News for hackers. Stuff that matters.
Re:Capabilities of GPT-4 on Medical Challenge Problems by sebzim4500 on Sunday March 26, 2023 @06:15PM (#35320215)
Is there any indication of whether the correlation was destroyed during the supervised finetuning or during the RLHF phase? Or are there even two phases any more?
Re:Capabilities of GPT-4 on Medical Challenge Problems by og_kalu on Sunday March 26, 2023 @06:18PM (#35320246)
Seems it was fine with instruction fine-tuning. Then gone with the RLHF.
Re:Capabilities of GPT-4 on Medical Challenge Problems by cubefox on Sunday March 26, 2023 @06:34PM (#35320401)
I don't think the paper says that. I would guess both SL and RL cause mode collapse.