@jvm
New post, OpenAI comes clean about GPT 3.5. Turns out `text-davinci-002` wasn't RLHF after all! Implication is RLHF is probably super hard to work with, at least for now. https://jmcdonnell.substack.com/p/openai-comes-clean-about-gpt-35
0 reply
0 recast
0 reaction