MIT CSAIL paper: frozen LLM agents design their own self-improvement without being told how

omarsar0 · x · 2026-11-19

A new paper from MIT CSAIL and the University of Tübingen directly tests whether LLM agents still need to be told how to self-improve. Answer: no.

Setup: a frozen model gets a compute budget and simple tools — it can write practice games, watch another copy of itself play them, and edit that copy's prompt and skill files. It never sees real tasks or validation scores and is given no learning instructions.

Findings:

The authors note this matters for safety too — when self-improvement no longer requires a human-specified loop, alignment assumptions need revisiting.

Related event: MIT Study: Frozen LLM Agents Can Invent Their Own Self-Improvement Algorithms(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →