from Hacker News

Llama 3.1: Our most capable models to date

by langitbiru on 7/23/24, 3:13 PM with 3 comments

• by ChrisArchitect on 7/23/24, 6:10 PM

  [dupe]

  More discussion: https://news.ycombinator.com/item?id=41046540

  https://news.ycombinator.com/item?id=41046773

• by sagz on 7/23/24, 3:44 PM

  405B is already being served on WhatsApp!

  https://ibb.co/kQ2tKX5

• by msoad on 7/23/24, 3:42 PM

  MMLU PRO is the benchmark I trust the most. I noticed they are using 5 shots and CoT. Is that true for GPT4 and Sonnet as well?