Toby Ord: Opus 5.5 system card yields lambda values matching OpenAI's agent parallelism estimates
tobyordoxford · x · 2026-09-30
Toby Ord points out that Anthropic's new Opus 5.5 System Card can be used to back out λ values for AI task parallelism, alongside OpenAI's published figures.
He reports getting values similar to OpenAI's for all tasks except one — generating a knowledge base — where λ ≈ 1, since the card states the task was simply cut into 100 pieces with each agent doing one, yielding no parallelism gains.
More from Research
- NVIDIA/MIT's Physis-Lang tops Physics-IQ leaderboard, +5.5 points via evolved captions — songhan_mit · 2026-09-30
- LeCun co-authors AD-E2E-JEPA: world model drives end-to-end without any driving policy — ylecun · 2026-09-30
- Three Meta Papers Show LLMs Struggle at Code Complexity Optimization, and How RL Can Fix It — TimDarcet · 2026-09-30
- Robot Policy Metadata Conditioning Fails: Text Prompt Completely Ignored — DominiqueCAPaul · 2026-09-30
- Researcher: Dropping 'everything must be differentiable' could unlock AI's next leap — PMinervini · 2026-09-30
- Writing evals with per-sample rubrics may be exploitable via subtle benchmax overfitting — teortaxesTex · 2026-09-30