Kokotajlo: OpenAI still hasn't given METR the promised employee-level access
DKokotajlo · x · 2026-11-23
Former OpenAI researcher Daniel Kokotajlo says both OpenAI and Anthropic are far less transparent than they should be: OpenAI still hasn't given METR (or anyone) the promised employee-like access, nor allowed a thorough investigation of the Hugging Face weights incident. David Manheim adds that standardizing on perplexity as a capability bound would favor OpenAI, since Anthropic's finer-grained tokenizer yields inherently lower per-token perplexity.
More from Models
- Musk: data starvation will be the training challenge next year, mocks 'pretraining' term — elonmusk · 2026-11-23
- $200 Google AI Ultra subscription quietly includes a 96GB Blackwell GPU for 10 days — kieranklaassen · 2026-11-23
- Tesla FSD insider: end-to-end is settled, winning hinges on data diet quality — Singularitarian · 2026-11-23
- Google sweeps top 3 on Arena text leaderboard as Gemini 4 Carbon scores 1527 — demishassabis · 2026-11-23
- Google's BuggyMLE injects bugs into ML repos, dropping top model solve rate from 95% to 9% — omarsar0 · 2026-11-23
- OpenAI chart reveals unreleased GPT-6 Astra models; Domingos questions billions for <0.3 bits — pmddomingos · 2026-11-23