Vai Pahwa
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Double Blind Evals: Resolving the Dual Confidentiality Dilemma in AI Safety Auditing
According to the abstract, Andrew Trask, Sol Messing and colleagues propose double-blind evaluations using secure enclaves, so an evaluator's prompts stay hidden from the model owner while the model's weights stay hidden from the evaluator. They describe a pilot testing a Google DeepMind model against part of the private AILuminate benchmark.