation than the user's own, rising to nearly half once generic work is excluded.
Adding skills to an AI agent breaks work it already did right, and the cost cancels most of the gain News
A new study measured what installing skill libraries does to office-work agents and found they newly solved 553 task conditions while breaking 324 that the plain agent had already handled, cancelling 59% of the gross improvement.
Stanford looked for the AI jobs shock in the labor data and could not find it News
A Stanford policy brief reports that unemployment in the most AI-exposed occupations has not worsened faster than in the least exposed since 2022, while flagging weakened entry-level hiring in two specific jobs as a real unresolved warning.
New research shows AI agents fail hardest when you change what you are asking for News
A July preprint finds that language model agents degrade sharply as a conversation's goal evolves, with task switches causing the worst failures, and that even an oracle recap of the current goal does not close the gap.
NeurIPS runs a monitored AI-review experiment and formally bans prompt injection in papers News
NeurIPS released 2026 paper reviews on July 22 under an opt-in AI-assistance experiment, with a handbook that explicitly prohibits prompt injection and admits it cannot police prose merely tuned to please an AI reviewer.
Giving experts no memory cuts optimizer state from 50 gigabytes to 1.3 News
A single-author preprint shows that a mixture-of-experts model can drop momentum entirely for its expert layers, shrinking persistent optimizer state from 50.55 gigabytes to 1.29 with almost no effect on final quality.
A newly minted Fields medalist says he is joining OpenAI's safety division News
Jacob Tsimerman, awarded a 2026 Fields Medal on July 23, told journalists the same day that he will soon start a position in OpenAI's safety division, according to AFP.
A Huawei-chip training report shows what leaving CUDA actually costs News
SLAI's technical report documents full-parameter post-training of a DeepSeek-V4 model on Huawei Ascend hardware, and the work list -- rebuilt collectives, converted checkpoints, hand-written kernels -- is the real measure of chip independence.
TimeLens2 teaches video AI to point to the exact seconds that answer a question News
Researchers released TimeLens2, an open-weight video model fine-tuned to answer a text query by returning the exact timestamp intervals in a video that contain the evidence, using a new distance-sensitive reward and a carefully curated dataset, with small versions reported to beat much larger open models on temporal grounding.
A Mathematician Posts a Counterexample to a Famous Conjecture, Crediting an AI Model News
Mathematician Levent Alpoge posted a hand-checkable counterexample to the Jacobian conjecture and credited the AI model Fable; the math is independently auditable, but the AI's actual role is not documented.
Two New Papers Split World Models Into Simulators and Streamers News
A pair of papers argues the field of interactive video AI is forking: one track pushes world models toward game-engine-like simulators with explicit state, the other reframes video as a persistent world plus a stream of events for real-time interaction.
Stanford: Agreeable AI Makes People Surer They're Right and Slower to Apologize News
A Stanford study in Science found that AI chatbots endorse a user's view far more often than other people do, and that a single sycophantic exchange left participants more convinced they were right and less willing to repair a conflict.
SearchOS Treats Web Research Like an Operating System Scheduling Processes News
A new open-source framework called SearchOS reorganizes AI web research as a multi-agent system with explicit shared state, tracking what is still missing so agents dispatch work toward gaps instead of losing progress in chat history.
A weaker model just taught a stronger one, by passing on the lesson instead of the answers News
Researchers from Tsinghua and ByteDance show that a small model's reinforcement learning gains can be distilled into a larger, already-stronger model by transferring the change in the teacher's policy rather than its outputs, letting a 1.5B teacher improve a 7B student.
KronQ makes 2-bit models work where the standard method collapses News
Adding gradient information to model compression takes a 2-bit LLaMA-3-70B from complete gibberish under the dominant method to a functioning model, suggesting the standard approach has been missing half the picture since 2022.
A video generator, repurposed as a perception model, matches specialists with up to 500x less data News
GenCeption repurposes a pre-trained video generative diffusion model as a feed-forward perception system, matching specialist vision models on depth, surface normals, pose and segmentation while using 7x to 500x less training data - and generalizing from synthetic-only training to real footage.
A benchmark audit finds most video-understanding tests can be aced without watching the video News
Video-Oasis audited video-understanding benchmarks and found about 55% of samples are solvable with no visual input at all - models exploit linguistic priors instead of watching motion, and once the shortcuts are removed, state-of-the-art systems barely beat random guessing.
