For people giving out AI advice, here’s mine: Go use crappy models. Use old Claude models, Kimi, locally hosted ones – get good at making bad models do good things through system prompts, evals, quality prompts, etc.
This is your value. Not “I told mythos to do a security scan”
— Mark Manning (@antitree)
Apr 25, 2026
from http://twitter.com/antitree
via IFTTT