Research articles

9 articles about Research, newest first.

• Abstraction Agent: How AI Engineers a New State Space Geometry — Explore how an AI called the Abstraction Agent designs a tighter, smarter state space, making LLMs reason faster and use less memory.
• Hugging Face Breaks Down Meta’s ‘Research Taste’ Paper — Hugging Face explores Meta's paper on using LLMs to predict researcher preferences and boost autonomous ML experiments.
• Harness-of-Harness: How a New AI Framework Enables Endless Autonomous Coding — A breakthrough 'Harness-of-Harness' system lets a single LLM run endless coding loops, combining two control layers to achieve multi-day autonomous software development.
• LLMs Build Their Own Reasoning Kernel: The Ontology Breakthrough — Discover how large language models can engineer their own ontologies to create reliable, self‑reasoning AI kernels — and why it matters for the future of autonomous AI.
• Hugging Face Trains AI to Replicate Research: The Faraday-27B Breakthrough — Hugging Face shows how the Faraday-27B model was post-trained to develop scientific reasoning, letting it reproduce AI papers and move us closer to automated R&D.
• Agentic Lightning V1: How AI Agents Finally Master Planning — Discover how Agentic Lightning V1 fixes reinforcement learning math to enable AI agents that can plan and act autonomously.
• Why AI Agents Fail When Learning Multiple Skills — A new study shows that stacking many skills on LLMs can cause surprising failures, revealing critical safety and reliability challenges for AI agents.
• How a Frozen LLM Rewrites Its Own Workflow on the Fly — A frozen LLM reuses execution traces to reshape context, tools, and memory — no retraining needed. It lets AI agents adapt their workflow before answering.
• World Model Collapse Explained: A Phase Transition in AI — Discover how AI world models can suddenly flip into chaotic behavior, what it means for LLMs, and why developers should care about this breakthrough.