OpenAI-knowledge: minimalist design that brings documents into context

The most annoying thing about writing code with AI is not that it can’t be written. It is written out, and the parameters are all wrong. The function name is correct, the idea is correct, even the error handling is decent, and then it hangs as soon as it runs. Because max_tokens is written as max_token, the valid range of temperature is written as 0-2 instead of 0-1 as clearly stated in the document. This experience of "everything looks right, but everything runs wrong" is especially serious when calling the OpenAI API. The model updates quickly, and the documentation changes accordingly. By the time you finish learning the parameter signatures of the previous version, the next version has already been changed. Letting AI write code based on the memory in the training data is essentially betting that it remembers it correctly. OpenAI-knowled