Today I have an RSS reader that sucks in articles from 110 feeds. Over the course of a week it ingests maybe 10,000 articles and picks 300 to show me which I make a thumbs up or thumbs down judgement which is then used to train the model that selects articles. It works great after getting a few thousand judgements and I love the feed I get. I can picture it as a consumer product with the caveat that I think most people would expect to get good recommendations after many fewer judgements. If I was going to productize it for consumers I'd probably use some kind of collaborative filtering (people who read these articles also liked X) as opposed to the content-based filtering it uses now (article X shares some characteristics with other articles you like) because I think it could cold-start for a new user more quickly but then it becomes another StumbleUpon, etc.
I think you can imagine the same kind of system could be used by anyone who is a "professional searcher". Instead of blog articles these could be resumes, or patents, or job listings, or scientific articles, etc. Personally I am more interested in this path for productization.
Does this sound interesting to you? I demo this system a lot and I can do it for you.
https://arstechnica.com/information-technology/2024/09/gener...