About a month ago I offered to help @k00b think through improved web of trust algorithms, which led to me to a ML topic called collaborative filtering.

Collaborative filtering is basically the task of using user ratings of content to determine which users like which content, and to then build a content recommender system. Netflix's recommendation algorithm can be considered a collaborative filtering algorithm. One cool thing about these systems is that you can build a similarity-matrix between users, i.e. How similar is user A to user B in their tastes?

I found out that Netflix ran a prize competition for building such a system, and I asked ChatGPT about the winners:

Ok great. Now I asked Chat about their underlying statistical model.

Cool, so you can derive embeddings for both users and posts using this method.

(There was a lot more stuff about how to actually implement the model on a computer, which I'll skip for now.)

Then, I asked Chat if there are any open source implementations of the algorithm. It pointed me to librec.net

So I go to librec.net and what do I find? It redirects me to a suspicious URL that asks me to install an extension! WTF!?

So I go back to ChatGPT and tell it librec.net is suspicious and it confirms.

Lesson of the story: Don't trust what ChatGPT spits out and always verify. A less cautious person could easily have been pwned here.

Also get your act together, OpenAI... make sure your bot at least checks the links it spits out. FFS