Elon talks about SpaceX renting compute to Anthropic and AI safety regulation
Elon said that before SpaceX rented compute to Anthropic, he first met with a group of people from Anthropic to test his "evil detector," and felt they were friendly and responsible, so he planned to move the cooperation forward; when asked whether he also talked with Sam Altman, he said bluntly that the man is too good at deceiving people. Regarding AI safety regulation, he proposed that several frontier model companies have a phone call once a week, and one week before a new model is released, open it to competitors for trial use first, and if problems are found, fix them before release, believing this is faster than waiting for regulators to make rules.
Body · AI translation
Why would @SpaceXSI provide compute to Anthropic? Elon said that before renting it out, he made a point of going to meet a bunch of people at Anthropic to test his "evil detector," and everyone turned out to be friendly and conscientious—he didn't run into anyone who seemed shifty 👀 So he plans to talk with Anthropic about this idea. As for whether he could also talk to Sam Altman? Elon said flatly: that guy is too good at lying 😆
Rather than waiting for regulators to slowly make rules, why not let competitors keep an eye on each other! On AI safety regulation, Elon's proposal is simple: several frontier model companies hold a call once a week, and a week before a new model is released, open it up for competitors to try out first. If someone says "I used your AI to build a nuclear bomb," then fix it quickly before releasing. In his view, regulators take a long time just to get established, and rulemaking is even slower; it's competitors who can best keep each other honest.