Issue 2026-11-22 · Industry · 研究 · 安全
OpenAI chief scientist urges perplexity as shared AI model metric

OpenAI Chief Scientist Jakub Pachocki argues for adopting perplexity (cross-entropy loss) as a shared metric to compare model developers and inform common safety standards, and publishes OpenAI's internal pretraining perplexity measurements to build a shared scientific basis for evaluation.
Read original ↗