MIT: capable frozen LLM agents autonomously invent their own self-improvement algorithms
MIT · hf · 2026-11-18
Research from MIT CSAIL and Tübingen (Sundaram, Sancaktar, Isola) studies an autonomous self-improvement paradigm: given basic primitives, resources, and a general self-improvement goal, a frozen LLM agent constructs its own context-based self-learning algorithm. With a compute budget and tools for collecting experience and updating text documents, the agent freely experiments without an externally imposed loop or real-world data access. Sufficiently capable models do self-improve, and the resulting text artifacts transfer to new game and 3D navigation tasks.
More from coding & agent
- Dev's Grok bot will auto-run a GTA 6 fan wiki, aggregating 5 platforms every 15 minutes — Rasmic · 2026-11-18
- AI-Generated Code Is Forcing Software to Become a Real Engineering Discipline — PaulShellDev · 2026-11-18
- Screenshot shows Claude Code injecting 'NOT USER INPUT' notices for background tasks — philhchen · 2026-11-18
- Cloudflare open-sources security-audit-skill, a 6-phase agent security auditor — _jaydeepkarale · 2026-11-18
- AI-era code is drowning teams in vulnerabilities—why scanners and manual reviews fall short — _jaydeepkarale · 2026-11-18
- Agentic on-policy distillation implicitly reshapes ReAct patterns and causes negative transfer — Deyang Kong · 2026-11-18