Yeah the “receiving” app is Java. Kind of comical that the Java code generated by the python generator is better than most of the Java code I have been able to get Claude to write :).

Cf. OpenAI “Why LMs Hallucinate.” 70%+ calibration error rates on code generation tasks. Any distributional “intuition” is bound to fail. If we could hone intuition to predict errors, so could the LLMs. If the code matters, you need to look at it carefully.