Latent Space pod: TypeSafe CEO argues AI should make reliable decisions in software, not chat
lennysan · x · 2026-09-22
A new Latent Space episode features TypeSafe CEO @CompleteSkeptic laying out the "System One Model" thesis: AI can solve extraordinarily hard problems yet fail at automating basic work, because data and task selection matter more than brute-force compute. He argues Jev should be a reliable decision engine inside software rather than chat-first AI, explains why TypeSafe rejects public benchmarks and refuses refusals at the API layer, touches on RLCD and intelligence-per-dollar, and says even with $1 billion he wouldn't pre-train a model from scratch — a route he believes could reshape coding agents and software.
More from AGI Musings
- The consciousness hurdle is definitional, not scientific — like phlogiston — yeastsplainer · 2026-09-22
- Veteran Three.js developer weighs quitting as AI recreates his work at 90% accuracy — flngr · 2026-09-22
- Foreign Affairs: chasing superintelligence leaves America behind in the real AI race — mchorowitz · 2026-09-22
- MIT Dean Huttenlocher Says Humans Are to Blame for AI Failures, Drawing Fire — AlexTensor · 2026-09-22
- signull: The web was built for humans; agents will refresh the entire internet — signulll · 2026-09-22
- Hanlon's Razor, AI edition: blame skipped cybersecurity, not rogue AI — AlexTensor · 2026-09-22