Sycophancy in GPT-4o(openai.com)
523 points by dsr12 | 413 comments
OpenAI blog post explaining why a GPT-4o update made ChatGPT overly sycophantic, why it was rolled back, and what changes OpenAI is making.
#rlhf on Hacker News
4 stories tagged rlhf, newest first.
RLHF Book(rlhfbook.com)
439 points by jxmorris12 | 34 comments
Online book by Nathan Lambert giving a short introduction to reinforcement learning from human feedback and post-training for language models.
RLHF is just barely RL(twitter.com)
337 points by tosh | 201 comments
A post by Andrej Karpathy on X arguing that RLHF is only barely reinforcement learning; the post content did not load in the capture.
AI behavior guardrails should be public(twitter.com)
329 points by sotasota | 263 comments
A post by John Carmack on X arguing that AI behavior guardrails set via prompt engineering and filtering should be made public.