If we cannot do compute governance, and we cannot do model-level governance, then I do not see an alternative solution. I only see bad options, a choice between an EU-style regime and doing essentially nothing.

It feels like this and some other parts of the post implies the EU AI Act does not apply compute governance and model-level governance, which seems false as far as I can tell (semantics?)

From a summary of the act: "All providers of GPAI models that present a systemic risk* – open or closed – must also conduct model evaluations, adversarial testing, track and report serious incidents and ensure cybersecurity protections."

*"GPAI models present systemic risks when the cumulative amount of compute used for its training is greater than 10²⁵ floating point operations (FLOPs)"

Interested

Agree that would be better. I think just links is useful too though, in case adding more context significantly increases the workload in practice

I wonder if there's an AI tool that could post-process it

Joscha Bach

Is there a group currently working on such a survey? If not, seems like it wouldn't be very hard to kickstart.

Link to mentioned survey: LINK

Maybe someone from AI Impacts could comment relevant thoughts (are they planning to do a re-run of the survey soon, would they encourage/discourage another group to do a similar survey, do they think now is a good time, do they have the right resources for it now, etc)

Thanks, I think this comment and the subsequent post will be very useful for me!

Here's what GPT-3 output for me