Such a weird timewarp feeling around this stuff. It's like you could go dig up an archived Twitter dev post from 2008, find and replace and post. What a time to be alive (again)
This could be interesting.
Twitter effectively doesn't have an API any more (what they do have is prohibitively expensive for tinkering).
Mastodon has a culture that actively pushes back against many of the things people might want to build - any experiments with things like search or discovery improvements or even potentially algorithmic feeds tend to get a very hostile reaction from long-time Mastodon users.
Bluesky's API is wide open for innovation right now - and you can even connect directly to a WebSocket stream (from JavaScript running on any page) for direct firehose access, like in this demo https://firehose3d.theo.io/ discussed previously: https://news.ycombinator.com/item?id=42159786
Here's my own quick firehose demo: https://tools.simonwillison.net/bluesky-firehose - code here: https://gist.github.com/simonw/af2e50446f6dfc5cb514a7d6aadbb...
I've always felt like the API enabled us to make Twitter a ~success, back before it had critical mass.
I built a small literary journal on early Twitter, and knowing we had a simple way to archive what we were publishing and do stuff like bio-driven contributors notes was a big part of why it made sense to bother investing creative energy on a new platform and social capital convincing writers to make accounts.
It's been frustrating to watch the past few years. I haven't fully decided if I have the energy to reinvent it for a new platform, but poking around at atproto stuff earlier this fall made me more hopeful about the prospect than a similar draft I have on top of gotosocial (even though thw former ia much further from complete).
I find that I want the opposite of the firehose: I want the thoughtful and smart people, and I want it to be about their interests.
I did sign up for bluesky, but it has the same failure as twitter: I don't want to hear about your politics just because I follow you.
I wish there was a way to force users to use 1 hashtag per post and then I could unfollow hashtags - but I suspect it would be playing wack-a-mole.
Really anything that aims to replace twitter should be forced to not optimize for engagement.
Gemini 1.5 Flash 8B is so cheap now that you could build your own little custom Bluesky reader that filters out political tweets for you based on LLM classification. It would probably cost less than a dollar a month.
Following a hash tag is basically what reddit is
> I wish there was a way to force users to use 1 hashtag per post and then I could unfollow hashtags - but I suspect it would be playing wack-a-mole.
I built something like this in a slightly different context for myself.
I doubt it'll ever really catch on widely, but I have a bunch of people who want to read my highlights about tech and nothing else who appreciate my software dev feed.[1]
A friend of mine built something[1] that should solve this problem, you subscribe to a tag of a person (or multiple) but not their full feed.
You can still add multiple tags per post, say if it's about music and technology, and if you subscribe to their technology tag, you'll see the post.
You can mute keywords whether they're hashtags or not.
I tried doing that with Twitter a few years ago in an attempt to reduce the amount of posts related to US politics. I don't follow any political figures or commentators, only tech people and artists, but about half the posts on my feed were about US politics.
Although I ended up with a blocklist of around 100 words it was very much a losing battle and a waste of time.
Not only do bluesky users(skeeters? did that catch on?) not tag posts correctly they do that "anti censorship" thing where they share a screenshot of a post so you are guaranteed to see things you don't care about.
There's a culture of encouraging alt text on images, and muting keywords works on the alt text.
yeah no regular user is ever gonna bother using alt texts and bluesky will not implement AI auto generated alt texts so i doubt this works at all
> Mastodon has a culture that actively pushes back against many of the things people might want to build - any experiments with things like search or discovery improvements or even potentially algorithmic feeds tend to get a very hostile reaction from long-time Mastodon users.
I noticed the same thing about AI in the Nix community. Was really surprised by the extreme negative reaction, as I could see so many applications of AI that could be useful in the Nix ecosystem.
Projects should just bite the bullet and say "AI friendly" so those conservative anti tech users can pre-emptively be pushed away. Might hurt adoption but better than holding back progress. One of the reasons i don't use bluesky much is coz they came out swinging anti AI
Suggested projects for the AT Protocol from the Bluesky devs.
How do you crawl or backfill the existing data?
Like if you wanted to implement a search engine or recommendation system it’s not clear to me how you retrieve a training set.