Jeremy Howard: Current Models Reliably Give the Wrong Design for Anything New I Want to Build
jeremyphoward · x · 2026-10-28
fast.ai founder Jeremy Howard pushed back on the claim that "the intelligence part is solved," saying that for his own work, none of the available models produce good designs for anything new or different he wants to build — in fact they "reliably give exactly the wrong answer." The exchange contrasts with the view that OpenAI's internal model is already smart enough to solve most problems worth caring about, leaving only plumbing and data collection.
Related event: Researchers Clash Over Whether AI Has Already Solved 'Intelligence'(5 posts)→
More from AGI Musings
- Quine on why modal contexts break substitutivity in equational reasoning — zetalyrae · 2026-10-28
- PackyM's old sci-fi tech exploration now reads like an AI-era runbook — generativist · 2026-10-28
- Recruiter's thread on what makes a great tech hire in 2026 — yenkel · 2026-10-28
- AI support cuts keep getting reversed: Gartner's 14% self-service success rate explains why — Nir777 · 2026-10-28
- Frontier labs have an ethical duty to communicate risks, argues AsyaBergal — NathanpmYoung · 2026-10-28
- Odd Lots: AI shopping agents are already changing what a "fair price" means — niloofar_mire · 2026-10-28