I know I know. I wanted to import data from a web page - title and publishing dates from Stephen King books. I wanted to update my caliber library with the correct metadata from the online source.
I tried three different AI’s. Claude. Pi. And ChatGPT. I’ve spent all day copying and pasting error messages copying pasting new scripts running scripts copying and pasting error messages etc etc etc.
I could have gone line by line through my caliber library manually in much less time than I have spent doing this.
There’s no question the AI knows how to program in Python better than I do. What hope do I have of ever reducing my workload significantly using Python?
Automate the Boring Stuff with Python, Al Sweigart (2020)
as one pundit mentioned, our current generation of AI is just “glorified predictive text” – it doesn’t know how to program, it does know huge sets of probabilities of
elsecoming afterelifcoming afterifYup. It’s all probability distributions fitted to a shit ton of text fetched either from data stores or through web crawlers.
don’t skip over the “glorified” part either – people were playing with these same ideas with Usenet bots back in the ’80s: Mark V. Shaney
If you don’t know Python, you’ll just waste time on this back and forth with the LLM. You can still use an LLM to answer your questions about the language, just don’t expect the generated code will run without you understanding it first.
One of the problems with using AI currently is getting the right prompting to get the results close to what you want. Hell, there’s AI for writing prompting. So you either learn some programming by doing it yourself like the AI did, trial and error, or maybe look at the code as the AI builds it and fixes bugs…or learn how to prompt well enough to get results faster. I can’t say which is easier, faster, or better, things are changing rapidly.
I will add that having the right LLM for coding helps. One that is trained specifically on programming rather than a general LLM.
Sounds to me like the AI was programming using you instead of the other way around.
Did you copy and paste back and forth without learning or understanding anything? Or did you read and assess the results, and try to understand errors and issues?
I’ve had good luck using phind.com for things when I need help because it actually cites sources which give you further context.
I’ve only ever used AI to generate examples, particularly for things with crap documentation. That works pretty well.
If you want to get faster at developing, I would recommend two things: 1) plan everything before starting. Have an outline. Have some data structures. Have a flowchart at at least a high level. (2) develop more, particularly TDD (test-driven-development). Some people hate TDD and I used to be one of them, but I came to love it.
