In work it’s common for me to put together working prototype software to demo feasibility of some approach. The knowns are libraries, languages and sometimes s

In work it’s common for me to put together working prototype software to demo feasibility of some approach. The knowns are libraries, languages and sometimes sample code. Usual workflow is lots of Googling and reading documentation to get something out in a few days. GPT saves me lots of time researching and has effectively replaced Google and StackOverflow. It allows me to design a plan of attack from “use these 3 libraries to do x” to a working prototype which I can then iteratively refine. It also is good at answering technical questions about library and language features. For example: “I have a pandas data frame, show me how to loop through it and access columns 3 and 4”. Once I see the API call is iloc I can research more. However there are a few issues with GPT: 1. Makes up APIs that don’t exist. 2. Uses APIs incorrectly. 3. Not up to speed on the latest APIs, or maybe chooses not to use them? For example I’m working with a library that has two functions: create_group() or get_or_create_group(). GPT seems to not acknowledge that get or create exists. Some usage tips: 1. I usually tell ChatGPT in my first question is: “don’t respond with too much text or detail. If I need more clarification I will ask.” 2. Use Shift+Enter to formulate my initial question in detail. For example if I say “let’s write a Python app” it will respond with “sure Python is a great language to…”. So give it enough detail to get to the point.