Judge feedback

Judge feedback is anonymous.

• - Set of standards: what is in those standards? It seems rather vague now. - Good you gave the political and implementation side detailed thought. Might make sense to have sanctions only coming from member state so the threshold to join AIRO is low. - Escalating daily costs for computing power or restricted access to shared data systems could be interesting - Seize 90% of its computing power, leaving 10% of computing capacity, the lab can develop safety measures to address the issues found in the audit. Interesting idea. - CAIRO: interesting idea! Not sure it this would necessarily slow things down, but it would centralize them, perhaps making ovrsight and safety easier to achieve. However, politically difficult to achieve. - Over all, the content of the audits that an AI company would have to meet are unspecified. Possibly, there is no (known) way to train a future large language model safely. If this would be the case, a significnt part of the proposal doesn't work anymore. - While the regulatory part is very well thought-out, the technical side is not very complete. I would say this proposal would be promising with more technical input to bolster e.g. the audit content.