upvote
> let the LLM generate slop, check it using formal methods

I'm much more bullish on the opposite approach. Perform the naive translation, let the LLM loose on cleaning it up...

reply