Updated
OpenAI chief scientist Jakub Pachocki published a blog post on November 22, 2026, proposing perplexity as a shared measure for setting limits on AI development, TechCrunch reported. The idea follows calls for a 'pace the frontier' approach: Dario Amodei, Anthropic's CEO, urged companies to agree on common safety standards and limits on the rate of unchecked AI progress, an approach Sam Altman and Elon Musk reportedly endorsed. TechCrunch said any speed limit first needs an agreed way to measure progress, and that benchmarks focused on downstream capabilities have quickly become outdated. Perplexity measures how well a model predicts the next token on average, with lower values meaning better predictions. Pachocki's post includes a chart showing OpenAI model perplexity falling since GPT-3.5. He wrote that GPT-4, at about 0.91, was proficient at chain-of-thought reasoning, and that after reaching losses below 0.8 the company saw transformative capabilities emerge in fields such as cybersecurity and mathematics. He argued perplexity could serve as a shared benchmark for general intelligence, with companies committing not to go past a certain threshold until their ability to understand and monitor AI behavior reaches a concrete milestone. Altman called the post the most important thing OpenAI has published in a while.