Tag
Omni-IO Skills presents a plug-and-play agent harness that enables omni-native capabilities across multiple modalities, significantly improving performance on multimodal benchmarks like UniM-90 without changing the agent's core reasoning.
SkillGym transforms human-written agent skills into executable training environments for LLMs, enabling supervised fine-tuning and reinforcement learning to enhance real-world problem-solving capabilities and performance on benchmarks.
The Large Knowledge Model (LKM) introduces a scientific knowledge infrastructure that transforms research papers into reasoning graphs, creating a Scientific Reasoning Landscape to support scientific search, question answering, and research planning with demonstrated accuracy improvements on benchmarks.
Kyutai Labs releases Voice of Reason, a speech-native model that solves spoken math problems without transcription or text LLM, achieving 77.1% on GSM8K, up from GLM-4-Voice's 27.3%.
The author argues that AI is far from AGI and that Silicon Valley CEOs are overhyping AI risks to raise funds, while actual progress is incremental and not revolutionary.
PaperGym is a framework that converts scientific papers into training environments by separating research questions from evaluation rubrics, enabling reinforcement learning to improve research planning. It demonstrates improved performance over existing methods, with trained models outperforming larger ones on benchmarks.
This paper proposes Intention Distillation (INDI) to distill behavior intent into the action decoder of Vision-Language-Action models, improving performance on benchmarks like SimplerEnv-Bridge and real-world tasks.
SkillGate addresses selector credit starvation in AI agent skill selection by using separate credit channels for execution and skill-naming tokens, improving success rates and reducing misleading skill exposure in long-horizon tasks.