Mastodawn

I love Mastodon, but sometimes I miss algorithmic timelines. Not because I want to be fed slop, but because federation makes it genuinely hard to discover people outside your bubble.

So I built a tiny script that scrapes the public timeline once a day and hands me 10 posts it thinks I'll care about. Static keyword matching (Opinions about AI, trans stuff, ADHD, Go, cooking) + a pass through an LLM to filter out engagement bait and generic motivational garbage.

Now I get a little digest every morning. No auto-likes, no bot spam, just: "hey, here's some stuff you might actually want to read."

AI for content generation in social media? Hard pass. AI to help me find content I wouldn't have seen otherwise? That's the good stuff

@owlex That's an interesting idea! Even though I've grown pretty skeptical towards LLMs, this seems like a decent application of them. Is it open-source? Are you using a local model?

Alexandra 6d ago

@sigmasternchen Not using a local model, but an openweight one. Wanted to look into Gemma-4 for local execution but did not have the time just yet (or the proper hardware).

Right now it's tied into my automation service, which hosts lots of small automations like a birthday reminder bot, which creates a small greeting to copy and paste with an LLM(yeah, shame on me)

Could open source the code if you're interested

@owlex Yeah, I get it. And with $companies paying up all the GPU stock, it'll probably be a while until I can properly test this stuff out locally. I mean, I can run an 8B model on my MacBook, but it's just not good enough for most applications...

Alexandra 6d ago

@sigmasternchen Yeah, but the trend seems to go towards local inference. In a few months it probably will be more and more viable

I played(vibed a mvp tbh) around with the Local OpenWhisper Models. They are really good even with my cheap 16GB M1 MacBook. But proper LLMs are sooooo slow on this hardware...

This is why I am really hyped about gemma-4. Google has released Google AI Edge Library for iOS and the model is sooo fast and even quite good

@owlex
This is pretty much what I tried to build and failed

Alexandra 6d ago

@curiousicae I feel you. It was not easy. Especially the static prefiltering drove me nuts

@owlex
Well, my main problem was that even with the smallest (1.3b) LLM it took at least a minute per invocation on the 10 year old laptop “gaming” GPU we have (and it would have required several). I didn’t want to feed the cloud of course (it’s not a good puppy ). And then it turned out those smallest models where a bit too stupid when it comes to any form of summation (unless you pass it, like, exactly one post but then what’s the point?).

Maybe just training a Basian scorer on my fav’d posts and dumping the LLM altogether would work? Didn’t even think of that...

Not quite what I had originally set out to do, but probably actually worthwhile.

(My original goal had been something like a summation of major events when I’m inevitably gone for a couple of days, but doesn’t seem like something that is possible with the hardware.)

Alexandra 6d ago

@curiousicae For a summary, I’d do the following. Take all posts in a given time frame. Make a bag of words. Filter out irrelevant words (I think there are already some OSS tools for this). Then get the post with the highest engagement for the“interesting” keywords.

If you want a summation, you could then cluster the posts of keywords and put them into the LLM

@owlex
Sounds a lot like SciKit-Learn. It’s been too long since that university course, but that line of thinking makes a lot of sense!

Not sure when I’ll get to this, but sounds interesting, thanks!

Alexandra 6d ago

@curiousicae when you do. Then tell me or even when you have troubles. I love this kind of stuff

Li ~ Crystal System 6d ago

@owlex something something .. need consent from people on local timelines to put their posts through this

Alexandra 6d ago

@Li even if I am using providers with a no tracking and no Training policy? Is the LLM the problem or the general automatic processing of posts? Just asking where the boundaries should be

Li ~ Crystal System 6d ago

@owlex i for one do not want my posts being fed to an LLM at all, and the reasons i am opposed to them extend far beyond merely 'tracking' .. i also don't want to be used for any 'algorithmic' determination, about them, .. ?