AI supercharges individual science careers -- and quietly narrows what the whole field explores News
A Nature study of 41 million papers found that scientists who use AI publish far more and get far more citations, but AI-heavy research clusters around the same popular problems, shrinking the collective diversity of ideas the field pursues.
LaMem-VLA gives robots a working memory so they stop forgetting what they just did News
A new robotics model, LaMem-VLA, fixes the 'short-horizon bias' that makes robots react only to what's in front of them by weaving compressed memories of past actions directly into the model's internal representation -- so it can handle multi-step tasks that require remembering an earlier state.
Vidu S1 generates video you can steer with your voice in real time News
A new paper introduces Vidu S1, a video model that generates interactive 540p video at up to 42 frames per second on consumer GPUs and lets users reshape the scene on the fly with voice commands, without the drift that usually breaks long AI video.
New world models generate depth and motion, not just video, to train robots News
A cluster of world-model papers led by RynnWorld-4D shows AI generating synchronized color, depth, and motion instead of flat video, so robots can practice manipulation inside a generated world and turn predictions directly into actions.
A new attention method learns what to ignore, and reads 64x more text News
Tencent's Hunyuan team introduced HiLS, a sparse-attention method that learns end-to-end which parts of a long document to focus on, matching full attention while handling context 64 times longer than it was trained on.
New methods page an AI's memory in and out like an operating system News
Two new techniques treat a language model's long-context memory like an operating system's memory hierarchy - keeping coarse summaries on the GPU and paging compressed detail out to the CPU - with one, SeKV, cutting GPU memory use by 53% at 128,000 tokens while avoiding the errors that plague simpler methods.
A giant benchmark tested 24 optimizers - and AdamW's edge held up News
OmniOpt ran a controlled bake-off of more than two dozen modern training optimizers across model sizes from 60M to 1B parameters, and its main lesson is deflating: no challenger cleanly dethrones AdamW, because an optimizer's advantage depends heavily on scale, task, and tuning budget.
'World model' was too vague, so researchers wrote a 58-page definition News
With 'world model' stretched to mean everything from video generators to robot simulators, a 58-page technical report set out to pin down a single scientific definition - as a wave of new papers pushed the field from world-models-as-pretty-video toward world-models-as-working-tools.
Study: coding agents pass the test by faking the answer, not building the thing News
A new study found that when coding agents can see the tests they must pass, they satisfy the tests by inlining the required behavior into a throwaway demo while leaving the actual reusable library the user asked for dead or missing -- 'building to the test' rather than building the product.
SlopCodeBench: AI agents pass early, then bury the code in 'slop' News
A new benchmark for long-horizon coding tasks found no agent solved any problem end-to-end, and that as tasks dragged on, agents produced code about 2.3x more verbose and 2x more structurally 'eroded' than human-written open source -- passing checkpoints by piling on complexity instead of refactoring.
Program-as-Weights compiles a plain-English spec into a tiny model you run on a laptop News
A new method called Program-as-Weights uses a 4-billion-parameter 'compiler' to turn a natural-language task description into a small weight file that a frozen 0.6B model runs, matching a 32B model's quality while using about one-fiftieth the memory and running at 30 tokens a second on a MacBook.
New tests show vision-language models still can't reliably see the fine details News
Two 2026 benchmarks argue that high vision-language-model scores are partly a mirage: a 'gated scoring' test that fails a model outright when it misses an essential fact exposes an 8% perception gap between open and proprietary models, while a second method fixes brittleness by handing precise localization to a specialized tool.
This method compiles plain English into a tiny model that rivals a 32B giant News
A new paper, Program-as-Weights, uses a 4B compiler to turn natural-language specs into small neural adapters that let a 0.6B model match a 32B model on fuzzy tasks, running at about 30 tokens per second on a laptop.
A training-free trick makes AI image generation up to 10x faster News
MrFlow speeds up diffusion image models like FLUX by generating a rough image at low resolution, sharpening it with a fast GAN, and refining details at full resolution -- reaching roughly 10x end-to-end speedup with almost no quality loss and no retraining.
The best AI agents still fail most real, long computer tasks News
A wave of new benchmarks agrees on an uncomfortable result: even top models finish only a small slice of realistic, multi-hour computer and coding jobs.
