OpenAI Chief Scientist Jakub Pachocki has proposed using perplexity as a shared industry metric for tracking AI progress and establishing safety standards.
OpenAI Chief Scientist Jakub Pachocki has proposed using perplexity as a shared industry metric for tracking AI progress and establishing safety standards.
The measure evaluates how accurately models predict the next token, with improvements associated with emerging capabilities in mathematics, cybersecurity, and other complex fields. Pachocki also suggested that AI could help accelerate fundamental research as early as 2027, emphasizing the need for coordinated safety limits across AI developers.