Dr. Casey FieslerMay 25, 2023
I'm actually kind of surprised that Open AI wrote this down. Like, straight up stating that though it's so so important that we regulate hypothetical technology that might harm people in the future, we should definitely not regulate the technology that exists today and is harming people in ways we can actually see... is quite a bold strategy. https://openai.com/blog/governance-of-superintelligence
Governance of superintelligence

Now is a good time to start thinking about the governance of superintelligence—future AI systems dramatically more capable than even AGI.

Show threadJames Grimmelmann
@cfiesler I don’t think that’s what they’re saying.
May 25, 2023 at 1:59pmWeb
Show threadFabian TranschelMay 25, 2023
@jtlg @cfiesler What does it mean, then?
Show threadJames GrimmelmannMay 25, 2023
@ftranschel @cfiesler I read it to say that they want licensing and audits for future high-capability models, but not *that type of regulation* for current models.
Show threadDr. Casey FieslerMay 25, 2023
@jtlg @ftranschel I think that's how I read it too... my summary didn't necessarily mean all regulation, but also like, licenses and audits aren't exactly over-the-top regulation?
Show threadKatherine O’KeefeMay 25, 2023

@ftranschel @cfiesler @jtlg

Considering licenses and audits are precisely some of the really basic mitigations and controls that have been proposed to partially address the current harms, it reads to me like exactly that.

Show threadÓscar Morales VivóMay 25, 2023
@jtlg @ftranschel @cfiesler do they define what a “high-capability model” looks like? I suspect these goalposts are going to travel at the speed of Moore’s law.
Show threadFabian TranschelMay 26, 2023
@MyLittleMetroid @jtlg @cfiesler Exactly.