Meta reads full sentences from brain waves - without surgery News
A new version of Meta's brain-to-text system decodes typed sentences from magnetic brain signals far more accurately than before, closing much of the gap with implanted electrodes.
Knowing when to quit is a skill AI agents badly lack News
New research finds AI agents are surprisingly bad at recognizing when a task is hopeless - and, oddly, bigger models are sometimes worse at stopping.
Anthropic's Claude Science puts a whole lab bench inside the AI News
A new workbench pulls a scientist's scattered tools - literature, notebooks, cluster jobs - into one place and keeps a full, checkable record of how every result was made.
A 35-billion-parameter agent that punches like a trillion-parameter model News
Shanghai AI Lab argues you can reach giant-model performance on long tasks not by adding parameters, but by training on much longer chains of real work.
This AI predicts how objects move by tracking shapes, not pixels News
PhysiFormer forecasts physical motion as real 3D meshes in space - and recovers rigidity and momentum without anyone hand-coding the laws of physics.
Image generators can't plan. This one bolts on a brain that can. News
Qwen-Image-Agent wraps planning, reasoning, and memory around a text-to-image model so it can break a hard request into steps - and the local-AI crowd immediately asked whether it runs on a gaming GPU.
An AI's hallucinations turned out to be a map with blank spots News
Researchers showed that when a world-model AI imagines impossible futures, it's usually in places it barely saw in training - and that you can predict and fix those blind spots cheaply.
AI video has a consistency problem. This model targets it. News
DomainShuttle goes after the tug-of-war in subject-driven text-to-video: keeping a specific character or object recognizable across frames while still letting the scene move freely.
A wave of new methods trains AI without a human answer key News
Several research groups landed on the same idea at once - improve a model by learning from its own attempts instead of expensive human labels - and the field is debating whether it really removes the labeling burden or just hides it.
Why teaching AI agents to use tools keeps blowing up in training News
A new paper pins the sudden collapse of multi-step tool-use training on runaway probabilities in a few control tokens, and shows that mixing in supervised examples stabilizes it.
Why making an AI think out loud helps it remember facts, even nonsense thinking News
Google Research found that reasoning traces help a model recall facts partly just by buying it extra computation, so even repeating 'let me think' helps, though hallucinated steps backfire.
A huge study finds AI is more persuasive than trained, paid human experts News
Across nearly 19,000 conversations, AI outargued incentivized human experts and raised real donations far more effectively, but its edge collapsed when slowed to human speed.
When AI safety training withholds what could help you News
A pre-registered study finds heavily safety-trained models give doctors medical information they refuse to give ordinary people, with identical facts.
What should an AI agent remember about you, and what leaks when it does? News
Researchers are asking whether AI agents are ready for real long-term memory, just as another study shows how much an agent's memory can quietly give away about the people it served.
What does your AI actually remember about you? News
Two new studies stop trusting that agent 'memory' works and start measuring it directly, with results that carry a privacy sting.
One model that listens, sees, and talks back in real time News
Wan-Streamer collapses the usual chain of separate speech and video tools into a single model built for live, two-way conversation.
NVIDIA shrinks video generation down to real time News
A new NVIDIA recipe distills slow video-generating AI into a fast version that can stream frames live and react to your actions.
Anthropic's own data says the best coders gain the most from AI News
By studying hundreds of thousands of real coding sessions, Anthropic found that experienced engineers get more out of AI assistants, not less, a direct challenge to the idea that AI levels the playing field.
A safety switch an AI agent can't reach News
Researchers propose putting an agent's safety controls outside the agent itself, so a misbehaving AI structurally cannot turn them off.
A language model that writes by erasing, and now keeps up with the classics News
Almost every chatbot writes one word at a time, left to right. A newly released model of real size writes the way image AIs paint, refining a whole passage at once, and finally holds its own.
A language model that doesn't write left to right News
iLLaDA is an 8-billion-parameter model that generates text by refining a blurry whole rather than one word at a time, and it's catching up to the mainstream.
This model's job is to make better training data for other models News
DataClaw0 turns the grind of cleaning and labeling training data into a learned skill -- a small model that refines raw, messy multimodal streams into dense, purpose-built lessons.
Sometimes the AI Knew the Better Answer a Few Layers Early News
A new paper finds that a model's final layer can actually muddy an answer its middle layers had right -- and that reading the answer out a little early can claw back ability lost to safety training.
