AI dev rips 'AI czar' Sacks: try post-training GLM 5.3 with just 3 rules
menhguin · x · 2026-10-12
- Practitioner menhguin strongly rebuts US AI czar David Sacks' dismissal of complex post-training rules in models.
- He proposes a thought experiment: give Sacks one day, a GLM 5.3 pretrain base, and all open internet data — and see if he can post-train a production-usable model with just "3 simple rules."
- He argues specifying behavior and thousands of edge cases for an autocomplete is not "woke bs": Chinese labs' post-training recipes use similarly complex rules, and models can't generate coherent responses without them.
- Follow-up reply calls the take one of the dumbest on the topic, saying the AI czar doesn't understand how models actually work.
More from AGI Musings
- Programming languages were just scaffolding for thinking — for LLMs and humans alike — MarcJSchmidt · 2026-10-12
- dioscuri publishes 'Bad AI Consciousness Takes Bingo' rebutting common LLM consciousness arguments — dioscuri · 2026-10-12
- Karpathy's drill analogy distilled: 5,000 people already got entire factories — _AustinCalvert_ · 2026-10-12
- Szegedy cites his May post: Terry Tao and Judit Polgar misjudge AI's trajectory — ChrSzegedy · 2026-10-12
- AI anxiety among mathematicians mirrors history's craft identity crises, argues researcher thread — soumitrashukla9 · 2026-10-12
- Midjourney founder: human brain and frontier models cost about the same energy per thought — DavidSHolz · 2026-10-12