Menu
Sign In Pricing Add Podcast

Lex Fridman Podcast

#434 – Aravind Srinivas: Perplexity CEO on Future of AI, Search & the Internet

4818.115 - 4841.299 Aravind Srinivas

It's not easy to make these systems controllable and well-behaved without the RHF step. By the way, there's this terminology for this. It's not very used in papers, but people talk about it as pre-trained, post-trained. And RHF and supervised fine-tuning are all in post-training phase. And the pre-training phase is the raw scaling on compute.

0
💬 0

Comments

There are no comments yet.

Log in to comment.