Stanford HAI

LLM agents can learn to proactively build optimized, hierarchical file system workspaces that anticipate future queries, outperforming reactive retrieval by 33 points on BrowseComp-Plus.

Unexplained observations, not just supporting evidence, are the key to LLMs discovering truly novel hypotheses.

Post-training LLMs for agentic tasks can drastically reduce solution coverage, a "Sharpening Tax" that a novel Bayesian sampler can mitigate.

Discarding irrelevant reasoning tokens can make language models up to 3x faster at test time without sacrificing performance.

Language models struggle to consistently encode the current year, with associative and declarative representations diverging in their update responses.

HyMeS enables robots to efficiently manage long-horizon interactions, achieving a 14.5-point increase in task success without requiring extensive demonstrations for every task configuration.

Test-time self-correction can boost LLM accuracy by over 30% on challenging reasoning tasks without the need for external reward models.

Skill-switching accuracy in LLMs drops significantly on complex tasks, but a new training approach boosts performance from 34.4% to 68.4% on challenging benchmarks.

Proprietary MLLMs may achieve high diagnostic accuracy, but they still struggle with reliable clinical reasoning, revealing significant gaps in their practical utility.

Fact-graph orchestration enables mathematical reasoning agents to tackle complex proofs more effectively than ever before.

Faithful reasoning in VLA models can boost policy responsiveness to rare scenarios by 1.6x compared to state-of-the-art approaches, revealing a critical gap in current alignment strategies.

Active-GRPO not only outperforms existing methods in molecular optimization but also redefines how models can adaptively balance imitation and self-discovery during training.

Learned stopping can significantly boost reasoning model performance in complex tasks, but its effectiveness hinges on the problem's characteristics rather than being a one-size-fits-all solution.

EQMs reveal that explanation quality can be quantitatively assessed, offering a more reliable indicator of forecasting accuracy than traditional methods.

Models that write intermediate states significantly outperform those that only report final answers, achieving up to 91% accuracy in predicting outcomes from edited states.

SPIRAL achieves a remarkable 15% performance increase by combining sequential, parallel, and aggregative reasoning in language models.

Despite high benchmark performance, LLMs often misrepresent logical reasoning, revealing a troubling gap between accuracy and faithfulness in legal contexts.

ExpRL outperforms traditional reinforcement learning methods by effectively rewarding intermediate reasoning steps, leading to better LLM performance on complex tasks.

INFUSER outperforms a frozen 32B model with just an 8B co-evolving generator, showcasing the power of adaptive question generation in self-evolution.

AI systems currently miss critical temporal and interpretive elements of clinical reasoning, limiting their effectiveness in real-world healthcare settings.

Current clinical AI systems often neglect the temporal dimension of patient care, limiting their effectiveness in longitudinal reasoning.

Design choices in agent memory systems can significantly shift operational costs, revealing critical trade-offs that impact long-horizon task performance.

CORE enables language models to rapidly enhance reasoning capabilities using minimal training data and rollouts, outperforming traditional methods.

AI-driven scientific discovery is closer than you think, but current systems still struggle with reproducibility, cross-domain robustness, and accountable scientific closure.

Looping language models isn't just for single agents anymore: Recursive Multi-Agent Systems (RecursiveMAS) show that agent collaboration itself can be scaled through recursion, yielding faster and more efficient problem-solving.

Language models can now learn to forget strategically, achieving 2-3x memory efficiency without sacrificing reasoning accuracy.

A lightweight, RL-trained context curator can match GPT-4o's context management abilities, slashing token consumption by 8x and opening the door to efficient long-horizon LLM agents.

LLMs can decide when they need more "thinking time" – and boost their accuracy while slashing compute costs by up to 65% – simply by checking if they agree with themselves.

LLMs' chain-of-thought reasoning often falls apart due to factual incompleteness, with errors compounding across multiple hops, as revealed by a new multi-hop QA dataset.

An AI agent can triage remote patient monitoring data with higher sensitivity than individual clinicians, suggesting a path to scalable and cost-effective patient monitoring.

LLMs struggle to explore multiple valid reasoning paths, often committing to a single route and missing alternative solutions, especially in complex, multi-step logical problems.

LLMs may grasp the broad strokes of causal strategies, but struggle with the devilish details of research design, as revealed by a new benchmark separating causal identification from estimation.