AI Papers of the Week
Every paper worth reading in AI, hand-picked one week at a time.
OLMo 3
Allen Institute for AI introduces OLMo 3, a fully open language model family that releases the complete “model flow”: every training stage, checkpoint, dataset, and dependency, enabling researchers to intervene at any development point. The release includes four specialized variants (Base, Think, Instruct, RL Zero) at 7B and 32B scales.
MAKER: Solving Million-Step LLM Tasks
MAKER is the first system to successfully solve tasks requiring over one million LLM steps with zero errors, overcoming a fundamental limitation where LLMs typically fail after a few hundred steps in complex multi-step processes. The approach demonstrates that massively decomposed agentic processes can efficiently handle lengthy sequences of dependent logical operations through extreme decomposition and error correction.
On the Fundamental Limits of LLMs at Scale
This work establishes rigorous mathematical foundations for theoretical limitations constraining LLMs, identifying five fundamental constraints: hallucination (rooted in computability theory), context compression, reasoning degradation, retrieval fragility, and multimodal misalignment. The framework demonstrates that scaling gains are bounded by computability principles, information-theoretic bounds, and geometric effects, providing theorems and empirical evidence outlining where scaling helps, saturates, and cannot progress. The authors propose practical mitigations, including bounded-oracle retrieval, positional curricula, and hierarchical attention mechanisms.
Olympiad-Level Formal Mathematical Reasoning with Reinforcement Learning
Google DeepMind introduces AlphaProof, an AlphaZero-inspired reinforcement learning agent that learns to find formal mathematical proofs within the Lean theorem prover, achieving the first-ever medal-level performance at the International Mathematical Olympiad by solving three problems, including the competition’s most difficult challenge.
The Era of Agentic Organization
Microsoft Research introduces asynchronous thinking (AsyncThink), a new reasoning paradigm where language models learn to organize their internal thinking into concurrently executable structures through an organizer-worker protocol, achieving 28% lower inference latency than parallel thinking while improving accuracy on mathematical reasoning and demonstrating zero-shot generalization to unseen tasks.
Unified Bayesian Account of LLM Control
Researchers from Stanford and MIT present a unified Bayesian framework explaining how prompting (in-context learning) and activation steering both control LLM behavior by altering beliefs in latent concepts, with steering modifying concept priors while ICL accumulates evidence.
RL Enhances Knowledge Navigation
Researchers show that RL-enhanced models outperform base models by 24pp on hierarchical knowledge retrieval tasks (e.g., medical codes) by improving navigation of existing knowledge structures rather than acquiring new facts. Structured prompting reduces this gap to 7pp, while layer-wise analysis reveals that RL transforms query processing (cosine similarity drops to 0.65-0.73) while preserving factual representations (0.85-0.92). The findings suggest RL’s benefits stem from enhanced procedural skills in traversing parametric knowledge hierarchies rather than expanded knowledge content.
Towards Robust Mathematical Reasoning
Google DeepMind introduces IMO-Bench, a comprehensive suite of benchmarks vetted by IMO medalists targeting International Mathematical Olympiad-level reasoning, featuring 400 diverse Olympiad problems with verifiable answers, 60 proof-writing problems with detailed grading schemes, and 1000 human-graded proofs, playing a crucial role in achieving historic gold-level performance at IMO 2025.
Context Engineering 2.0
Researchers from SJTU, SII, and GAIR trace the 20+ year evolution of context engineering, reframing it as a fundamental challenge in human-machine communication spanning from primitive computing (Era 1.0) to today’s intelligent agents (Era 2.0) and beyond. It defines context engineering as systematic entropy reduction where humans preprocess high-entropy contexts into low-entropy machine-understandable representations. This gap narrows as machine intelligence increases.
Scaling Agent RL via Experience Synthesis
Meta researchers introduce DreamGym, a unified framework that synthesizes diverse training experiences to enable scalable reinforcement learning for LLM agents without costly real-environment rollouts. It addresses fundamental barriers of expensive interactions, limited task diversity, unreliable rewards, and infrastructure complexity.
TIR-Judge
Google and collaborators introduce TIR-Judge, an end-to-end reinforcement learning framework that trains LLM judges to integrate code execution for precise evaluation. It surpasses reasoning-only judges by up to 6.4% (pointwise) and 7.7% (pairwise) while demonstrating that tool-augmented judges can self-evolve without distillation.
