So, on the topic of AI - if we were to assume that AI would follow approximately the same trend as the internet did - a massive early bubble, followed by gradually becoming more important and more and more of a Big Deal - what parallels could we make? If we assume that it will move faster, is there any way we could measure our progress along that path, and estimate it's speed? :/
Eh, you could say the same about a lot of early industrial machinery, too. "My textile mill just got a new mechanical loom, and it's terrible! Does bad work, cost way too much, breaks down all the time, mulched 5 children this month! Absolute trash!" And I imagine there were a few people who did say such things, and were quite right about the facts on the ground where they worked - but the industrial revolution happened anyway. :/
I do think silicon valley has jumped the gun on this by a fair bit, and executed the whole thing clumsily and carelessly - it's the dotcom bubble all over again - but dotcom bubble or no, the internet was still a big deal, and I think this will be too, for better or worse. :/
but the dotcom bubble was capitalising on the tech a lot of us were using back then, this AI slop is just turbo snapshot video enhance and "write me an essay" it's comic sans not a new paradigm imo.
Most people were not using the internet at the time of the dotcom bubble, I'm pretty sure? I was born 1994, so I didn't exactly get a good view of it, but that's my impression.
As for AI, I do use local AI (Devstral, mostly), and while it's definitely an early prototype that still needs a lot of work done, it is useful none the less.
@Angle🖇 I disagree with @Nick , AI is a new paradigm, but that nobody wants for good reason. Even you only use it locally.
One of the reasons Internet has been successful is that people have seen the advantage of putting some personal time and effort into it, either for clout, money or for the greater good. Reddit, Instagram, YouTube, none of these (and many more) wouldn’t be what they were 5 years ago without people actively wanting to contribute to it.
GenAI is the total opposite, it takes from existing human creations without consent and outputs plausible nonsense at scale. The adoption model consists in subscriptions not enough people will want to pay for as is, so it has be crammed in everything which then justifies a price hike. However people already are either tired of the normalized output or having a psychotic break at the moment, so it won’t be sustainable either.
Uber eventually made a fortune destroying taxi cabs because it was successful at providing a convenience to regular people even as it was man-in-the-middling taxis. But GenAI only provides convenience to very few people while man-in-the-middling the entirety of human arts which is absolutely ludicrous.
Local AI has a lot less of the associated issues... And I'm not sure they won't get away with 'man-in-the-middling the entirety of human arts', ludicrous or not. I'm sure there are lots of things about our current society that would seem ludicrous to someone thirty years ago, yet, here we are... -_-
@Angle🖇 @Nick I don't think that's a great argument for the potential staying power of AI. A number of ludicrous inventions simply didn't make it. I'll go back to asbestos and radium which I feel have a lifecycle I expect GenAI to follow. Asbestos was deemed revolutionary at the end of the 19th century and used in most new buildings until 1970 and it's been a pain to extract ever since. Radium was also deemed revolutionary before we quickly found out it wasn't good to be exposed so much. And both asbestos and radium have niche and safe uses, the problem was that both were pushed in most things.
However, it wasn't a concerted effort by a handful of massive multinational corporations like GenAI is so there's some uncertainty about how long the AI push will last for even as the popular support for the tech dropped below 50%.
I mean, just because it shouldn't necessarily be pushed into most things doesn't mean it won't be. And I still think you're underestimating the technology. Yes, the current versions are buggy and underdeveloped, and a lot of people are doing stupid things with them, but that's normal for a new technology. There is still a lot of potential there, and it will be realized eventually - the main question is by who, and for what ends. :/
I'll admit that I don;t know what to do about it... But acknowledging that the problem exists, telling other people about it, and maybe building a consensus that it needs to be addressed would be a good first step... T_T
@Angle🖇 My complete and utter reject of this technology isn't born out of underestimating it. Even now in its "infancy" stage, it has been shown to wreck massive havoc on several intellectual disciplines for the benefit of ridiculously few people. I don't need more evidence from more mature applications.
You're talking about the terminator, but I'm just talking about LLMs/GenAI which is an outsized application of machine learning. And like I said, the benefits are felt by extremely few people while the downsides are felt globally. I expect this imbalance to be its undoing, no matter the validity of the few niche uses that do not only benefit mindbogglingly rich people.
To go back to asbestos, it doesn't matter if your building is cheaply fireproof if the people get cancer just by existing within its confines.
I will also note, it's not enough to just 'reject' the technology - you have to actually do something about it. Otherwise you're just going to find yourself getting shoved into a smaller and smaller corner. :/
Like, I really want to see people actually learning about the technology and how it works and what it's good for - but at a bare minimum, learning what other people are doing with it and how you can prevent them from fucking you over seems important...
You'll only get shoved into smaller and smaller corners if we, collectively, allow that. Gen AI is destroying industries and lives and does so entirely through theft for the benefit of a tiny, tiny minority of people.
"Just ignore it" is a viable tactic in some circumstances. Blockchain and NFT guys promised just as much revolution as the current AI bubble (cont)
but just like AI, those fads required enough people to sign on. And it didn't happen, and it faded.
AI isn't inevitable, unless we make it so. It's already starting to crumble as profits fail to appear, as lawsuits start catching up. The people pushing for it will find something else to try, undoubtedly.
And personally I think "trying to find the good uses" is just giving it legitimacy. Any good use will be corrupted by capital.
AI has a way, way lower bar, though. Like, you can generally tell if you're getting caught up in blockchain nonsense, whereas telling if a company is using AI is a lot harder. Yeah, a lot of it is really shit right now, but you shouldn't assume it's going to stay that way forever.
As for good uses, the first requirement is to cut capital out of the equation as much as possible. Ideally, you want AI developed by the people, for the people.
I mean, Ideally, I want no genAI at all. I don't see a use case that doesn't just waste time and resources.
What is the point of producing lots of text? What social purpose does that perform? IDK if its built "by the people" and "for the people" meaningless but grammatically correct text is just chaff.
And yeah Capital will keep trying I don't doubt that. They do it with child slavery too. Anything to make a profit.
