Whilst LLMs still perform weakly in APL, the situation is improving at pace, and giving it a “skill” to evaluate code makes a dramatic difference. I gave a conference talk about it recently (video): https://youtu.be/H_wdKeJ8gt4