Iirc, the translation example was in some comment on one of the AI Alignment blog posts from ~2016, though I can’t find it right now.
I wrote about translation in connection with IDA in this 2016 comment thread, although the form of the problem I wrote down is different enough from Paul’s 2020 writeup that I’m not sure how much connection there is between the two.
I wrote about translation in connection with IDA in this 2016 comment thread, although the form of the problem I wrote down is different enough from Paul’s 2020 writeup that I’m not sure how much connection there is between the two.