And like child slavery you can't be like "well there might be a good use case!" (I'm not saying you're in any way in support of CS, just using an example). And when laws wont stop it it's up to us to deny it a part in our society.
I don't think individual rejection actually accomplishes that much, though.
Like, I use AI to help me write Minecraft mods - it's far from perfect, but it's still a 10-20% productivity boost, even in the current early prototype form that I have access too. And if I weren't telling people about, nobody would know. Even with me telling people about it, I think it's extremely unlikely that I face significant pressure over it - and that's just me! I'm a small player here. :/
@KolaMagpie @hypolite The big companies have so many more resources to cover up and deflect their AI bullshit, the same way they do all their other bullshit. I just don;t think your rejection means anything. :/
@hypolite I get that. But the same argument could be used and is used to justify all sorts of bullshit.
Individual action doesn't mean anything, so why bother. One person shopping Walmart doesn't make the community store fail. But everyone believing that, does make it fail.
One person trying to win rights never works, so why try at all?
I'd personally be okay with 20% fewer minecraft mods, in exchange for a healthy society.
@KolaMagpie @hypolite But that's not actually the deal on offer. Like, I'd look at it like guns in a warzone. Should everyone have them? Probably not. But *some* people absolutely should, and a blanket 'guns are bad' when the Russians are coming over the horizon won't help anyone. -_-
@hypolite But this isn't a warzone, not like that at least. In war, if you don't have a weapon, you're a victim. In society, we are only victims /if we let ourselves be/ (please don't take this out of context).
NFT bros wanted us to be victims, but we rejected that because that future sucks. We /do/ have the power to do that. Collectively. And enough individual people were like "fuck that" that it wasn't the future. The same can be true of AI.
@KolaMagpie @hypolite I mean, the average person is answering a definitive 'yes' to the first too. So right there, we're off to a bad start. All my points about the futility of individual rejection apply to those things too. :/
@hypolite Sure, in which case, we're fucked. Cuz Capital will /always/ be able to out-spend, out-propoganda, out-/anything/ us. Capital are experts at making us complacent.
I'd posit that the lack of education and leftist outreach during the rise of those industries lead to their current monopolies. And I'd rather try for a better future than give up and give more power to the weapons that oppress us.
@KolaMagpie @hypolite I'm not entirely sold? It's not just a matter of spending, propaganda, organization, etc, but the combination of those with other factors - but that still requires that we put forth the spending, organization, agitprop, and appropriate other factors to make things work. :/
@KolaMagpie @hypolite Like, vegans and vegetarians and their assorted fellow travelers have been rejecting meat for quite a while, but they've barely made a dent in the meat industry. If you want to make headway on this, then it takes more than individual rejection, you have to find a way to make it appealing to the average person, and moralizing just won't cut it. :/
@Angle🖇 @KolaMagpie I don't know about that, in most cities restaurants will have at least a well thought-out vegetarian option, which is probably more than you could have expected 20-25 years ago. There are thriving vegan restaurants, which is a good sign that there's some significant adoption. Speaking of dent, meat production has decreased significantly in Europe between 2020 and 2023, and stagnated in the US. Only in Asia did it rebound from the pre-pandemic level. The dent is there, see ourworldindata.org/grapher/glo…
@hypolite @KolaMagpie Eh, that's great for the vegetarians, but most places are showing roughly a 100% increase in meat production since 1990. I wouldn't call that a victory for vegetarianism. -_-
@Angle🖇 @KolaMagpie The important isn't the absolute number, but the comparison. Meat production has been growing way slower in developed countries, while it still grows in developing countries. For me this means that at a certain level of development, meat has become way less important as a status symbol.
@hypolite @KolaMagpie As a status symbol, sure. But status aside, meat is tasty and it's easier to have a healthy diet with it than without it. And until there's a decent substitute, I don't expect consumption to drop much.
@Angle🖇 @KolaMagpie No, you're right, which means even a stagnation in the production means for me that vegetarian/eco-conscious lifestyles are gaining ground as there isn't much of a reason for this production to slow down otherwise.
@hypolite To circle back to my original point: the people rejecting AI aren't your enemy nor are they the antithesis for keeping society safe from the AI crisis.
Just like vegans aren't the enemy when fighting wasteful land use policies.
I stand by that AI, currently, isn't inevitable so rejection might be by itself valid. It might need more active work than NFTs did, but it's possible it can be nipped in the bud.
@KolaMagpie @hypolite It's not the rejection that bothers me, per se, but rather the insistence that it's not capable and that rejection is sufficient. Like, a vegan is fine, but a vegan who insists 'Meat is bad for you actually, the meat eaters will all die out any day now' is much less so - especially if they have lots of company and like, you actually want to cut down on meat consumption. -_-
@Angle🖇 @KolaMagpie I believe we disagree on capabilities. I believe GenAI is capable of mesmerizing an entire generations of humans into believing a human-shaped parrot serves them using a language they are familiar with.
On the other hand I don't think GenAI is capable or will ever be capable of writing or drawing masterpieces, generating production-ready code for complex real-world applications, or make any progress in research and development. I believe its widespread adoption would lead to an overall stagnation of the artistic, scientific and technologic production, while decreasing humans' very ability to learn and make stuff ourselves.
The capability that is currently being developed is fooling the training staff that it's reaching its arbitrary metrics goals. Machine Learning has always been able to do that, and the introduction of natural language only reinforces this because as humans, we strongly associate language with intelligence.
But even this is showing its limits. To "increase productivity", training staff is now running the output of a model by another GenAI model to check for "quality". Predictably, the trained model has started optimizing for this very task and outputs very well-formed nonsense that the other model is applauding as the pinnacle of literacy: futurism.com/gpt-5-literary-ou…
@KolaMagpie @hypolite Ehhhh, it's closer to that then I'd like. Like, if a company uses AI to automate your job or whatever, what are you going to do about it? If they use AI to automate someone else's job, how will you even know, let alone do something about it? :/
@hypolite Tell people about it! Be a society about it. Target's imploding cuz people stopped wanting to go there because they made anti-societal decisions.
