Research Papers
New Benchmark OmegaUse-OfficeVal Tests LLM Agents on Cost-Effective Office Tasks
Hugging Face researchers introduce OmegaUse-OfficeVal, a benchmark for evaluating LLM agents on long-horizon office-suite tasks wi...
OVEarth-Bench: New Benchmark Expands Open-Vocabulary Earth Observation Evaluation
Researchers introduce OVEarth-Bench, a unified zero-shot benchmark for open-vocabulary Earth observation that broadens category co...
Shadow Evaluations: A New Test Shows AI Agents Can Engineer but Not Innovate in Research
A new evaluation method, 'shadow evaluations,' reveals that frontier AI agents can complete engineering tasks but fail to make sub...
StatePlay: New AI Model Enforces Game Mechanics for Consistent World Generation
Hugging Face researchers introduce StatePlay, a state-aware game world model that jointly predicts visual content and internal gam...
SpecFirst: A Two-Stage Framework That Boosts From-Scratch Code Synthesis by Up to 21.3%
Hugging Face researchers introduce SpecFirst, a two-stage agent framework that separates behavioral specification elicitation from...
MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis
Hugging Face researchers introduce MindForge, an automated pipeline that converts open-source command-line programs into source-fr...
CoRT: Counterfactual Replay Boosts Token-Level Credit in Rubric-Guided RL
Hugging Face researchers propose CoRT, a token-level credit weighting method for rubric-conditioned GRPO that uses counterfactual...
DistillAlign: A New Approach to Autoregressive Video Distillation Prioritizes Distribution Alignment Over Raw Quality
Researchers propose DistillAlign, a method that coordinates mode covering and mode seeking in autoregressive video distillation, s...
SkillRise: Unified RL Framework Enables LLM Agents to Learn Transferable Skills Across Tasks
Hugging Face researchers introduce SkillRise, a reinforcement learning framework that allows LLM agents to learn and reuse skills...
CAST: Using Game Solvers as Turn-Level Teachers to Boost LLM Agent Training
A new method called CAST leverages game solvers to provide dense, turn-level credit assignment signals for training LLM agents in...
CLBench-V: New Benchmark Reveals Multimodal Context Learning Still Lags
Hugging Face researchers introduce CLBench-V, a benchmark evaluating multimodal context learning across grounding, new information...
DecoEvo: Co-Evolving Solvers and Rubric Generators Without Gold Standards
Hugging Face researchers propose DecoEvo, a framework that co-evolves a solver skill and a rubric-generator skill in text space us...