Ethan Perez comments on Discovering Language Model Behaviors with Model-Written Evaluations

[Ethan Perez]

Thanks for catching this—It’s not about sycophancy but rather about the AI’s stated opinions (this was a bug in the plotting code)