@KolaMagpie @hypolite ...Did Target actually do something noticeably worse than other companies? I mean, yeah, communicating and rejecting the companies that seem especially bad is something, but I still think we need something more organized and larger scale... :/
More organized and larger scale's great, but that doesn't happen without individual action or awareness. The union wars didn't happen (and the subsequent workers' rights) because people were all "darn, the company's forcing me to live in their house but there's nothing I can do about it" or "darn wages suck but what can I do about it?" or even "darn they're gonna throw bombs at me so why bother trying?"
@KolaMagpie @hypolite And sure, current AI is far from automating jobs wholesale - but that just makes detection even harder. Like, a company uses AI and hires 20% fewer people then they would have otherwise... How do you know? What do you do? If the majority of companies do that, how do you possibly address things? :/
@Angle🖇 @KolaMagpie I write software for a living, and I would never use GenAI at my job for a few reasons: 1. I like to write code. 2. I don't like to read code someone else wrote. 3. I'm not confident I would be able to measure productivity change with GenAI. Studies show that people who believe they are more productive with AI actually aren't. Because that's what GenAI is good at and optimized for, giving the change about its actual limited capabilities.
@Angle🖇 @KolaMagpie oh, and I forgot: 4. I don't know what code it's been trained on, if their authors are aware and consenting to their work being used this way.
@hypolite Like, if some guy has a fucking gun, and looks like he's bad news, and I go 'gee that looks like it could be trouble, maybe we should do something about it, or prepare or whatever', and your response is to cross your arms and stick your nose in the air and go 'I don't like guns', well I'm sorry but... :/
@hypolite All around me I see people rapidly developing more powerful and more capable AI, while the people I would expect to turn to for help are sticking their heads in the sand and going "LALALA AI can't do anything it's not important nothing to worry about LALALA!", and I just want to fucking scream. T_T
I've been struggling with this. like, what's "more powerful" mean? I play with local models, which is probably ethically equivalent to eating meat. there's that new swiss one you have to have a huggingface account and a newer GPU to use, so even though it's fully open-source, it's still questionable. I compared qwen3-coder and qwen2.5-coder and qwen2.5-coder seems better at x86 assembly, but, I'm not a good judge, and
the kind of "powerful" I see is more like, "given access to more systems", AI weapons and shit, not "better at creating the illusion of thought" (or thinking, or whatever they do. transformer models are useful, demucs is fun, beat_this is unbeatable for beat detection, image generation seems... idk, more capable? but maybe rejecting it /is/ doing something about it, and I'm more a part of the problem than I'll admit.
@amsomniac Eh, there are some things where it could be worth rejecting it? But I'd look at it like guns in a warzone. Should everyone have them? Probably not. But *some* people absolutely should, and a blanket 'guns are bad' when the Russians are coming over the horizon won't help anyone. -_-
@Angle🖇 @upbeat estimation software Again, I have to reject the analogy. GenAI is (only?) good at making people believe it is competent, regardless of their actual competency. What kind of people do you think wants to wield such "weapon"? What kind of people do you think are the prime targets?
Using local models isn't fighting back against megacorps (I assume the Russians in your military analogy), it's potentially a softer self-subjugation with a much smaller energy footprint. Which is still a plus in my book.
@hypolite @amsomniac ...I'm sorry, but you're just fucking delusional on this. You're noticing a handful of obvious failures and insisting that's all there is, because that's what you want to believe. Here's a video examining the effects:
TLDR: AI absolutely can be useful if you use it right, and it already *is* transforming our economies in big ways. We will need to figure out how to grapple with it, and ignoring it won't get you anywhere.
Profitez des vidéos et de la musique que vous aimez, mettez en ligne des contenus originaux, et partagez-les avec vos amis, vos proches et le monde entier.
@hypolite @amsomniac If you don't want to use it, fine, but you will still need a better strategy than just turning your back and sticking your nose in the air.
@Angle🖇 @upbeat estimation software Nope, I did my due diligence, I learned how the tech work, what it is capable of and mostly not capable of, I've been impressed with some of the output (mainly the generated video) but I don't want any part of it.
There's no valid strategy for me that involves actually using the stuff in any form. Just like drugs, you don't have to try them to know how they work and how harmful they can be. I don't judge too harshly people who casually use chat bots or GenAI because they are designed to be enticing and I know there are valid niche use cases, but I won't get near it myself.
I make sure to make my point known when the topic comes up, and that's pretty much all I can do. Like I said, it might reduce my opportunities, but I prefer the idea of finding my people in the corner rather than submitting to this humankind-sucking trend. Either it will come to pass and things will keep on for me practically unchanged, or it will reign and I don't want to have any part of it.
@hypolite @amsomniac ...I mean, I took my dexmethylphenidate this morning. 5 MG, extended release. Plus 100 MG caffiene. Helps me focus and get moving. :/
So, yeah, drugs can be useful, actually, if used correctly. Lots of finicky details, of course, and a lot of people probably shouldn't try and use them, but they're far from useless. A good parallel to AI, honestly. XD
@Angle🖇 @upbeat estimation software Still not. Medicine (which I separate from drugs even if there's a single word for both 😖) has to undergo extensive clinical trials before being put on the market which AI currently isn't bound to.
Would you use a new bootleg drug its manufacturer says it will cure all your ills if only you trust them, although they didn't go through any serious trials? This is what LLMs feel to me at the moment. I don't doubt it has useful applications, LLMs existed before chat bots and was pretty useful as a first pass for translation for example. I've happily used DeepL for years, for example.
But I don't want to have to talk to machines, it just isn't the most efficient way of interacting with them and it tricks people into awkward situations where they trust the machine based on the presentation of the output rather than the output accuracy itself.
