Raw LLM Responses

Sounds like they did something along those lines, given they said they were using Puppeteer to scrape websites, but when they sent the scraped contents to AI (which were garbage), it still cost them tokens.  Although it is crazy to me a novice would set up their AI API without limits in place (at the very least). The problem is probably not so much to do with their architecture per se, but how poorly thought out their process was when you get into the nitty gritty, and the lack of guardrails in place for when things go wrong.  Admittedly, it can be like playing a game of wack-a-mole.