METR's Eli Lifland: Coding Uplift Should Update Beliefs on Less Verifiable Tasks Too
eli_lifland · x · 2026-11-14
METR researcher Eli Lifland pushes back on the view that coding uplift says little about the main crux of very short AI timelines — generalization to non-verifiable tasks. He argues that treating it as binary is wrong: AI R&D spans a spectrum of task verifiability, so observing larger-than-expected acceleration should update you toward tasks only marginally less verifiable than those already automated.
Related event: METR's "AI doubles AI research speed" report sparks timeline debate(8 posts)→
More from AGI Musings
- Nathan Lambert: premature 'this is AI' claims are wrecking the AI discourse — natolambert · 2026-11-14
- Ex-SSI researcher: rampant cyberattacks are inevitable in any AGI scenario — minhnguyen · 2026-11-14
- Hinton Warns EU Leaders in Dublin: Superintelligence Will Control You, Not Vice Versa — Chris_Armstrong · 2026-11-14
- Google's Multi-Agent Systems Solve Open Problems in Math and Chemistry, Now External — pycui64 · 2026-11-14
- Humans controlling superhuman AI is like eight-year-olds running society, writer argues — binarybits · 2026-11-14
- AI finds shared metabolic signature of five lifespan-extending treatments in mice — Dr_Singularity · 2026-11-14