@hypolite @amsomniac Drugs are just medicine used wrong. I could walk outside and start selling the same pills I use, straight from the pill bottle, and bam, I'd be a drug dealer. Not something I want, but it's a thin line. :/
Of course, for some chemicals, there an argument to be made that there is no good use for them - but that doesn't mean we can dismiss all chemicals. -_-
And I'm kinda iffy about talking to machines too? I use them in Cline, a Codium plugin, mainly, where I can pretty much just give them a few sentences, shove some files at them, and have them figure out the rest. Summarizing error logs or code, suggesting what might be the cause of problems I encounter, that sort of thing.
@hypolite @amsomniac As for the manufacturers, yeah you should absolutely be picky about what chemicals you accept, and from who. But jumping from there to "Chemistry bad! No chemicals!" is just too much of a leap. :/
@hypolite @amsomniac Oh, I would be interested in hearing about the equivalent of clinical trials for AI? I mean, there's an argument that that treatment has made medicine way worse and less accessible, which I think is broadly true... But if it protects us from bad medicine, maybe it's worth it? I dunno. I am would generally lean towards a bit less regulation for medicine... But not too strongly... :/
@hypolite Less options, for one thing. I'd love to get my dexmethyphenidate in 1 MG increments, but with the limited suppliers, there's little reason for them to diversify and offer me that option.
@hypolite Also less competition means less pressure to drive down prices. Lots more competition to sell caffeine tablets, cause they're less regulated, so I can buy a giant jar of those on Amazon for $10 and have them delivered to my door, whereas the meds cost way more, for way less, and I have to pick them up from a pharmacy.
@Angle🖇 Ok, yes, I see what you mean.I believe it is the cost of drug safety, although if pharmaceutical research was a public service and patents were free, there would be better availability without sacrificing safety through deregulation. Drat, I criticized capitalism again.
@hypolite @amsomniac Like, the early industrial revolution analogy really works, IMO. You're in the position of a worker seeing the new machines and going 'These things are terrible! They do bad work, they cost too much, they're trying to take my job, I saw one grind a small child into paste right in front of me! I hate them!"
It's a fair viewpoint. From the perspective of the worker, those things are all true - his mistake is in thinking that they're universal and will never change.
@hypolite @amsomniac I'll admit I'm not entirely sure where to proceed - if you find yourself advising that early industrial worker, what do you tell him? He's not in a good position... I do think the truth is a necessary starting point, though. "Sorry dude, but the machines will eventually do good work, or good enough work at least, and they will be economically viable. They will eventually come to form the backbone of the economy. They will reshape your world, whether you like it or not..."
@hypolite @amsomniac Probably not what he wants to hear. What can he do about it? I'm not sure. If I found myself dropped back at the dawn of the industrial revolution, I'd probably try and organize some kind of alternative to capitalism? Some kind of decentralized, grass roots economic system, with the power as close to the people as possible... Though I'm not sure it *is* possible. Maybe power is always destined to centralize, and the best you can hope for is a center that you can live with?
@hypolite @amsomniac But standing at the dawn of the industrial revolution and thinking you can just skip out on the industrial machinery... Yeah, I'm sorry, but no. There's no way around it, the only way out is through. -_-
@Angle🖇 @upbeat estimation software I understand the appeal of the Industrial Revolution analogy, but I'm weary of it. On one hand, we know the exact way it went down for industrial machines. They became more precise and safer to use for workers, and then they got relocated to developing countries to save even more on labor. But they were things that needed to be engineered and built by people who built the skills to make them more precise and accurate with safety features.
On the GenAI/LLM front however, we're building global black boxes that aren't engineered for accuracy but for human likeness. No industrial machine engineer was ever convinced their design was their friend and deserved personhood rights.
Since they are global, we all have a personal choice whether to use them or not, unlike industrial machinery that had an expensive upfront cost for business owners only.
Since they are black boxes, we have no reliably way of increasing the precision of the models or measuring the accuracy of their output, and that isn't even an optimization goal for the three main players in the space.
And then for safety, using local models sounds better but they are purposefully mesmerizing machines made to exploit human instincts, so it is only marginally safer.
Criticizing the systemic harm while not letting the few niche valid use cases weaken the larger point.
Shedding the assumption it's there to stay. Rejecting the false marketing promises in favor of real, current harm.
The tech already is unpopular after just a few years of mass deployment, so the only thing currently sustaining it is the massive sunk costs by the handful of companies who poured everything they had into an intangible moon shot.
@hypolite @nicksalt I think you should be picky about what AI you use, and what for... But it *is* a powerful tool, if you use it right. So, skipping out on it entirely is like finding yourself in a warzone and refusing to touch any guns, or worse, insisting that nobody else does so. Nooot necessarily a good decision.
@Angle🖇 I understand what you are trying to get at with the warzone analogy, but I don't think it reaches the mark. Mainly because the current crop of AI directly exploits human weaknesses: our tendency to look for humanity anywhere and our longing for convenience in all things. This means the only people wielding it as a weapon are the megacorps. Individuals running homegrown models aren't "fighting back", they just contribute slop at a much smaller scale.
This also means to me that the only way to defeat the megacorps is to reject their premise entirely. No, GenAI isn't an inexorable future, no, I do not want any of its output in any of my workflows, and yes, I prefer humans writing text, drawing images or coding slower and more deliberately. For me, running local models is already giving them the benefit of the doubt, which is more than I want to do.
To take your analogy literally, not everybody in a warzone takes up arms. You still need plenty of civilian support, and that's what I feel closer to than direct combat if it even meant something in the GenAI paradigm, which I don't think it does.
But for me my main policy would rather be to avoid companies using AI at all. I know this limits my opportunities and I'm not impatient for my possible layoff which will throw me into the IT job search meat grinder, but I'd still favor direct approaches at tech event and mixers rather than looking to automate job search using AI. I'm hoping that I would find my people by doing the legwork and standing firm on my principles, but I'm not eager to see them tested in real life.
