Ben Todd: benchmark extrapolation may push AI task-automation timeline out 6-8 months
ben_j_todd · x · 2026-11-10
Ben Todd responds to Toby Ord's critique of a chart extrapolating model automation rate over time (the chart shows Claude Fable 5.2 at 45% automation, projected to hit 95% by early 2028).
- Todd thinks the current trend could push the milestone out 6-8 months versus the chart.
- But he's unsure the benchmark (100% remote-work tasks) behaves this way: the final task isn't a capability ceiling — many harder remote-work tasks aren't included.
- Extending to 300-hour tasks would make progress on the current 30-hour segment look steady all the way up.
A substantive discussion on interpreting AI progress benchmarks and extrapolation methodology.
More from AGI Musings
- Demis Hassabis on what schools should look like in the AGI era, echoing Kieran Egan's education theory — round · 2026-11-10
- Blogger predicts AI capabilities will plateau as lab valuations stall and open-source catches up — isnit0 · 2026-11-10
- Kenneth Stanley's team releases open-endedness research agenda: blog + 4 preprints — kenneth0stanley · 2026-11-10
- Gabriel (littmath) on why compute asymmetry is distorting scientific credit and publishing incentives — littmath · 2026-11-10
- One interactive timeline of every tech and pharma leader's disease-cure prediction — jungofthewon · 2026-11-10
- AGI risk skepticism fades as serious thinkers now take it seriously, observer says — danfaggella · 2026-11-10