Conversation

Notices

1. Embed this notice
   ⚡Lord of Misrule⚡ (toiletpaper@shitposter.world)'s status on Thursday, 16-Jul-2026 19:07:28 JST ⚡Lord of Misrule⚡
   "But we cease to feel struggle when we use AI."
   
   Sounds like this person has little experience using AI.
   
   I get the TLDR; (ie. the tool has negative externalities). But the counter is, the cat's out of the bag. No amount of public pressure is likely to affect that trajectory very much, especially when it comes to how nation-states and the military have already been using it against the general public since long before LLMs became widely available to the public. Nor will it affect the vast majority of sheeple and bandwagoneers, who frankly don't give a fuck, and can't get much more retarded than they already are.
   
   Boycotting it entirely might make one feel morally superior, but I don't think it will meaningfully affect how bad actors leverage the technology, mainly because you have almost no control over how your taxes are spent. The best we can do is be careful about who we pay for the service. You get what you pay for (taxes notwithstanding) and there's no need to throw gasoline on the forest fire for the sake of personal convenience. In other words, choose ethical providers and local models whenever possible. But for the same token we can also use AI to enable us to fight back against the bad actors. In effect like Tai Chi, to redirect their own destructive energy against them. Think USA's 2A, swords and military munitions are meant for killing people, but is it really better for us when only the enemy has them at their disposal? Clearly not.
   
   https://www.frank.computer/blog/2025/05/just-a-tool.html
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 16-Jul-2026 19:07:24 JST iced depresso

     in reply to

     • 🌈ᚩ🌈
     @toiletpaper @bonifartius its much less if you're fine tuning an existing one (unsloth does a lot of work about this.)
     
     the astounding part was always that training just a token predictor was able to make something that could speak and approximate reason at all. i still remember reading one of the earliest papers about this when someone curiously typed in "The purpose of life is _" and the robot deadass finished it as "to have sex."
     
     i keep being bothered by how much of ML models come back to the same discrete token + mixture model pattern, though.
   • Embed this notice
     ⚡Lord of Misrule⚡ (toiletpaper@shitposter.world)'s status on Thursday, 16-Jul-2026 19:07:25 JST ⚡Lord of Misrule⚡

     in reply to

     • 🌈ᚩ🌈
     @bonifartius
     
     > llm produces semi random combinations
     
     yeah. this is something a lot of people have a hard time wrapping their head around. it's just a fancy autocomplete, which in the most basic sense can be implemented as a spreadsheet with a minimal algorithm as the front-end. nothing "intelligent" about it. but that also doesn't negate it's usefulness. however quasi-deterministic it may be, it's still a startlingly convincing simulacrum. one thing that I think of often in this context is the story of the golem of prague, except the modern incarnation isn't likely to come with an off switch (מֵת).
     
     > if they were trained locally.
     
     would be nice. I've created toy models this way purely for educational purposes, but to do anything particularly useful it seems like I'd need at least a couple $100K or an order of magnitude more, and dedicated access to a data centre and team of other engineers. so, I rent instead, and use RAG with strong guardrails for domain specific knowledge. it mostly does the job I want it to, albeit with a lot of back and forth and hand holding to avoid it producing a pile of incoherent slop.
   • Embed this notice
     🌈ᚩ🌈 (bonifartius@noauthority.social)'s status on Thursday, 16-Jul-2026 19:07:26 JST 🌈ᚩ🌈

     in reply to

     @toiletpaper hannah arendt lets libtards heads explode 💁

     everything changes one, hence i have to decide if i want to use something and thus be changed. fedi is ok for me, i can talk to interesting people here. X etc. just fry my brain.

     i can't really place my finger on it, but LLMs feel functionally different to me. a plough is a plough, a lathe is a lathe, a compiler is a compiler. a llm produces semi random combinations of things. fwiw, i'd be leas critical if they were trained locally.

   • Embed this notice
     🌈ᚩ🌈 (bonifartius@noauthority.social)'s status on Thursday, 16-Jul-2026 19:07:27 JST 🌈ᚩ🌈

     in reply to

     @toiletpaper i mainly don't use it because of the heidegger argument. it would change me, and the slope is downward - stopping it's usage would be much harder for me than starting it.

     if i can't do things on my own, i have to reflect if i really need to do them. if i'm too lazy to do the menial tasks, is the whole important enough for me? is it really MY work if 90% is done by some automaton? (isn't the compiler the very same?)

     not easy and of course different for everyone.

   • Embed this notice
     ⚡Lord of Misrule⚡ (toiletpaper@shitposter.world)'s status on Thursday, 16-Jul-2026 19:07:27 JST ⚡Lord of Misrule⚡

     in reply to

     • 🌈ᚩ🌈
     @bonifartius
     
     > the heidegger argument. it would change me
     
     I get it. personally I'm not opposed to changing. in the words of heraclitus, you can never step in the same river twice. but fwiw, I think citing heidegger is also a great argument that refusing to use AI is inherently fascist (libtard logic). 🤪
     
     > is it really MY work if 90% is done by some automaton
     
     you could make the same argument about whether it's your work because you used a plough and a combine to plant & harvest crops instead of digging the furrows with a stick and harvesting with a sharp rock. at the end of the day "MY work" is just a question of ego, whereas the products of the labour is what's really the point in the first place. and given that AI is trained on a large chunk of human knowledge, how is it different from planting an acre by yourself, vs planting that same acre with the help of your family or community? there's no case where we're not leveraging work done by others, whether directly or in the form of tools we didn't invent and manufacture from scratch, and thus we're always necessarily sharing the credit for the end result whether we acknowledge that explicitly or not.
     
     that being said, I don't fault or judge you for your choice. just being the devil's advocate for the sake of argument.
   • Embed this notice
     🌈ᚩ🌈 (bonifartius@noauthority.social)'s status on Thursday, 16-Jul-2026 19:07:28 JST 🌈ᚩ🌈

     in reply to

     @toiletpaper
     > Boycotting it entirely might make one feel morally superior, but I don't think it will meaningfully affect how bad actors leverage the technology

     well, you only can truly change what you do so that's the only thing that can be changed anyway. feeling superior isn't good though. i have good reasons not to touch LLMs, it's for my own sake, not because i can save the world with it. when others choose to not use LLMs, i will support them.

