Agree with everything you said. > One thing that concerns me with making such a system is that one should be really careful when collecting people's data and p

Agree with everything you said. > One thing that concerns me with making such a system is that one should be really careful when collecting people's data and preferences and then recommending based on that (thinking of YouTube recommendation slippery slopes). The plan is for the entire recommendation system to be open source and customisable. Because it's using LLMs as opposed to traditional ML models, it can be more easily reviewed by the end user and customised with natural language prompts. You can see the repo here if you are curious: https://github.com/OpenPipe/open-recommender https://github.com/OpenPipe/open-recommender