‘Doomers’ push for compute thresholds in comments to NIST on AI misuse risk

Stakeholders expressing concern over catastrophic risks from the development of foundational artificial intelligence models advocated the use of controversial thresholds for mitigating misuse of the technology, in comments to the National Institute of Standards and Technology.

“Include dynamic thresholds based on training compute as a factor in initially estimating a model’s capabilities of concern,” read the comments from the Center for a New American Security, which NIST published Sept. 10 following a request for public feedback on its ...