Alexandra 6d ago

@Li can you please explain me why? If you’d be willing to? I am trying to understand your position

Li ~ Crystal System 6d ago

@owlex i do not approve of llms in any context .. for many reasons (https://silica.codes/ramblings/ai.htm for a few..) .. therefore i do not want what in saying given to them in any way? i also dont like social algorithms because its basically interact with who you see etc, and gives wheoever controls the algorithm a massive amount of cotntil over whomever uses it, which isnt too relevqnt here since its you using it for yourself second reason is simply* that i do not consent and thats honestly its a computer being used in place of actually meeting people and such which i dont want to be involved in ? honestly tbh its also just things were saying posting eyc being used in ways were not okay with and tbh thats enough reason

Artificial Inteligence & Why "No AI Webring"

Alexandra 6d ago

@Li Thanks for taking the time and answering me.

I kinda understand your arguments but I came to a different conclusion for myself.
We have different views on this topic, which is okay. But I do not think any further discussion would be productive. From my perspective you'd only be satisfied if I'd forsake any kind of LLM use. Which I won't.
So lets stop that here.

I wish you much success on your further projects

Li ~ Crystal System 5d ago

@owlex i want you to specifically not use random peoples fedi posts in an LLM to do recommendations, because said people may not want that and you have no possible way of knowing actually .. .. but anyways ..

⠀ ⠀ ⠀ ⠀ stg 6d ago

@Li @owlex this is kind of a silly take. would it be okay if it was just a bunch of if statements with keyword matching? what makes a neural network any different?

Li ~ Crystal System 6d ago

@stag @owlex "a bunch of if statements" - you have just described every program in existence ever ...

.. doing based on keyword matching is a completely different thing and would be different because it is a different thing-

⠀ ⠀ ⠀ ⠀ stg 6d ago

@Li @owlex ok? what makes a statistical model so different? its just some vectors no?

edit: fixed typo

Li ~ Crystal System 6d ago

@stag @owlex its not searching for and not displaying based on keywords.

⠀ ⠀ ⠀ ⠀ stg 6d ago

@Li @owlex i mean, it kind of is in a way though. and even if you dont believe that, so what if its not purely based on a simple substring filter? what gives?

Li ~ Crystal System 6d ago

@stag @owlex the technical implementation details are not the problem it is a social issue .. again 'a bunch of if statements' describes nearly every program ever

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex

you're moving the goalposts

you first said that consent must be given. then i said that if it was just a keyword match, you would have no issue, and you said that wasn't fair, because of the technical differences.

and now you're saying "the technical implementation details are not the problem"

also, separate point: by publishing your writing publicly (in whatever form) online, you implicitly agree that people will download and use that. this does not harm the authors, it's not like theres a model being trained on the posts.

Li ~ Crystal System 5d ago

@stag @owlex you said "what makes a bunch of if cases any different" .. you asserted that the difference was technical not me, i always asserted that it was a consent/social issue from the start, and said "its no different" was dumb because the reasoning applied to litterally any algorithm ever .. blocking based on keywords is a litteral function in mastodon and is about trying to prevent yourself being exposed to triggering content this serves a different purpose and is something i come here explicitly to avoid being done

and no posting online is not consent to throw my shit into an LLM thanks, stop telling me the terms of my own posts and what is and isnt harmful to me -- thanks.

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex

i find it very hard to agree with your take

if you publish online it obviously doesnt mean consent for every use of the work. and if a model was being trained on, this would be fair.

but this is not exploitive, or really harmful to the authors in any way at all. it is a personal reading aid, one could say. when you post publicly, you consent to being read, but you do not have the right to how a reader reads your words.

i presume you will not object to a screen reader, or a translator, or even just skimming past a post. if i can do that, why not a model? assuming the model is local, of course. a model is just another tool, and you above stated that it was not a technical issue. what makes the model different as a "social issue"?

do you believe that people should be forced to seek consent for running private tools to filter the content they receive?

Li ~ Crystal System 5d ago

@stag @owlex when the "tools" are not tools and are just LLM bullshit trying to create an algorithmic feed, yes you should, by this logic why even have follow requests or blocks or whatever the fuck else which also let you decide what way your stuff is used (with follow requests you can even still read it but cant "subscribe" to it) i am fundamentally opposed to LLMs in all usecases and algorithmic feeds using them while intentionally disregarding that and going "nah dont need your approval to use you for this" is harmful your practically going "my body your choice" to me .. fuck off ..

