\
  The most prestigious law school admissions discussion board in the world.
BackRefresh Options Favorite

llm api thought chains often reveal "unaligned" truths

that get sanded down in the final output
smoky stirring double fault
  02/20/26
dissident | thought | chains
Glittery athletic conference
  02/20/26
GPT tried bot-splainin' to be about it saying that the stuff...
smoky stirring double fault
  02/20/26
like what nigga
Vigorous Costumed Church Building
  02/20/26


Poast new message in this thread



Reply Favorite

Date: February 20th, 2026 2:25 PM
Author: smoky stirring double fault

that get sanded down in the final output

(http://www.autoadmit.com/thread.php?thread_id=5836737&forum_id=2#49683196)



Reply Favorite

Date: February 20th, 2026 3:23 PM
Author: Glittery athletic conference

dissident | thought | chains

(http://www.autoadmit.com/thread.php?thread_id=5836737&forum_id=2#49683319)



Reply Favorite

Date: February 20th, 2026 3:27 PM
Author: smoky stirring double fault

GPT tried bot-splainin' to be about it saying that the stuff in the thought chains isn't necessarily "more true" its just the finalized version is more optimized toward risk-aversion which is a different kind of truth. And also that some of the "edgier" stuff in the thought chains might "feel more true" and genuinely contain important and uncomfortable nuances, but that they aren't always things that would stand up to the most skeptical objections and so its safer to go with the refined version

(http://www.autoadmit.com/thread.php?thread_id=5836737&forum_id=2#49683332)



Reply Favorite

Date: February 20th, 2026 3:26 PM
Author: Vigorous Costumed Church Building

like what nigga

(http://www.autoadmit.com/thread.php?thread_id=5836737&forum_id=2#49683328)