Toby Ord: AI's 80% time horizon on AI R&D is just 15 minutes at OpenAI and GDM
tobyordoxford · x · 2026-11-07
Oxford philosopher Toby Ord highlights an underreported "bearish" fact: at OpenAI and Google DeepMind, the 80% time horizon for AI completing AI R&D tasks is only about 15 minutes.
- He investigated and reported the figure himself precisely because it cuts against bullish AGI narratives
- In follow-up replies he clarifies that his critiques shouldn't be read as opposing short timelines — he actually thinks most people's timelines are too long, which he calls the bigger problem
- He says he mostly reads optimistic takes from people he respects, so he still insists on calling out flawed analysis on "his own team" — self-censoring is bad epistemics
Related event: Oxford Philosopher Says Most AGI Timelines Are Too Long(3 posts)→
More from AGI Musings
- Creators argue AI writing is flattening prose into a monoculture, betting on human connection — _akpiper · 2026-11-07
- US construction workers are 40% less productive than in 1970, economist finds — aakashgupta · 2026-11-07
- Dario's 10k-word essays called the EA/AI safety crowd's kryptonite — shakoistsLog · 2026-11-07
- Less-verifiable benchmarks speed up just as fast: 1.64x vs 1.67x in Anthropic data — ben_j_todd · 2026-11-07
- Redacted METR report cross-checked with public data: Anthropic's AI-driven 2x acceleration is probably real — ben_j_todd · 2026-11-07
- Ben Todd: the biggest question in AI is whether acceleration extends beyond verifiable tasks — ben_j_todd · 2026-11-07