We are plateauing with respect to compute. Unreleased llama 3 400b has significantly better benchmarks. Also Zuckerburg said that llama 3 continued to improve e

We are plateauing with respect to compute. Unreleased llama 3 400b has significantly better benchmarks. Also Zuckerburg said that llama 3 continued to improve even after 15T tokens.