When you say system cards, do you just mean “artifacts that AI companies publishing discussing risk/alignment/safety-relevant information about their AIs and their house view on risk (as in the view they are using to make decisions”? If so, this also includes at least risk reports.
Yep, this should include risk reports that the lab publishes themselves. I think there’s less juice in scrutinising third-party risk reports like METR’s.
When you say system cards, do you just mean “artifacts that AI companies publishing discussing risk/alignment/safety-relevant information about their AIs and their house view on risk (as in the view they are using to make decisions”? If so, this also includes at least risk reports.
Yep, this should include risk reports that the lab publishes themselves. I think there’s less juice in scrutinising third-party risk reports like METR’s.