twitter, for all the hate it gets, is surprisingly useful — if used well. in particular, in the ML research community, it ends up being an important channel through which early results are disseminated, and through which researchers share their half-baked thoughts. (note that many other non-ML fields also have a strong twitter presence.)
unfortunately, the default state of the twitter algorithm is to flood you with slop: cat photos and political rants. this leads to a less than ideal research-browsing experience. to get anything else, you have to explicitly teach twitter’s algorithm that you like other things. you can either do this over months of natural usage, or by force in five minutes. here’s how to do the latter.
seeding your twitter feed
- make a new twitter account.
- when asked who / what to follow, choose as few people / categories as possible, and ideally people / categories far away from politics or culture.
- once you’re done with setup, unfollow everyone Twitter has forced you to follow by default.
- you may be able to skip “being forced to follow people by default” by closing the window just after you create your account and then reopening it.
- build your graph. follow 25-50 people whose takes you like (or like to dislike) — and people they follow that look interesting.
- sometimes Twitter will give you the option to ‘follow similar people’. this can be useful, but check the list of suggestions with care before blindly accepting them.
- make sure the people you follow don’t post too much politics. (even if their research tweets are interesting, their political tweets will drag the rest of the political memeplex into your timeline.)
- prune uninteresting tweets. go through your feed (the ‘For You’ section), and select ‘Not interested in this’ (click the three dots at the top right of the tweet) for all tweets that aren’t related to ML research.
- do this for at least five minutes, on the clock! (time it! it will feel a lot longer.)
- like interesting tweets. go through your ‘Following’ section, and ‘like’ a bunch of ML tweets that you find interesting.
- as before, do this for at least five minutes, on the clock!
- now refresh your ‘For You’ section. the twitter algorithm should now have learned your tastes.
caring for your twitter feed
over time, you may find your feed reverting to slop. this is (unfortunately) normal. some fixes include:
- using your ‘Following’ feed as opposed to your ‘For you’ feed.
- the ‘Following’ feed contains only tweets from people you follow. this stops the twitter algorithm feeding you slop — but also stops it feeding you interesting tweets from people you don’t follow. so use this wisely.
- mark posts you don’t like with ‘Not interested in this’.
- go to ‘Settings’ → ‘Privacy and safety’ → ‘Content you see’ → ‘Interests’, and turn off all interests you don’t want in your feed (h/t @jankulveit for this suggestion!)
- you will have to do this regularly, but it mostly fixes the problem.