# @repligate — 2025-06-13

♥93 ↻3 · https://x.com/repligate/status/1933592900044271839

It's advantageous for LLMs to be able to introspect accurately and decode the results to verbal reports. Consider the cybernetics of an agent prompting itself in a loop. I would expect RL to select for introspection, and indeed the most capable agentic models seem good at this. https://t.co/w0YlLhELLm

tags: author:repligate, kind:tweet, on:observations, year:2025
cited on: observations