   • Embed this notice
     ⚡Lord of Misrule⚡ (toiletpaper@shitposter.world)'s status on Thursday, 16-Jul-2026 19:07:28 JST ⚡Lord of Misrule⚡

     in reply to

     • 🌈ᚩ🌈
     @bonifartius
     
     I get chided a lot for not having a cell phone, not using the internet without a VPN, not having traditional social media, not wanting to be digitally recorded every time I step into the public domain, not using ecommerce if it means compromising my privacy, etc. So I get the urge to boycott invasive/destructive technologies in a visceral sense. That said, I think AI is a bit different in this respect. I've been able to do a lot with it that would have simply remained a perpetual pipe dream if it was just me left to my own devices. I don't really do less programming or research or other stuff I would normally do because I use AI. In effect I do a lot more of those things, but cut out the menial tasks and things which were previously out of reach due to time constraints, skills gaps, disinterest, etc. That includes creating software that enables me to have better privacy, including from AI systems. That said, I don't use just any AI/provider. I choose carefully and implement safeguards, because it heavily intersects with my privacy and related ethical concerns. I don't see it as a black and white decision to use or not use it, but in shades of gray, and I'm strongly motivated to use it to fight back. I even use it to write my local politicians when they're weighing in on policy decisions which have an impact in these areas.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 16-Jul-2026 19:43:24 JST iced depresso

     in reply to

     • 🌈ᚩ🌈
     @toiletpaper @bonifartius smaller models perform better with narrow scopes. the people getting by with 9B models and prayers do an awful lot of prompt engineering, MCPs and loras.
     
     some of which is basically making up datasets that actively consult the MCPs, which is where tricks like letting models write and run scripts instead of chaining tool turns come in.
     
     but, yes, if even that is still beyond you then :comfyshrug: it is me too.
   • Embed this notice
     ⚡Lord of Misrule⚡ (toiletpaper@shitposter.world)'s status on Thursday, 16-Jul-2026 19:43:25 JST ⚡Lord of Misrule⚡

     in reply to

     • iced depresso
     • 🌈ᚩ🌈
     @icedquinn @bonifartius
     
     > fine tuning an existing one
     
     this is something I've not tried yet, mainly because RAG does a good enough job for my use cases so far. also depending on the model size there're costs involved that are currently not practical for me.
     
     > "to have sex."
     
     pretty close. more to the point is, "to reproduce". the entirety of the meaning of life can really be distilled down to "copy" and "mix" (kopimism?), which are embodied in the form of sexual reproduction. everything else is a means to that end, even if the journey is more rewarding to an individual than the destination.
   • Embed this notice
     ⚡Lord of Misrule⚡ (toiletpaper@shitposter.world)'s status on Thursday, 16-Jul-2026 20:18:10 JST ⚡Lord of Misrule⚡

     in reply to

     • iced depresso
     • 🌈ᚩ🌈
     @icedquinn @bonifartius
     
     I want to try this, even if only for personal interest, but too much on my plate and not enough resources to spare yet. but if I manage to think up a way to make it pay for itself I'll definitely do it.
     
     right now my main focus is an LLM based tool that helps people with navigating a specific set of legal paperwork and procedures. RAG is good enough to start off with, but if the error rate ultimately proves to be too high I might need to go with fine-tuning. but it adds to the complexity (eg. choosing the best base model), cost (training, hosting, etc) and time involved before the product can ship, so I'm hoping it's not necessary. TBD.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 16-Jul-2026 20:18:10 JST iced depresso

     in reply to

     • 🌈ᚩ🌈
     @toiletpaper @bonifartius sounds like a product that doesn't have much business going beyond hybrid search then. :gutkato_nervuma:
     
     I've been watching google gemini answer questions about factory town 2 confidently wrong for the past day and its been making shit up about various games for a couple weeks now.
     
     where the LLM earns anything in a high risk environment is extracting semantic webs, maybe to help with complex indexing, but law and medical aren't great places to be confidently wrong.
   • Embed this notice
     ⚡Lord of Misrule⚡ (toiletpaper@shitposter.world)'s status on Thursday, 16-Jul-2026 20:36:10 JST ⚡Lord of Misrule⚡

     in reply to

     • iced depresso
     • 🌈ᚩ🌈
     @icedquinn @bonifartius
     
     Yeah. Hence not having much resources to spare, since after guardrails are fully implemented, there has to be tens of hours of legal review (never mind the surrounding non-AI based infra). I'm optimistic about the revenue potential, but it'll be a ways off before it's operating in the black. Given some recent international case law, simply having a disclaimer won't be enough. Realistically there's no way to guarantee 100% accuracy even in a best case scenario, but it will be providing exact citations, and as long as can demonstrate due-diligence with everything else, that's the best possible.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 16-Jul-2026 20:36:10 JST iced depresso

     in reply to

     • 🌈ᚩ🌈
     @toiletpaper @bonifartius well. i wouldn't let it generate user facing answers in a high risk setting. :ablobcatcoffee:
     
     the takedown of zuck being behind outsourcing a bunch of age laws shows its useful for investigation passes, but letting it actually answer stuff is a shortcut to "wikipedia is not a source but the judge used it to rule on a case" town.