ChatGPT show AI-generated sources

I used a fair amount of AI this weekend. My efforts were mainly around Claud Code, as I created an app for personal consumption (basically making recipe viewing, storing, and using while cooking easier). It was satisfying to ideate, build, and ship something live in a weekend. Much more satisfying than getting a half-baked response from an LLM when discussing things in ChatGPT. I used to be impressed by ChatGPT, but now I almost think of it like Google. Helpful for quick things, but not great for more nuanced work that involves critical thought.

This led me to consider the sources LLMs are built on. Language models are underpinned by enormous amounts of human data. The fascinating issue with this is that we’ve already run out of human data to build upon. Reddit is a popular platform for LLMs to cite. However, I’ve noticed more and more AI-generated posts on Reddit.

What happens when LLMs start citing AI-generated content that’s been placed on platforms historically generated by humans? The doomer mentality would argue it’s the beginning of a degradation of AI products, though there has to be some way to link content with the human behind it (though I’m not certain what this process is, maybe blockchain?). Regardless, this is likely going to be a real problem in the near future we will have to figure out how to solve.

Have a friend who uses ChatGPT for everything? Send them this newsletter. https://sorens.beehiiv.com/