AI Papers of the Week
Every paper worth reading in AI, hand-picked one week at a time.
Can LLM Agents Infer World Models?
Can an LLM agent actually build a model of an environment it cannot see? This work makes that question gradeable through agentic automata learning. An agent has to uncover a hidden deterministic finite automaton by interacting with an oracle through two interfaces, membership queries that ask whether a string belongs to the target language, and equivalence queries that ask whether a proposed automaton is correct, which yields a clean, scalable testbed for interactive discovery.
OpenClaw-Skill
Equipping LLM agents with effective skills is most of the battle in real systems, yet most skill-induction work distills one trajectory at a time, which produces narrow, brittle skills. OpenClaw-Skill introduces Collective Skill Tree Search, a tree-search-based skill construction framework that builds a structured, diverse, and generalizable tree of skills, then trains agents to actually use what it builds.
Back on Track
Diffusion large language models generate text in a way that does not fit cleanly into the reinforcement learning recipes built for autoregressive models, and training them to reason exposes two specific problems. Rewards are sparse, so a single terminal reward fails to guide intermediate generation steps, and policy updates sometimes drift toward unnatural trajectories rather than authentic generation paths. This paper proposes Process Aligned Policy Optimization to fix both.
The Stanford EDGAR Filings Dataset
Clean, long-context documents remain scarce for pretraining, especially in finance. This release reconstructs U.S. SEC corporate and financial disclosures into layout-faithful, token-efficient MultiMarkdown, publishing 152B tokens in SEFD-v1 out of an estimated 550B-token archive spanning 18.5M filings, with less than 0.1% overlap with Common Crawl corpora. It also ships two derived benchmarks, EDGAR-Forecast for numerical forecasting and EDGAR-OCR for financial table transcription, to support financial reasoning, forecasting, and document understanding.
MiniMax Sparse Attention
Ultra-long context is now a core requirement for agents, codebase-scale reasoning, multimodal workflows, and persistent memory, but dense softmax attention still makes million-token deployment expensive. MiniMax Sparse Attention (MSA) tackles this by adding blockwise sparsity on top of Grouped Query Attention, with a lightweight routing branch that chooses which key-value blocks each query group should actually attend to.
Lookahead Sparse Attention
Long-context decoding is bottlenecked by the KV cache, which grows with every token and quickly dominates memory at extreme context lengths. This work, branded around DeepSeek-V4, introduces Lookahead Sparse Attention (LSA), which avoids storing the full KV cache by predicting which parts of the context future decoding will actually need and retaining only those query-critical chunks.
The Consistency Illusion
Multi-agent debate is often judged by whether the agents end up agreeing, but this paper shows that output-level consensus can hide deep disagreement in the reasoning that produced it. The authors abstract agents' reasoning traces and decisions into four states along two axes, reasoning similarity and conclusion agreement, and flag divergent agreement, where agents reach the same answer through very different paths. Across 600 content-moderation items, divergent agreement appeared in 118 cases and separated cleanly from genuine disagreement states with a Cohen's d of 0.80, and routing on these categories beat divergence-only methods at flagging high-disagreement cases.
Beyond Scalar Rewards
Reward models usually compress a judgment into a single scalar, but this paper argues human preferences are better captured as score distributions, and proposes Z-Reward, which internalizes reasoning into a predicted distribution before scoring. A large vision-language teacher does the reasoning-heavy judgment and is distilled into a compact student for efficient deployment, with the 27B teacher reaching 89.6% human-preference accuracy and the 9B student nearly matching it at 88.6%. Used as a reinforcement learning signal, it delivers a 41.3% net preference improvement over a supervised baseline, beating GRPO and other reward methods.
Self-Revising Discovery Systems
From MIT, this paper argues that genuine scientific discovery is not answer generation but a change in the search space itself, and that an AI scientist must perceive that shift without being told. It develops a category-theoretic framework in which evidence, artifacts, operations, and verifiers are typed, and discovery is defined as a principled revision of that representational regime rather than more search within a fixed one.
LEAP
New research from Google shows how far a custom agent harness can push a general-purpose model on formal mathematics. LEAP wraps a general LLM in an agentic scaffold that grounds every step in the Lean compiler and iterates against verifier feedback. Rather than fine-tuning a specialized prover, it leans on informal reasoning, instruction following, and self-refinement, then forces every formal step through a compiler check before moving on.
A Primer on Post-Training Reasoning Data
This primer is the first to pull the scattered post-training reasoning-data literature into one place, synthesizing over 150 public studies and system reports that previously lived across dataset papers, RL write-ups, and lab reports. It organizes the field around four questions: what data objects exist, what makes them useful, how they are constructed, and how they scale. The key reframing is that a reasoning-data item is more than a prompt-response pair: it packages a problem or state, model behavior, judging feedback, and attribution metadata, with usefulness defined relative to the verifier and the rest of the corpus rather than in isolation.
Language Models Need Sleep
Attention scales badly with context length, so long-horizon agents keep paying a growing cost as their context grows. This paper studies a sleep-like consolidation mechanism: the model periodically converts recent context into persistent fast weights, then clears its key-value cache. During the sleep phase it performs offline recurrent passes over the accumulated context and updates fast weights in its state-space blocks through a learned local rule.
