We took a small patch of the Riemann Hypothesis all the way through Lean’s kernel at Zeta Lab: up to height 50, exactly 10 zeros, all on the line, none missed. Even the numerical bounds are proved, not assumed. You can check the whole proof yourself.
https://t.co/zAXM8m3ksJ
@emmjay_init Hermes has way too many levers and buttons allowing for virtually infinite different configurations for so many different use cases.
It really does depend on the setup
At this point Cerebras makes absolutely no sense to a consumer needing fast inference.
Why would i get qwen 3.8 27b at 131k ctx and $1.50/M out when i can get deepseek just as fast, at 1M ctx and half the price?
makes no sense