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex

ok... is this ragebait. this is getting absurd.

why even have follow requests or blocks or whatever the fuck elseyou want the privacy of a completely private account, but you choose to publish on a public one. if you really hate it that much, make all your posts private! this is saying you hung up a painting in a museum and you're complaining that people are taking pictures of it.practically going "my body your choice"when you publish something publicly, that text (which is not a human body) is independent of you. denying a reader to the right to privately manage data that you sent to their computer is violating their autonomy, not yours. you're the one saying "my body your choice"

and i presume you have a hate for ai in general, by the way you write. that's fine, i can't control that. but your own distaste does not mean you get the right to govern someone else's computer. a reader's right to organize their own data outweights your desire to micromanage them.

you seem to believe that once you publish something online publicly, you should still have full control over anything that anyone does to it.

if you don't like this, it's simple: don't post publicly

Li ~ Crystal System 5d ago

@stag @owlex follow requests and blocks allow a public user to restrict access to their posts even though their all "public" clearly using this violates their autonomy and god given right to see and use my personal information however the fuck you want /j

my posts are inly public in the context of the platform i share it on, which is a platform that doesnt do algorithmic feeds and ai bullshit if i wanted to do that id go on twitter. its not the same as taking photos of an art peice thats also a completely different thing,

1- my posts are not art
2- its not in a museum
3- i am not physically restraining anyone, i am telling you to please not do this (mistakingly) thinking any of you care about consent in any meaningful way (more reason to not like ai guess, :shrug:)

my posts are more personal and not detached from me/us as a person, so its more like asking not to take photos of (me) or not to record what im saying or .. then insisting i should just not go outside or speak ever if i dont like that

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex

first, ftr i am not an aibro, i do not heavily use ai, only for mudane coding tasks

clearly using this violates their autonomyi'm not sure which part the /j applies to, or the entire thing. i'll just ignore the /j, because i don't know which part to respond to.

i did not say follow reqs and blocks violate their autonomy. my point is that sending your data to their computer, then trying to retain complete control over it violates their autonomy. if you block someone, you don't send your data, so it does NOT violate their autonomy.my posts are inly public in the context of the platform i share it on, which is a platform that doesnt do algorithmic feeds and ai bullshit if i wanted to do that id go on twitterhave you misunderstood how the fediverse works? it runs open protocols specifically so users can choose exactly how they want to see things. while most clients dont have algorithmic feeds, i could very well make a client with one right now. and one that uses ai to do so.

the entire point is "bring your own client." if you say are trying to say no one on the fedi can use algorithmic feeds, you're building a centralized walled-garden! exactly what the platform is trying to avoid. and apparently what you're doing to, though you seem to be inconsistent on that.my posts are more personal and not detached from me/us as a person, so its more like asking not to take photos of (me) or not to record what im saying or .. then insisting i should just not go outside or speak ever if i dont like that> maniaintevengotagifforthisone.png

you posting publicly is like announcing something to a crowd in a park. the model is not like a camera taking a picture of you, but it is like a pair of ear plugs.

again, (and i noticed you didn't respond to this point) do you really think that once you post something, you should still have complete control over what anyone does anything to it? once you post it, that text is independent of you.then insisting i should just not go outside or speak ever if i dont like thatso you're saying i'm violating your autonomy. again, i make the same point. you're violating mine. you send data to my device, but also feel the right to dictate what i use to read and process that data? seems a bit a authoritarian, no? i dont care how you feel about your posts, how personal they are, you chose to send it to my device and as my right as the owner of the device, i can do a lot of things (within reason).

to put it simply: do you think that when you send an email, you should be able to control what client the person receiving it uses and what folder they put it in? if no, then i'm glad you agree with me. if yes, then you might be a lost cause

Li ~ Crystal System 5d ago

> you posting publicly is like announcing something to a crowd in a park.

no it isnt.

> do you think you should control what email client someone does, i mean it litterally lets you do that though can not send to certain user agent on web, or domain such as @gmail.com or google MX records .. for example .. and on fedi you can not allow certain clients? i could block all akkoma instances from interacting if i wanted, atleast if you tun the instance anyway.

but it isnt a fedi "client" though its not intended to present the posts to you doing something unwanted with them, >_>

Li ~ Crystal System 5d ago

@stag @owlex which leads to; i would not "send you my posts" if i knew youd do LLM bullshit with them, and definitely not if you act like every single public post ever is free to use for litterally anything you ever could because its "on your computer" or whatever the fuck.. you sound like a advertising firm trying to justify scanning everyones messages for certain "interests" i mean its "theirs to do that with" right how dare you take issue with it .. its like the exact same shit ..

