Pattern comments on [AN #85]: The normative questions we should be asking for AI alignment, and a surprisingly good chatbot

[Pattern]

Errata:

we only get data [from] the outer objective on the training distribution,

[Rohin Shah]

Thanks, fixed.