Post

Re the regressions - have you tried getting the LLM to write some unit tests for the things that work well (eg your 2017 docs) and then telling it not to change those? I’ve found that quite useful for directing it to make more orthogonal changes that don’t break existing stuff.