Comment by Sam Altman

We need a mechanism for complementary national and international frontier AI standards, standards for measuring capabilities, assessing risks, determining whether safeguards are sufficient, and preserving meaningful human oversight as systems become ...

In the sense of, can we really teach the models what collective human values look like? I think the answer is clearly yes, and we have a big effort going on there.

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks. Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We...