gwern comments on How might we make better use of AI capabilities research for alignment purposes?

[gwern]

(Your link seems to be missing.)