OpenAI's Chief Scientist Jakub Pachocki thinks it's time for a new rule: before AI gets too advanced, companies should measure how well it predicts what comes next: a score called "perplexity."
Lower perplexity means smarter predictions.
The idea? Slow things down until we actually understand what our AIs are doing.
Industry leaders push shared safety rules
This move isn't just about numbers: it's about safety.
Industry leaders like Anthropic's Dario Amodei have been calling for shared rules to keep AI from running wild, and now Sam Altman and Elon Musk have seemingly endorsed the broader pace-the-frontier plan, while Pachocki proposed "a shared benchmark for general intelligence" in his post.
If companies work together on this, we might get smarter AI without the scary unknowns.