Here are some palm oil free makeup, cosmetics and perfume brands that do not use rainforest-destroying palm oil. If you are ever in doubt look for the prefixes: LAUR, STEAR, GYLC and PALM in the in…
Vertice della coalizione a Parigi, Macron assicura: “Appoggio Usa alla sicurezza”. Ma la telefonata con la Casa Bianca è tesa. Ok di Londra ai missili a lungo …
I've been to a lot of restaurants with veggie burritos that had lard in the beans or chicken in the rice. It's nice of them to mention it (but maybe, they shouldn't name it for a lifestyle it isn't appropriate for?)
Qui studenti israeliani e palestinesi, dai 3 ai 18 anni, siedono fianco a fianco e seguono le stesse lezioni nelle loro due lingue: "Qui tutto si basa sul…
One day you wake up and learn that blues thrives in the Sahara, Turkish funk is a full-blown genre, and hard rock is practically a science in Eastern India.
Parla lo storico russo-britannico: "Quello nato a Pechino non è un asse, né un modello per un mondo post-occidentale. Dietro allo show di unità si nascond…
Il presidente israeliano ribadisce l’impegno alla difesa della comunità cristiana. Il pontefice: “I due Stati sono l’unica soluzione”. L’Idf: “Preso il 40% di …
BREAKING: New York Governor Kathy Hochul is set to sign an executive order allowing pharmacists to prescribe and administer Covid-19 vaccines to anyone who requests them, regardless of age or underlying conditions.
4 likes, 0 comments - matthewfacciani on September 4, 2025: "Excited to share these new slides breaking down the 5 R's from my book Misguided!
A huge thank you to Hui Li, a recent media studies master’s graduate, and participant in my online course …
- DeepMind received its first major investment by Peter Thiel (introduced by Eliezer), and Jaan Tallinn later invested for a 1% stake. One founder, Mustafa Suleyman, got fired for abusively controlling employees, but Demis Hassabis kept heading the company. DeepMind lost so much money that it had to be acquired by Google. - Distrusting DeepMind (as now directed by Demis under Google), Sam Altman and Elon Musk founded the nonprofit OpenAI. Sam and Elon fought for the CEO position, and Sam gained control. Holden made a grant to this nonprofit, which subsequently acted illegally as a for-profit under investments by Microsoft. - Distrusting OpenAI (as now directed by Sam to appease Microsoft), Daniela and Dario left to found Anthropic. Then all the top billionaires in the safety community invested. Then Anthropic received $8 billion in investments from Amazon. We are dealing with a gnarly situation. (...) One take on this is a brutal realist stance: That’s just how business gets done. They convince us to part with our time and money and drop us when we’re no longer needed, they gather their loyal lackeys and climb to the top, and then they just keep playing this game of extraction until they’ve won.
It is true that’s how business gets done. But I don’t think any of us here are just in it for the business. Safety researchers went to work at Anthropic because they care. I wouldn’t want us to tune out our values – but it’s important to discern where Anthropic’s leaders are losing integrity with the values we shared."
I will say, while plenty of neolib shitheads have remained neolib shitheads, I have been very pleased to watch as more and more people who had been center-left now swinging far left.
Like Dem Party, there's a reason you're sitting at like 10% approval. You've alienated literally everyone. Your base, the moderates, the center left, NOBODY likes your wishy washy do fuck all but send "sternly worded letters" bullshit in the face of the biggest extitential crisis the union has seen since 1861.
oh my god the gung ho vote blue nomatter who wait not like that bullshit. Protip dems you keep losing to Trump and the republicans because nobody in what was your base will ever vote for a republican in everything but name and why would a republican vote for Diet Republican when they can have full flavor Republican max. You dont need republicans to win but you cant win without leftists...
I mean yes, I admit freely that come November of any given election year, I show up and vote for whichever candidate who has a chance in fuck to win that will do less harm than any other candidate who has a chance to win. But that's the absolute maximum I will give these sniveling whining spineless fucks who can't seem to even do the goddamn minimum thing they could possibly do: Show up every day Do whatever procedural bullshit you can to slow shit down. Demand an actual roll call vote for EVERYTHING. Ensure your whole caucus votes no on every single thing that would advance Trump's agenda in the slightest no matter what.
If they can't even show up and vote no to fascist fucks when that's literally their jobs, they ain't gonna get a dime from me or a single afternoon stuffing envelopes.
@Scoll And alternative doesn't have to specifically be doing violence. Direct action civil disobedience works wonders to force people to pay attention and pick sides, even without actually doing violence against people.
I am not a hardcore Bolshevik, to be clear. I agree with a decent amount of what Lenin said, but I reject the Leninist vanguard party, "democratic" centralism, and the associated one party state that arises from that.
I am also not entirely opposed to anarchism (ancoms and anarcho-syndicalists especially are alright, and I could get on board with them if they actually gained momentum)
I also partially disagree with the Leninist answer to the national question, because it has been used to justify reactionary nationalism on the part of people like Stalin and Jackson Hinkle.
@burnoutqueen Are you going to purge us anarchists after the revolution? I gotta be honest, that's a worry with a lot of Leninists; some of them are very clearly looking forward to the purges as the main draw of the revolution.
There's a truly frustrating amount of people that call themselves leftists, but in actuality, are just authoritarians, but they want THEIR authoritarian.
I also hate imperialism so much. It really really really bothers me how so many leftists overlook or ignore the absolute horrific oppression the global south is enduring, in favor of maintaining a gradualist social democratic status quo.
And I feel like the body of work that expands on Lenin is one of the richest criticisms of imperialism and its nature as the most developed form of capitalist expropriation.
In a short-med future I still pursue soc-dem policies in elections and such, because the Grand Glorious Revolution is such a roll of the dice on what the fuck would come out of it, and tbh, our odds against the counterrevolutionary authoritarians is really really really damn bad. So while I hate making slow progress dragging the Republic leftwards, I don't trust the process of revolution.
@passenger I believe that in the meantime before there is a strong revolutionary movement, I support parliamentary engagement and public agitation/coalition building
@passenger I am not a Leninist. I support a democratic party of the working class, a broad leftist coalition that includes anarchists and other non Marxist market abolitionist socialists, and the implementation of a council democracy in the same vein as Rosa Luxemburg, council communists, and some anarchists (or what I sometimes call Soviet democracy)
Where I agree with Lenin most is his written perspective on imperialism, the state, the abolition of the police (which he blatantly didn't do), and the question of parliamentary engagement.