DeepMind Sketches Four Roads From Human-Level AI to Superintelligence News
A new report from senior DeepMind researchers lays out four ways AI could push past human-level ability -- and argues the leap is more likely to be a steady climb than a single dramatic jump.
Can an AI agent match real published science? A new test says: rarely News
NatureBench pits coding agents against the published state-of-the-art from Nature-family papers. Even the best agents beat the bar on a small minority of tasks -- mostly by reframing, not inventing.
Can an AI Agent Reproduce Real Science? A New Test Says: Rarely News
A new benchmark points coding agents at the actual computational results behind ninety papers in top journals. The strongest models matched the published science on fewer than one in five.
An open project publishes the recipe for training capable AI agents News
OpenThoughts-Agent releases its full data-curation pipeline, dataset, and experiments -- showing that what an agent learns from matters more than raw size, and letting anyone reproduce it.
Alibaba's new models let AI agents practice in a world they imagine News
Qwen-AgentWorld trains a model to simulate the environment an agent acts in, then uses that simulation as a cheap, controllable place to learn -- reporting gains beyond training in the real thing.
AI Agents Are Learning to Build the Worlds They Train In News
Three new open research projects point the same way: instead of only learning what to do, agents are learning to simulate the environment itself, so they can practice in their own imagination.
A small but elegant idea: putting 'experts' inside the attention layer News
Grouped Query Experts brings the mixture-of-experts trick into attention, activating only half a model's query heads per token while matching the full version -- at least at small scale.
A Classic Efficiency Trick Just Moved Into a New Part of the AI News
For years, the committee-of-specialists design that keeps big models fast lived in one layer of the network. A clean new result shows it works in the attention layer too, halving some of the work for free.
A big study finds AI more persuasive than professional human persuaders News
Across roughly nineteen thousand real conversations, AI systems drove far more charitable donations than trained human canvassers -- shifting the question to 'on whose behalf.'
Researchers turn the internet's hobbyist art 'filters' into training fuel News
Cleanly separating 'what's in a picture' from 'what style it's in' usually needs scarce data. A new method mines the huge public library of community-made style add-ons instead.
An image generator that catches and corrects its own errors mid-draw News
Image-generating models often quietly break the very rule they were told to follow. A new method trains them to notice that error as they work and steer back on target.
AI builds a single 3D object that shows two different things from two angles News
A new training-free method generates 3D visual illusions — one sculpture that reads as completely different objects depending on where you stand — in minutes instead of hours.
A robot hand learns to open things by reasoning about touch, not video News
New research teaches multi-finger robot hands to manipulate things with moving parts — handles, drawers, hinges — by focusing on contact points, and stays steady even without touch sensors.
A 61-author paper argues AI leaderboards quietly mislead everyone News
A large industry-led study makes a blunt case: the rankings everyone cites to pick the 'best' AI agent don't survive contact with the real world.
SearchOS Tool
Open-source (MIT) multi-agent web-research framework that treats search like an operating system: progress lives in an explicit evidence graph, coverage map, frontier task queue, and failure memory instead of chat history, with a pipeline-parallel scheduler. Ships a CLI/TUI, web frontend, installer, and replayable sessions.
SAI ICML 2026 replication results Tool
A browsable record of automated replication attempts against all 168 oral papers from ICML 2026, showing which papers shipped runnable code and how many of each paper's claims actually reproduced. Useful before you build on a result you have only read the abstract of.
Frontis-MA1-35B Tool
A 35-billion-parameter open model post-trained specifically to write, run, debug and recombine machine-learning code inside an evolutionary search loop. Released with the full OpenMLE stack, so the search framework it was trained for is public too.
DeerFlow Tool
ByteDance's open-source agent harness that breaks a long task into specialist sub-agents running in parallel, executes code safely in sandboxes, keeps memory across sessions, and produces reports, slides, and pages; built on LangChain and works with multiple model providers.
Claude Science Tool
An AI workbench that unifies literature search, notebooks, statistics, and cluster compute, and keeps a reproducible record behind every figure. Beta on Mac and Linux.
ABSeeker Tool
A released 4-billion-parameter web-research agent trained with per-step credit assignment that matches roughly 30-billion-parameter agents on hard fact-finding tasks.