I think users are probably capable of providing "seeds" and "preferences" but tbh, I still think actual algorithm development will live in bigger units/teams/businesses. Like "give me the rage algorithm with particular attention to politics, lobbying and this one critter in particular" or "give me the optimistic science algorithm paying particular attention to ageing technologies, gene therapy and space travel". The underlying algorithms that are either optimistic or ragey, or are good for feeling -toned or factual things are made by professionals, but specific contents are chosen by users

great post, thanks for sharing 🎇 Imo there are 2 types of data worth differentiating as of rn. 1 - transaction related data (available on the #blockchain) -> who posted/commented/... when. 2 - content (available on #storage - #ipfs, #arweave, #ceramic, ...) -> what someone posted/commented. Imo we will see logic to order / filter the feed based on 1 pretty soon. And that might cover 95% of your requests. The - to me ;-) - exciting part is 2. My guess is that high quality content based #recommendations require some data aggregation layer for the content to train AI/ML on top of. Or enable simple content scoring. ---> Would love to hear other people's opinion on that. <--- I think a "workaround" good enough to exclude at least posts containing keywords like "FTX" might be available pretty soon as well.

Yup, 100%. I've been wanting to create a timeline showing you posts that have received the most positive sentiment, that's what I did this analysis for: mirror.xyz/rickydata.eth/krF2safQf84KevrCq_S6f3lV2oUuge2PKVnDvIPcw6M

I definitely like the idea of highly customizable timelines, I created a little app as an experiment, but found it tough to gather 100% of the content. You can't really have an indexer and API endpoint which lets you say "don't show me FTX-related content" because the content is stored elsewhere, so you'd need to collect that data, which can be slow.