I even fixed my Pixel 6a back in December. Unfortunately, I broke the new screen a few weeks ago, and the microphone and speakers also started to fail, and the battery was showing its age.
I'm glad I upgraded to my Pixel 9a when I did. I love GrapheneOS and have been using it since early May 2023 with my 6a.
Pedophile maga went from wanting the death penalty for child rapists to being apologists for sending them to do their time in a minimum security club fed.
All for their hero. Convicted child sex trafficker Maxwell can hurt rapist Trump, so she gets what she wants. And none of them will say a word about it.
Republicans protecting child rapist. Release the #EpsteinFiles
Senator Bill Cassidy deftly grilled Health and Human Services Secretary Robert F. Kennedy Jr. in a Senate hearing Thursday, forcing Kennedy to contradict himself on the topic of vaccines.
love2d stavolta che gira, nonostante la octo-oriented programming!
Sorprendentemente, appena qualche ora di sonno e qualche ora di scrittura magica un pochino avanti e indietro più tardi, e ho effettivamente trovato una soluzione al problema problemoso delle prestazioni imbarazzanti di Love2D caricato di una tale OOP che non gira affatto bene su una viemmina come quella di Lua… e, anche se come previsto il modo che ho dovuto mettere in atto è abbastanza spaventoso, non è nemmeno inadatto alla produzione, e anzi: è gnammastico. 😳
L’obiettivo in mente era una roba del tipo: avere nel possibile una programmazione orientata ad oggetti che, per ridurre l’overhead causato da troppi lookup in tabelle e troppe chiamate di funzioni in poco tempo, fosse basata principalmente sulla composizione, desiderio che è anche comune in Lua… ma, volendo evitare Lua, perché voglio invece qualcosa di fortemente tipizzato, perché altrimenti so che finisce rapidamente tutto a spacc. In questo senso, Teal è interessante, però, per motivi che ora non frecano, non mi convince più di tanto… e allora ho ragionato su cosa si potesse fare con TypeScript… 😨
Ecco: sorprendentemente, sfruttando semplicemente gli oggetti anonimi (uguali a quelli di JavaScript, che si mappano perfettamente a tabelle di Lua) in congiunzione con il sistema di tipi composti di TypeScript (che funzionano come le interfacce nella OOP, ma indicano tipi di oggetti), evitando completamente le classi del linguaggio… con la proprietà intrinseca degli oggetti in JavaScript (e in Lua, duh, in qualunque linguaggio interpretato) di essere componibili, ma combinati coi tipi lì, si riesce ad avere a livello di sviluppo tutta la sicurezza dei tipi di TypeScript, ma in output codice Lua estremamente pulito!!! (E che, per inciso, evita completamente l’uso delle metatabelle, anch’esse causa di rallentamenti.) 🤯
Benchmark stavolta niente, poiché palle, e anche perché i “fottuti rettangoli” hanno mostrato prestazioni negative inaspettate rispetto alle 2 versioni scritte a mano ieri in Lua… ma non perché il codice sputato fuori da TypeScriptToLua in questo caso sia sporco, quanto più perché ho già iniziato a reimplementare con questo nuovo paradigma il mio motorino desiderato, che ovviamente dell’overhead in più lo ha comunque, ma… Stavolta, la demo di Breakout sul 3DS è magicamente giocabile, non va più a 5 secondi al frame!!! (E sul PC mi si aggira su 1-2% di CPU, che è wow.) 🗽
Ora… boh, solo le pareti che mi tengono compagnia quando programmo sapranno dirmi come andrà avanti questo affarino. A parte il fatto che ho dovuto già ripensare abbastanza la API da come l’avrei voluta inizialmente — dovendo farla deviare già parecchio da HaxeFlixel, perché non sembra esserci modo di avere i tipi completamente sicuri dovendo allo stesso tempo minimizzare gli oggetti nidificati e le catene di funzioni — ci sono alcuni dettagli per cui questa cosa degli oggetti pseudoclassisti funzionano che mi sanno di strano, perché praticamente devo tenere le definizioni all’effettivo completamente separate dalle implementazioni (quindi, per esempio, devo usare Cacca.new() per creare una nuova cacca, ma TCacca per riferirmi al tipo…), ma sarà un TypeScript skill issue. 😶
C’è anche da dire che con questo mio accrocco non c’è incapsulazione, implementarla sarebbe un casino e costerebbe (per via di come funziona Lua, che costringe ad usare funzioni anonime per implementare questa cosa; funzioni che verrebbero interamente copiate su ogni singolo oggetto) lo spreco di un fottio di memoria (termine tecnico)… ma non lo vedo come un problema; casomai dovesse servire il distinguere campi pubblici da privati, basterà rubare la convenzione di Python per cui le variabili che iniziano con gli underscore sono ad uso interno. E, davvero, l’unico aspetto negativo di questa macchinazione credo sia il fatto di non poter ottimizzare ulteriormente senza ridurre il riuso del codice, avendo svariate chiamate a funzioni miste che per quanto piccole sarebbero meglio inlinate, cioè copincollate dal compilatore anziché lo sviluppatore, se solo Lua lo permettesse… (…E se scrivessi un postprocessore Lua per fare esattamente ciò???)
