#Paper/Research

AIMIT Statistics and Data Science Center director Sasha Rakhlin argues that AI progress is fastest where results can be verified quickly, citing a model reaching gold-medal level at the International Mathematical Olympiad a year before models produced new research results. He says departments should reconsider how they reward work, emphasizing question-asking, replication, and disclosure of AI's role in a researcher's contributions. He also urges universities to build shared lab infrastructure that captures failed experiments and tacit expertise.

AILewis Tunstall quotes Kyle Cranmer's post about a paper by Nate Gunnarsson on a non-perturbative approach to chiral fermions in the Standard Model, extending Lüscher's abelian result. The paper's acknowledgments state that OpenAI's GPT-6 Astra model was essential, proposing refinement strategies, writing rewrites of the proof, and carrying out Lean verification.

AISantiago says physical consistency is the most important and hardest feature of a world model, and that many generated videos show objects defying gravity. He reports that Seedance 2.5 is currently the best among the evaluated world models. The post links to a physics evaluation benchmark in which eight video world models reached a top score of 57.76/100.

AIAmazon Scholar and University of Texas at Austin professor Matthew Lease will give an Expo Talk at COLM on Thursday at 1pm PT. The talk explores how to harness AI for scientific discovery while assessing potential risks, drawing on work from UT's Good Systems and the Cosmic AI Institute.

AIOpenAI released 722 mathematical manuscripts in 372 families, produced by an unreleased frontier model, with the average result taking the equivalent of three hours of ChatGPT Pro thinking. The author notes many results are verified in Lean but not all, and suggests mathematics could divide into vast machine-verified work and a compressed human 'effective theory' that people can actually understand.

AILucas Beyer, who says he is not a mathematician, finds it bizarre that an arXiv paper credits a "meatproxy" and lists an author "in personal capacity." He comments on the awkward form of the attribution while congratulating the authors on the content, which is linked in a reply from @mahdi_tcs_.

AIDust is a zeroth-order method that pretrains transformers and sometimes matches or exceeds backprop given large compute. The authors report it is about 1,000 to 10,000x more compute efficient than EGGROLL, the state-of-the-art ES method, for training transformers. The post also cites the gradient-alignment result up to 1B tokens and the virtual population idea for scaling.

AIA Latent.Space podcast episode features MIT researcher Alex Zhang explaining recursive language models (RLMs). He discusses why Claude Code, Codex, and Pi are basically the same, and how RLMs use code, context offloading, and recursive subagents to generalize across tasks. The episode also covers OpenAI's 10,000-agent, 130B-output-token experiment and academia's freedom to pursue ambitious research bets.

AIarXiv has introduced stricter rate limiting for all submitters to fairly distribute moderator time. The post links this move to Vibe research, where turning ideas into papers is easier, while noting that standards for judging research output have not kept pace.

AIHarvard physicist Matthew Schwartz argues that LLMs are poorly matched to science when used as human-style collaborators, so he built a toolkit for exact quantitative calculations. Working with Claude, the approach surfaced connections to ecology, population genetics, and a dozen other fields, with domain experts steering it toward interesting questions.

AIJohn Schulman says two papers on counterfactual simulatability used Tinker for fine-tuning experiments, one of them building on earlier work by Hase et al. Tinker's background post frames the approach as predicting how an LLM's output changes under a different prompt, and cites both papers.

AILewis Tunstall praises a new Meta paper on research preference models (RPMs), which instill "research taste" in agents by treating experiments as tree nodes. An RPM acts as an LLM judge that selects the most promising candidate experiment before it is run, reducing wasted compute. Tunstall notes the resulting trajectories could train domain-specific RPMs, which would be valuable in hard fields such as the natural sciences.

AIThe article argues that scientific papers omit the failed experiments and rejected branches that could train AI systems to develop scientific judgment. It describes Alasdair Russell's Cambridge group logging discovery paths as graphs of ideas, and proposes recording six fields per decision, including candidates and outcomes, to test whether this taste transfers to unfamiliar projects.

AIJohn Schulman says PPO gained a second wave in the LLM era for reasons not anticipated in the original paper. He points to the importance-ratio objective, which corrects biases from numeric error, asynchronous training, and forward-pass noise, and to the clipping objective, whose effect on entropy was unknown at publication, citing DAPO's arXiv paper.

AIJan Leike says he is excited about NLAs as a new tool in Anthropic's interpretability toolkit. The quoted post from Sam Marks describes NLAs as an unsupervised method that converts an LLM's internal state into human-readable text, which he says can advance understanding of model thinking and safety auditing.

AITri Dao says the dynamical system view yields very clean conditions for looped transformers to be stable. The post's background links looping blocks of layers during training to predictable scaling laws that can match a Transformer twice the size at a fixed parameter count.

AIMost alignment research is not crisp and requires research taste to evaluate, according to Jan Leike. He explains that Anthropic chose to point the AAR at a scalable oversight problem because progress there could let AARs tackle fuzzier alignment problems where humans can only provide weak supervision.