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li

i do notice you seem to be (purposefully?) ignoring many of my counterpoints

no it isnt. we can argue nuance but when you post something online, you really are announcing it to people. thats how it works.i mean it litterally lets you do that though can not send to certain user agent on web, or domain such as @[email protected] or google MX records .. for example .. and on fedi you can not allow certain clients? i could block all akkoma instances from interacting if i wanted, atleast if you tun the instance anyway. yes you can do that. for example, i could very well block all chrome users if i wanted to do that. but beacuse the web is an open protocol, just like fedi. it would be stupid, but i could.

however, because it is an open protocol, i could simply just pretend to be a different client.

you can choose to block certain users, but at the end of the day, the content is public and it is an open protocol, so you really cannot control what client people usebut it isnt a fedi "client" though i have no idea what you mean by thisits not intended to present the posts to you doing something unwanted with them, >_>you still believe that you should have complete control over what people do on their own computers?i would not "send you my posts" if i knew youd do LLM bullshit with themi am not running any model on your posts ftr. however, if i did, because you are publicly publishing, then you are in practice sending it to everyone who requests ityou act like every single public post ever is free to use for litterally anything you ever couldi have said before that there are limitations, for example i do not believe i should train a commercial model, impersonate you, etc.

but running a personal aid tool, which is of zero harm to you, is fine. by what right do you have to control what i do on my own device?sound like a advertising firm trying to justify scanning everyones messages for certain "interests" [...] .. its like the exact same shit ..what you are saying is completely different. i presume you are talking about scanning in messaging apps like whatsapp. your posts on the fedi are in no way private. whatsapp texts are supposed to be private.

if i am publicly releasing things, i do not have any right to object if someone wants to scan my content, because i put it out there. if i didn't want it, i wouldnt release it. that is how the internet works. you cannot act like you want an open internet and then procede to say that you should get to decide what people do.

Li ~ Crystal System 5d ago

@stag my posts are not shouting publicly for anyone to see, it is you insisting that is the case so you can avoid asking permission to force shit into AI/LLMs because you know they would not give it,

"it does not hurt you" yes it does though? the fact im making such a fucking huge deal about it is because im fucking hurt by it? do you think id do this if it was nothing? you just dont care about the harm its causing.

⠀ ⠀ ⠀ ⠀ stg 5d ago

my posts are not shouting publicly for anyone to seeokay, so the only logical negation of this statement is that your posts are private, for no one to see.

well clearly they're not. as a matter of fact, they're not even followers only. the facts speak for themselves here.yes it does though?okay, so apart from this emotional crash out, how exactly does someone else, on their own computer, completely locally, running a model, hurt you? i really don't see how that works.

Li ~ Crystal System 5d ago

@stag there is not actually a hard binary of "public for litterally anyone to use for litterally anything ever" and "private for no one" .. actually, why the fuck do you insist these are the only two options ???

Li ~ Crystal System 5d ago

@stag @owlex also if i have followreqs on there is nothing technically stopping you just occasionally fetching my posts and putting them into the feed in the first place, and if i block you nothing is stopping you technically from fetching unauthenticated or via an alternate account, clearly doing either of these is completley fine and definitely who cares about block evasion just dont post anything ever if you have a problem with that totally a reasonable response to that .. since otherwise id be telling you what to do w my posts and how dare i??!

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex never claimed block evasion was a right thing, but its an open protocol, people are perfectly within their right to do that. it may not be moral, but they have the right to.

because if they do not, then it is no longer an open protocol. my issue with you is that you claim people do not have the right to do what they want to run on their own devices. i don't have an issue if you think that's just not a good thing, but the fact that you think people shouldn't even have the ability in the first place.

Li ~ Crystal System 5d ago

@stag @owlex ? what the fuck is the difference between "having the right too" and "not being moral" exactly? you dont have a right to do that? rights are like your ability to exist, not be physically harmed, not be controlled, etc- "you shouldnt have the ability in the first place" .. ??? i said you shouldnt do it, of course you physically are able too, obviously?? theres no way to solve that technically?

