But for a new language, there isn’t a large body of work to train the LLM on what idiomatic code looks like. It really worries me, that we’re going to be stuck at this local maximum