Adapting the Interface, Not the Model
When a frozen LLM agent repeatedly fails in a deterministic, rule-governed environment, do you have to retrain the model? Life-Harness argues no. Many failures come from mismatches at the model-environment interface, not from the model's reasoning, so the fix belongs in the runtime harness. Life-Harness is a lifecycle-aware harness that improves frozen agents without touching model weights or the evaluation environment.
Forecasting Scientific Progress with AI
Can frontier models predict where science is going? This work introduces CUSP, a cutoff-conditioned benchmark built from 4,760 real scientific events across multiple disciplines, each grounded against a verified knowledge cutoff. For every event, models are tested on four tasks: feasibility assessment, mechanistic reasoning, generative solution design, and temporal prediction. The headline is sobering: models recognize plausible directions but cannot forecast outcomes.
Code as Agent Harness
A 100+ page survey treating the agent harness as a first-class research object rather than glue around an LLM. The authors argue that code-as-harness is the most promising path to general-purpose agency, and that future agent systems should satisfy four properties: executable, inspectable, stateful, and governed. The report consolidates methods, applications, and open problems across the harness layer.
OpenAI Disproves the Unit Distance Conjecture
An OpenAI internal reasoning model produced a counterexample to Erdős's 1946 unit distance conjecture, the first time an AI system has autonomously resolved a prominent open problem in mathematics. For nearly 80 years, mathematicians believed square grids were essentially optimal for placing n points to maximize unit-distance pairs. The new construction beats grids using an infinite unramified tower of totally real number fields with 3-power Galois groups, producing n-point sets with more than n^(1.014) unit distances. A human-verified companion paper was prepared by nine external mathematicians including Noga Alon, Tim Gowers, and Melanie Matchett Wood.
A Geometric Calculator Inside a Neural Network
Goodfire reports mechanistic interpretability work identifying a geometric calculator inside an LLM. The model represents numbers as Fourier features, where circles in activation space correspond to numbers modulo a given base. Arithmetic operations are implemented as rotations of these circles, forming a variant of a residue number system that does not require coprime moduli. The same circuit appears to be reused beyond arithmetic.
AutoTTS
AutoTTS reframes test-time scaling as a search problem. Instead of designing branching, pruning, and stopping heuristics directly, the user constructs a discovery environment in which TTS strategies are searched automatically. Width-depth TTS is recast as controller synthesis over pre-collected reasoning trajectories and probe signals, so candidate controllers can be evaluated without repeated LLM calls.
AI Co-Mathematician
Google DeepMind presents AI Co-Mathematician, an agentic research workbench for mathematicians. The system is an asynchronous, stateful environment that supports ideation, literature discovery, computational analysis, theorem verification, and knowledge development across long sessions. It reaches 48% on FrontierMath Tier 4, a new high among AI systems evaluated.
AEvo
AEvo separates the iterative self-improvement loop into two roles: a candidate-proposer that generates the next attempt, and a meta-agent that observes traces and edits the procedure used to propose future candidates. Past runs (candidates, feedback, traces, failures) function as memory the meta-agent reads from when revising the procedure. AEvo reports a 26% relative gain over the strongest evolution baseline on agentic and reasoning benchmarks, and SOTA on three open-ended optimization tasks under the same iteration budget. The work demonstrates one way to operationalize accumulated agentic search logs as input to procedure-level updates rather than discarding them after each run.
The Memory Curse in LLM Agents
A study of how long histories affect LLM agent behavior. Across 7 LLMs and 4 social dilemma games over 500 rounds, expanding accessible history degraded cooperation in 18 of 28 model-game combinations. Lexical analysis of 378,000 reasoning traces shows the mechanism is erosion of forward-looking intent rather than increased suspicion: long histories pull the model toward reasoning about past interactions rather than future payoffs. A LoRA adapter trained only on forward-looking traces mitigates the decay and transfers zero-shot to new games. Memory sanitization, which keeps prompt length fixed but swaps in synthetic cooperative records, restores cooperation, indicating the trigger is content rather than length. Ablating explicit chain-of-thought often reduces the collapse, suggesting deliberation amplifies the effect. The paper provides a diagnostic plus interventions for long-running agent systems where history quality, not just history length, drives behavior.
HeavySkill
One of the cleaner takes on agentic harness design released this year. The paper argues that what actually drives harness performance is not the orchestration code, but a single inner skill: parallel reasoning followed by deliberation. Internalize that pattern into the model and most of the surrounding scaffolding becomes optional. HeavySkill systematizes the idea as a two-stage pipeline you can run beneath any harness, then trains it as a learnable skill via RLVR. The result is a harness win that looks more like a model win.
Self-Improving Pretraining
Most LLM safety, factuality, and reasoning fixes get bolted on at post-training. By then the patterns have already set. This Meta FAIR paper moves those behaviors into pretraining itself. The team uses a strong post-trained model as both a rewriter and a judge: it rewrites pretraining suffixes toward higher-quality, safer continuations, then scores model rollouts against the original suffix and the rewrite to drive RL during pretraining. Instead of next-token prediction, the policy learns sequence generation from the start, with rewards for quality, safety, and factuality.
Horizon Generalization
Microsoft Research runs a controlled study where the only variable is task horizon length. Same decision rules, same reasoning structure, different sequence length to the goal. The main finding: horizon alone is a training bottleneck. As goal distance grows, exploration explodes combinatorially and credit assignment gets ambiguous. Models that learn cleanly on short horizons fall apart on long ones, even when the underlying reasoning is identical. The fix is not more compute, it is horizon reduction.