Analyzing METR dates to estimate Anthropic's internal capability lead — a 7-day shift moves the integral over 40%
teortaxesTex · x · 2026-11-06
tenobrus shows you can reverse-engineer from METR's "majority internal access" dates how long frontier models are used internally before public release, then compute a cumulative "internal capability gap integral" (AECI-days) tracking Anthropic's accumulated advantage over time.
Skeptics counter that the implied gaps — often just days to two weeks — are implausible: employees almost certainly had access to earlier, similarly capable checkpoints long before the named model's official internal date. The reported date likely only reflects the exact final checkpoint. By this reading, even a 7-day earlier access shift pushes total AECI-days up over 40%, meaning these internal-vs-external gap charts are highly misleading.
More from AGI Musings
- Palisade's Jeff Ladish: skip the 40 benchmarks, ask if agents finish real work — FinanceYF5 · 2026-11-06
- Safety critics slam Anthropic's acceleration plan, urge government to stop it — AndyMasley · 2026-11-06
- Naval on why innovation is always geographically concentrated, from Venice to Silicon Valley — naval · 2026-11-06
- Ex-OpenAI safety lead: AI policy world overindexes on OpenAI and Anthropic — Miles_Brundage · 2026-11-06
- Ethan Mollick: skip the nation-of-geniuses wait, fund achievable national AI crash projects now — emollick · 2026-11-06
- Laid-off Microsoft manager making $320k can't find a job, runs out of money in 10 months — RexDouglass · 2026-11-06