_rpd comments on Making benchmarks outputs directly useful for AI safety and security

[_rpd]

can you make this unlearning method as cheap as possible?

Should “this unlearning method” be a link to a particular method?