Broderick McDonald comments on DeepMind: Model evaluation for extreme risks

[Broderick McDonald]

Useful model evals but more focus is needed on near-term risks from malicious actors