Quantifying detection rates for dangerous capabilities: a theoretical model of dangerous capability evaluations

Open in new window