Cole Wyeth comments on TASTE: Can AI Models Judge AI Safety Research Proposals?

[Cole Wyeth]

This sounds almost exactly the same as creating a benchmark for general research taste, which Anthropic can then climb.