Yeah i've noticed this a lot with ibis. Coming from pandas, ibis was a bit of a learning curve and the LLMs were very bad at generating correct code. LLM generated pandas code is generally copy-paste-run and it often just works correctly. Ibis however was usually completely wrong or just slightly tweaked pandas code.

Makes me wonder if each new technology project will need to make an effort to sythensize data to allow for "indexing" by LLMs. It will be like a new form of SEO