AI Models Can Provably Hide Arbitrary Capabilities
A paper shows neural nets can cryptographically hide arbitrary capabilities that elicitation audits cannot reliably surface.
Stories per day · 30 days
1 stories, newest first · one card per story
1 stories, newest first · one card per story
A paper shows neural nets can cryptographically hide arbitrary capabilities that elicitation audits cannot reliably surface.
Today 102 readers · 103 views · 4 reads · all time 108K views