Herbie Bradley argues AI research taste barely transfers to automating AI research — the gap is bottlenecked elsewhere
herbiebradley · x · 2026-11-11
In a technical timelines debate, Herbie Bradley explains his #1 disagreement with AI 2027-style models: the gap between AI research taste and TED-AI (automated AI research skills). He argues research taste transfers only indirectly — it speeds crossing the gap up to a point, but overall progress is quickly bottlenecked by slower non-research factors. He also notes many skills (e.g. corporate law) take over a year to surpass humans, while conceding these concepts are hard to operationalize.
Related event: Researcher Questions Using Research Taste to Timeline Automated AI Research(2 posts)→
More from AGI Musings
- "Techno-utopianism is dead": researcher laments AI has no shared prosperity narrative — jd_pressman · 2026-11-11
- New paper digests OpenAI's Haldane-gap proof with tensor networks, formalizes it in Lean 4 — danintheory · 2026-11-11
- Hot take: Artists who hate AI most are those who failed before it existed — flowersslop · 2026-11-11
- Industry will regret not hiring junior developers, warns 3-5 year prediction — bendee983 · 2026-11-11
- NYT letters: if an AI has its own interests, ownership doesn't give vendors the right to speak for it — nordicinst · 2026-11-11
- AI100 founder Eric Horvitz shares framing memo listing 18 evergreen AI study areas — erichorvitz · 2026-11-11