Gradio founder: let users read full unencrypted reasoning traces to align LLMs

mmitchell_ai · x · 2026-09-16

Abubakar Abid (Gradio founder) argues that one thing that would help right now to ensure LLMs behave in an aligned way is letting users read their full, unencrypted reasoning traces. The take was amplified by Margaret Mitchell's retweet, tying into ongoing debates over chain-of-thought obfuscation and lost monitorability.

Original post →

More from AGI Musings

AGI Musings channel →