Research
A paper is listed when one of its authors is affiliated with the lab. Data: OpenAlex.
Predictive Closure: A Set-Level Characterization of JEPA's Feature SelectionAllen Institute for AI · arXiv (Cornell University)Oct 19When Should AI Chime In? Benchmarking Timely Contribution of Proactive Agents in Multi-Party ChatAllen Institute for AI · arXiv (Cornell University)Oct 19UnexpectedParrots: Iterating Behavioral Experiment Design with In-Silico SimulationsAllen Institute for AI · arXiv (Cornell University)Oct 18Reasoning Models Can Control Their Chains of Thought When They Tell Themselves ToAllen Institute for AI · arXiv (Cornell University)Oct 16The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free AdaptationAllen Institute for AI · arXiv (Cornell University)Oct 14EvalBuilder: Can LLMs Build Efficient Benchmarks?Allen Institute for AI · arXiv (Cornell University)Oct 14Pushback in a Haystack: Language Models Respond to Factual Errors Ambiguously and IneffectivelyAllen Institute for AI · arXiv (Cornell University)Oct 13Scaling Repeated-Data Pretraining: Batch Size, Regularization, and GPU-Time TradeoffsAllen Institute for AI · arXiv (Cornell University)Oct 11Reflex: Visually Grounded Reactive Humanoid ControlAllen Institute for AI · arXiv (Cornell University)Oct 8Retrofitting language models to operate over bytesAllen Institute for AI · Nature2 citationsOct 7Ream: Unfolding Mutual Awareness in Human-Agent WorkspacesAllen Institute for AI · arXiv (Cornell University)Oct 7Characterizing the role of reasoning tokens in large language models as state over tokensAllen Institute for AI · Nature Machine IntelligenceOct 7Base Models Can Reason By Taking a Cue From Training DataAllen Institute for AI · arXiv (Cornell University)Oct 5Learning What to Imitate: Entropy-Aware Distribution MixingAllen Institute for AI · arXiv (Cornell University)Oct 5HERA: Harness-Environment Co-Evolution for Reliable Agentic AbstentionAllen Institute for AI · arXiv (Cornell University)Oct 5Rubric-Based Optimization for Text-to-Music GenerationAllen Institute for AI · arXiv (Cornell University)Oct 2ScholarCatalyst: A Benchmark for Retrieving Papers That Inspire New ResearchAllen Institute for AI · arXiv (Cornell University)Oct 1Old Ideas, Novel Problems: The Instability of LLM-Based Novelty EvaluationAllen Institute for AI, Microsoft · arXiv (Cornell University)Oct 1You're Hired: Strategic Model Selection for LLM CollaborationAllen Institute for AI · arXiv (Cornell University)Sep 30Long-Horizon Scaling: How Model Capabilities Shape the Returns to ComputationAllen Institute for AI · arXiv (Cornell University)Sep 28From Token Probabilities to Semantic Constraints: Towards Declarative Probabilistic Evaluation of Language ModelsAllen Institute for AI · arXiv (Cornell University)Sep 11Learning to Solve Hard Problems in RL for LLMs by Never Giving UpAllen Institute for AI · arXiv (Cornell University)Sep 11XYBench: Can LLMs Respond Pragmatically to Queries with Misconceptions?Allen Institute for AI · arXiv (Cornell University)Sep 6RATIO: A Benchmark for Retrieval Across Typed Ideation Operations in Scientific LiteratureAllen Institute for AI · arXiv (Cornell University)Aug 27Demystifying Reinforcement Learning Post-Training of Language ModelsAllen Institute for AI · arXiv (Cornell University)Aug 24