l’amore 2d e gli oggetti orientifici, ma quello che accade non fa divertire (test e valutazioni prestazioni OOP su Love2D)
Visti gli imprevisti con HaxeFlixel che non ho ancora avuto il tempo di elaborare qui, stavo (ri)considerando il basato Love2D che, ultimamente mi sono (ri)accorta, gira su talmente tante piattaforme da rendere inutile anche fare degli esempi qui. La cosa seccante di quel coso, però, è che non è esattamente un motore di gioco, quanto più un framework multimediale… e quindi, a differenza di Flixel e altri robini, non ha tutte le varie utilità che è bene avere per poter sviluppare qualcosa senza partire dallo zero assoluto… e quindi, l’idea sarebbe di creare una specie di motorino per esso per gestire cose comuni come sprite, fisica, boh, queste cose (non) belle. 🤥
Ovviamente, il problema inevitabile è sorto immediatamente dopo una giornata di lavoro iniziale — a dire il vero, fatta per fortuna con le pinze, perché ero sotto sotto pronta a vedere cose storte accadere — e cioè che, con una dose di OOP in realtà nemmeno troppo grossa per gli standard comuni, le prestazioni sono crollate così tanto a picco che, per un semplice giochino di Breakout (la demo di HaxeFlixel, che ho adattato strada facendo per testare), da un lato su PC l’uso di CPU si aggirava attorno al 10-15% (che è tipo il quintuplo di cosa fa la stessa demo in HaxeFlixel)… mentre, su piattaforme pazzurde come il 3DS era letteralmente ingiocabile, facendo 5 secondi a frame (e pensare che io ho il new, che è più veloce). 💀
Ho fatto un po’ di ricerca e — per quanto fosse a me comunque ovvio che una programmazione ad oggetti basata principalmente sull’ereditarietà rende un programma più lento, perché i sassi elettronici sono fatti per eseguire istruzioni sequenziali e lavorare con memoria quanto più continua possibile, che è il contrario di cosa succede con tutte quelle classi che estendono classi che estendono il mercato che mio padre comprò — non immaginavo che su Lua il calo prestazionale fosse tale da essere non solo evidente, ma proprio fuori scala in certi casi. E ora, dunque, i problemi sono grossissimi. 😤
Anche stavolta ho raccolto molti link a riguardo di questa ennesima causa di sofferenza per me, e in realtà ancora non ho capito bene la questione, ma un grande problema sembra essere causato dagli accessi a tabelle nidificate, e alle chiamate di funzioni fatte più del necessario anche per operazioni altrimenti veloci… e nel mio caso certamente una buona parte di overhead in questo senso sarà causata dal fatto di non scrivere Lua nativo, bensì usare Haxe (o in alternativa, TypescriptToLua) per traspilare a Lua, ma sentivo che il problema non poteva essere solo il codice bloattato generato da questi affari… 🧨
E infatti, scrivendo in Lua puro un piccolo benchmark (battezzato al volo solo per dare un titolo al memo: “Love2D fucking rectangles“, genera innumerevoli rettangoli e li fa muovere calcolando le collisioni), prima in modo classico e poi con un minimo di OOP, ed eseguendolo oltre i limiti del ragionevole, ho visto le cose brutte: la versione OOP è in effetti più lenta. Non tanto più lenta, e comunque dipende dalle opzioni con cui la si fa girare, ma solo perché è comunque molto semplice… a differenza del motorino che tanto vorrei creare per replicare la API di HaxeFlixel in Love2D per quanto possibile (evidentemente, non molto possibile). 😭 Dopo ben 4 (quattro) immagini non so se ho voglia di elaborare oltre… Ma, in sostanza: in una modalità, il programma genera solo X (200mila) rettangoli all’avvio, mentre nell’altra ne genera X (200) a frame, andando all’infinito, calcolando sempre le collisioni… e quindi con la prima si esclude una lentezza dovuta alla continua istanziazione di oggetti, mentre la seconda da modo di vedere come un programma rallenta nel tempo rispetto all’altro (generando meno oggetti a parità di tempo). 💥
Nella prima modalità, il carico è basato principalmente sull’accesso alla chiamata draw, quindi non potevo limitare il numero di quadrati effettivamente visibili, e quindi ho potuto eseguirla solo su PC, dove si nota in media un rallentamento di circa il doppio per la versione OOP, che accede a svariate proprietà nidificate per fare il disegno… Mentre, nella seconda la prova il carico era più il resto, quindi ho deciso di limitare il numero di rettangoli visibili a schermo ad ogni frame agli ultimi X (500), e questo mi ha permesso di eseguire il programma pure sul 3DS senza che crashasse (credo ci siano limiti di VRAM lì), ma sia su PC che su 3DS si vede che la versione non-OOP riesce a generare in media 1,2 sprite in più per delta di tempo, differenza che nel corso di minuti diventa di migliaia di sprite. 😵
Incredibile, spassoso, magicante, ma… e adesso??? Boh! Dovrò ingegnarmi pesantemente per creare un motorino sufficientemente generalizzato da poter essere usato come comoda libreria per molti giochi Love2D, ma che allo stesso tempo sia efficiente… ma qui casca l’asino, perché per implementare concetti come uno sprite, che oltre ai classici dati come posizione X e Y ha un oggetto “disegnabile”, che può essere un’immagine o una forma geometrica, che quindi richiede chiamate della API Love2D completamente diverse dietro le quinte, non vedo alternativa non incasinata se non l’OOP; ma non basterà usare più la composizione che l’ereditarietà, bensì per sconfiggere l’overhead serviranno mosse di design interne talmente scomode che ho davvero tanta paura anche solo a pensare di scriverle… 😱
comunque, pensavo che probabilmente una versione di Love2D basata su una VM WebAssembly anziché Lua permetterebbe di spremere ancora più prestazioni (pur senza JIT, a quanto so WASM è più efficiente di Lua)
“A charlatan,” chaos, and lies: Anti-vaccine activist RFK Jr. faces Senate RFK Jr. accused senators of making things up as they made factual points on vaccines. arstechnica.com/health/2025/09…
Profitez des vidéos et de la musique que vous aimez, mettez en ligne des contenus originaux, et partagez-les avec vos amis, vos proches et le monde entier.
