Paper traces limits on chain-of-thought controllability to first tokens after thinking tag

1. 10:27 AM
   elvis@omarsar0X

   Paper traces chain-of-thought controllability limits to the first few tokens

   AIThe paper finds that chain-of-thought monitoring depends partly on models being unable to control what they write in their reasoning. It traces much of that limit to entropy collapse in the first few tokens after the opening thinking tag, where models fall into a few common openings. Prefilling a single token, or banning frequent opening tokens and adding in-context examples, raised average compliance by 34 to 41 points across models.

   

The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.