⠀ ⠀ ⠀ ⠀ stg 5d ago

difference between "having the right too" and "not being moral"believing "X should have the right to Y" means that i believe that X can physically be able to do Y, and do so without any repercussions (apart from purely social)

believing that "X doing Y is immoral" means that i do not support X doing Y, and that if X does Y they should face social repercussions

morality and ability are pretty simple concepts really.i said you shouldnt do it, of course you physically are able too, obviously?? theres no way to solve that technically?if it is possible or not on technical level is not relevant. you believe that, if it was possible and nothing else was affected, people would just have no way of running a model over your posts. am i wrong?

Li ~ Crystal System 5d ago

@stag @owlex "people should not have a way of running an LLM model over your posts" i dont think i ever once said this, i said they should not do so, it isnt possible to stop them, and i would not support any means of forcefully trying to prevent this, but i do not support doing it, what do you think me complaining about it is for ?? changing the laws of physics or some shit? i am not capable of doing that,

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex

you seem to misunderstand.

let's say you are given a button, where if you press it, people will no longer have the physical ability to run a model over your posts. it works exactly the way you think it does, and it does not impact anything else, just this specifically.

would you press it?

Alexandra 5d ago

@stag they won’t change their opinion. You’re wasting you’re energy. In my opinion they only see black and white there are no nuances. They could block non followers from processing their requests but they post publicly on the internet. Thanks for discussing this topic

⠀ ⠀ ⠀ ⠀ stg 5d ago

@owlex yeah honestly lol its giving me "never argue with stupid people..." vibes im giving up atp

Alexandra 5d ago

@stag i gave up. I even tried to do it in a really polite way. They didn’t accept me having a different opinion

Li ~ Crystal System 5d ago

@owlex @stag "having a different different opinion"

no its "i am throwing your posts into an LLM without consent. and no i will not stop"

that is not "polite" nor is it a "different opinion"

fuck you ^^

Alexandra 5d ago

@Li @stag then please also tell me which apps I am allowed to use to view your posts. I was polite. You did not actually ask me what I am really doing with my automation. Thank you for this unproductive interaction

Li ~ Crystal System 5d ago

@owlex @stag but you are not "viewing" them you are "processing" them, which is the actual issue. >_>

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex really, please tell me what is the difference between them? is there a point in this distinction?

if my viewer can hide posts with certain words, is that processing? if my viewer can translate posts, is that processing? if my viewer can generate a tts audio, is that processing?

Li ~ Crystal System 5d ago

@stag @owlex ? the translation is so you can read it, the tts audio is so you can read it, or well hear it, your displaying the same thing just as audio instead of as text, translation is maybe the only real issue here since it could say something different to what you actually wrote .. or said. but this isnt that -- stop trying to correlate these..

⠀ ⠀ ⠀ ⠀ stg 5d ago

@Li @owlex and so what if i put it through a tool that tells me if it is relevant to me? if it works by non-neural based methods, is that okay? again, you claim its not a technical issue but it sure seems like thats not the case

Li ~ Crystal System 5d ago

@stag @owlex then it is not "reading the posts" it is doing something else with them, that they may not be okay with being used, in the same fucking way the CIA scanning all of the internet for anything that says "bomb" isnt actually fine or an advertiser doing it for "computer" to then show you ads for computers .. fucking

Mattie → Margo

@owlex Nice! How do I block it

Alexandra 5d ago

@matt_rethyu is there a standard yet, which I can adhere to skip profiles like yours? I see you have hashtags in you bio. Will build in something, that posts from Profiles like yours are skipped

Alexandra 5d ago

@matt_rethyu but generally you can’t block these kind of processing except when you turn your profile private. Me respecting your hashtags doesn’t make the big labs skip processing you content

Mattie → Margo

@owlex the #nobot and #nosearch seem to be common ones used by many people, the rest are less common

Alexandra 5d ago

@matt_rethyu I also discovered that you can exclude your posts from public search in your profile. That way they won’t show up in the api I am using. Which is also probably used by everyone else

Alexandra 5d ago

@matt_rethyu Okay further research shows. That this is not doing anything. Mastodon makes it hard to follow you wishes easy. Integrated the hashtags as a filter in my process though

Li ~ Crystal System 4d ago

@matt_rethyu @owlex why was this so fucking hard ..