Hugging Face Highlights 10 AI Papers Spanning Agents, RL, and Diffusion Models

On July 6, 2026, the Hugging Face community upvoted 10 research papers reflecting key trends in artificial intelligence development. The highlighted works cover areas including long-horizon AI agents, reinforcement learning for large language models, inference optimization, and controllable generative models. Notable among them is the Program-as-Weights concept, which proposes converting natural language task descriptions into compact neural artifacts executable without relying on large online models. Another standout paper addresses the training-inference policy gap in RL for LLMs, arguing that optimization should target inference-time performance rather than training reward alone. A third paper, AgenticSTS, introduces a bounded-memory testbed that enables controlled analysis of how different memory components affect agent performance.