I recently moved my wiki notes to a different platform and wanted to build a Python app to read each article and add a list of keywords and categories describing it, so things would be easier to find on the new site. I ended up attempting to create a natural language model, but literally every unique word in each article kept becoming a keyword and category. After hours of toil and StackOverflow, I realized that I was essentially trying to recreate ChatGPT or at least an aspect of what it could do seamlessly. Instead of just pivoting to its API’s, which would have only cost a few cents, I spent a few more hours trying to use a wrapper around Bard that was broken because the service didn’t want people to build free automation atop it. I finally wrote a script that used ChatGPT/OpenaAI API’s to feed the articles and it worked almost perfectly. Had to run some parsing, but it got where it needed to be. TL;DR: I tried to write an AI in a day that is difficult for seasoned experts, failed pretty gloriously, realized I should use an existing AI, then wasted a few more hours trying to save a few cents, then did it correctly. I hyper-focus when coding, which can lead to these rabbit holes.