Enhancing Long-Term Memory in LLMs
Researchers from the University of Alberta and UMass Amherst introduce BEAM, a new benchmark for evaluating long-term memory in LLMs with conversations up to 10M tokens, and LIGHT, a framework that enhances memory performance through three complementary systems.
Mathematical Exploration and Discovery at Scale
Google DeepMind, Princeton, Brown, and Terence Tao apply AlphaEvolve, an AI system using LLM-guided evolutionary search to autonomously discover mathematical constructions across analysis, combinatorics, geometry, and number theory. Across 67 problems, AlphaEvolve rediscovered best-known solutions, improved several problems, and extended finite solutions into general formulas with significantly reduced computation time.
Unlocking the Power of Multi-Agent LLM for Reasoning
Researchers from Penn State, Harvard, Microsoft, and collaborators introduce Dr. MAMR, addressing the “lazy agent” problem in multi-agent LLM reasoning through Shapley-inspired causal influence measurement and verifiable restart mechanisms. The framework achieves 78.6% on MATH500 (+4.2% over ReMA), 20.0% on AIME24, and maintains balanced agent contributions where baseline approaches collapse into single-agent dominance.
AgentFold
AgentFold introduces proactive context management for long-horizon web agents, addressing context saturation through dynamic “folding” operations that balance detail preservation with efficient compression. The 30B parameter model outperforms dramatically larger competitors while achieving state-of-the-art results on web browsing benchmarks.
Introspective Awareness
Anthropic research demonstrates that contemporary LLMs possess limited but functional introspective capabilities, the ability to recognize and accurately report on their own internal states. Using activation steering to inject known concepts into model activations, the study measures whether models can detect these manipulations through self-report, revealing that introspection remains highly unreliable and context-dependent.
Multi-Agent Evolve
Multi-Agent Evolve (MAE) enables LLMs to self-improve their reasoning capabilities without human-annotated data through a co-evolving multi-agent framework. Three interacting agents (Proposer, Solver, Judge) instantiated from a single LLM undergo reinforcement learning optimization together, creating a scalable self-improving system that extends beyond game-based environments to general reasoning domains.
SmolLM2
SmolLM2 demonstrates that strategic data curation beats scale through a 1.7B parameter model trained on 11 trillion tokens using iterative data mixing optimization. The data-centric approach introduces three specialized datasets (FineMath, Stack-Edu, SmolTalk) and dynamically refines composition across training stages, achieving superior performance over Qwen2.5-1.5B and Llama3.2-1B while enabling practical on-device deployment.
Global PIQA
Global PIQA extends physical commonsense reasoning evaluation to 100+ languages and cultural contexts, revealing how language models handle everyday practical scenarios across diverse linguistic communities. The benchmark goes beyond translation to include culturally-contextualized scenarios, uncovering significant performance variations that challenge assumptions about universal physical understanding in AI systems.
GAP
GAP introduces graph-based agent planning with parallel tool execution and reinforcement learning, enabling AI agents to coordinate multiple specialized capabilities simultaneously rather than sequentially. The framework significantly accelerates task completion and improves success rates on complex multi-step problems through optimized tool selection and execution ordering.
When Models Manipulate Manifolds
Anthropic researchers investigate how Claude 3.5 Haiku learns to predict line breaks in fixed-width text, revealing geometric representations analogous to biological place cells and boundary cells in biological brains.
Reasoning with Sampling
Base language models achieve reasoning performance matching or exceeding RL-posttraining through inference-time power distribution sampling, using MCMC techniques that require no training, datasets, or verifiers.
Ring-1T
Ring-1T is the first open-source thinking model with 1 trillion parameters (~50B active per token), achieving breakthrough results through three innovations for trillion-scale RL training.
Cell2Sentence-Scale 27B
C2S-Scale extends Cell2Sentence by converting gene expression into “cell sentences” and training LLMs on 50M+ cells plus biological text. Models scale to 27B params and unify prediction, generation, and NL interpretation. A dual-context virtual screen then led to a wet-lab validated finding: silmitasertib acts as an interferon-conditional amplifier of MHC-I antigen presentation.