> Second, then the safety around the model. Ad runtime. We have lots of classifiers around harmful content or bias, which we then catch. And then, of course, the takedown. Ultimately, in the application layer, you also have more of the safety net for it. So this is all going to come down to, I would call it, the everyday engineering practice.

Is the piece I’m remembering

They want to avoid their new chat bot revealing their secret love of Hitler like the last one.