Back to Latest SignalTECHMEME / MODEL RELEASE

OpenAI shares its models' perplexity, or how well they predict data, and urges pacing limits based on such metrics, citing a possible 2027 avalanche of AI gains (Jakub Pachocki/OpenAI)

Source: TECHMEME · NOV 22, 2026

Jakub Pachocki / OpenAI : OpenAI shares its models' perplexity, or how well they predict data, and urges pacing limits based on such metrics, citing a possible 2027 avalanche of AI gains — - Difficulty of comparison across labs. … - Detachment from actionable evaluations. … - Not accounting for advances outside of pretraining.

IN THIS STORY
  • OpenAI
  • AI Safety
  • Jakub Pachocki

Sources

  • Read the original at TECHMEME

Related on Frontier

  • Jony Ive, ex-OpenAI product designer Joey Flynn, and others on Silicon Valley's obsession with "taste" as a moat against AI slop and whether AI can learn it (Jonah Weiner/New York Times) Nov 22, 2026
  • NYU's Tristan Buckmaster and Anthropic's Levent Alpöge publish an AI-aided proof of Navier-Stokes blowup, saying it closes a loophole in OpenAI's solution (Peter Hall/Scientific American) Nov 21, 2026
  • Oracle says it has secured funding for its infrastructure contract with OpenAI and "any further private discussions with vendors are just additional options" (@oracle) Nov 20, 2026
  • Combatting adversarial distillation Nov 20, 2026
  • How OpenAI’s Performance Factory keeps ChatGPT fast Nov 19, 2026
MORE LATEST SIGNALSIGNALS