Loved this jargon from a recent #hifi show in Melbourne - the winning exhibitor's description of his system:
"front end is all denafrips: arcas streamer / terra clock / gaia ddc / terminator plus 15th dac and cables are by dacman with diamond optics: shi-tone ethernet, interconnects and speaker cables in addition lhy fmc and sw6 and uptone ether regen (total not including speaker cables is around $28k)"
Nick
in reply to Angle🖇 • • •Angle🖇
in reply to Nick • • •Angle🖇
in reply to Angle🖇 • • •Nick
in reply to Angle🖇 • • •Angle🖇
in reply to Nick • • •Most people were not using the internet at the time of the dotcom bubble, I'm pretty sure? I was born 1994, so I didn't exactly get a good view of it, but that's my impression.
As for AI, I do use local AI (Devstral, mostly), and while it's definitely an early prototype that still needs a lot of work done, it is useful none the less.
Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 I disagree with @Nick , AI is a new paradigm, but that nobody wants for good reason. Even you only use it locally.
One of the reasons Internet has been successful is that people have seen the advantage of putting some personal time and effort into it, either for clout, money or for the greater good. Reddit, Instagram, YouTube, none of these (and many more) wouldn’t be what they were 5 years ago without people actively wanting to contribute to it.
GenAI is the total opposite, it takes from existing human creations without consent and outputs plausible nonsense at scale. The adoption model consists in subscriptions not enough people will want to pay for as is, so it has be crammed in everything which then justifies a price hike. However people already are either tired of the normalized output or having a psychotic break at the moment, so it won’t be sustainable either.
Uber eventually made a fortune destroying taxi cabs because it was successful at providing a convenience to regular people even as it was man-in-the-middling taxis. But GenAI only provides convenience to very few people while man-in-the-middling the entirety of human arts which is absolutely ludicrous.
Angle🖇
in reply to Hypolite Petovan • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 @Nick I don't think that's a great argument for the potential staying power of AI. A number of ludicrous inventions simply didn't make it. I'll go back to asbestos and radium which I feel have a lifecycle I expect GenAI to follow. Asbestos was deemed revolutionary at the end of the 19th century and used in most new buildings until 1970 and it's been a pain to extract ever since. Radium was also deemed revolutionary before we quickly found out it wasn't good to be exposed so much. And both asbestos and radium have niche and safe uses, the problem was that both were pushed in most things.
However, it wasn't a concerted effort by a handful of massive multinational corporations like GenAI is so there's some uncertainty about how long the AI push will last for even as the popular support for the tech dropped below 50%.
Angle🖇
in reply to Hypolite Petovan • • •Angle🖇
in reply to Angle🖇 • • •Angle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 My complete and utter reject of this technology isn't born out of underestimating it. Even now in its "infancy" stage, it has been shown to wreck massive havoc on several intellectual disciplines for the benefit of ridiculously few people. I don't need more evidence from more mature applications.
You're talking about the terminator, but I'm just talking about LLMs/GenAI which is an outsized application of machine learning. And like I said, the benefits are felt by extremely few people while the downsides are felt globally. I expect this imbalance to be its undoing, no matter the validity of the few niche uses that do not only benefit mindbogglingly rich people.
To go back to asbestos, it doesn't matter if your building is cheaply fireproof if the people get cancer just by existing within its confines.
Angle🖇
in reply to Hypolite Petovan • • •I will also note, it's not enough to just 'reject' the technology - you have to actually do something about it. Otherwise you're just going to find yourself getting shoved into a smaller and smaller corner. :/
Like, I really want to see people actually learning about the technology and how it works and what it's good for - but at a bare minimum, learning what other people are doing with it and how you can prevent them from fucking you over seems important...
KolaMagpie
in reply to Angle🖇 • • •You'll only get shoved into smaller and smaller corners if we, collectively, allow that. Gen AI is destroying industries and lives and does so entirely through theft for the benefit of a tiny, tiny minority of people.
"Just ignore it" is a viable tactic in some circumstances. Blockchain and NFT guys promised just as much revolution as the current AI bubble (cont)
KolaMagpie
in reply to KolaMagpie • • •but just like AI, those fads required enough people to sign on. And it didn't happen, and it faded.
AI isn't inevitable, unless we make it so. It's already starting to crumble as profits fail to appear, as lawsuits start catching up. The people pushing for it will find something else to try, undoubtedly.
And personally I think "trying to find the good uses" is just giving it legitimacy. Any good use will be corrupted by capital.
Angle🖇
in reply to KolaMagpie • • •AI has a way, way lower bar, though. Like, you can generally tell if you're getting caught up in blockchain nonsense, whereas telling if a company is using AI is a lot harder. Yeah, a lot of it is really shit right now, but you shouldn't assume it's going to stay that way forever.
As for good uses, the first requirement is to cut capital out of the equation as much as possible. Ideally, you want AI developed by the people, for the people.
KolaMagpie
in reply to Angle🖇 • • •I mean, Ideally, I want no genAI at all. I don't see a use case that doesn't just waste time and resources.
What is the point of producing lots of text? What social purpose does that perform? IDK if its built "by the people" and "for the people" meaningless but grammatically correct text is just chaff.
And yeah Capital will keep trying I don't doubt that. They do it with child slavery too. Anything to make a profit.
KolaMagpie
in reply to KolaMagpie • • •And like child slavery you can't be like "well there might be a good use case!" (I'm not saying you're in any way in support of CS, just using an example). And when laws wont stop it it's up to us to deny it a part in our society.
And if we wont do that, then we get the bullshit.
Angle🖇
in reply to KolaMagpie • • •I don't think individual rejection actually accomplishes that much, though.
Like, I use AI to help me write Minecraft mods - it's far from perfect, but it's still a 10-20% productivity boost, even in the current early prototype form that I have access too. And if I weren't telling people about, nobody would know. Even with me telling people about it, I think it's extremely unlikely that I face significant pressure over it - and that's just me! I'm a small player here. :/
Angle🖇
in reply to Angle🖇 • • •KolaMagpie
in reply to Angle🖇 • • •@hypolite I get that. But the same argument could be used and is used to justify all sorts of bullshit.
Individual action doesn't mean anything, so why bother. One person shopping Walmart doesn't make the community store fail. But everyone believing that, does make it fail.
One person trying to win rights never works, so why try at all?
I'd personally be okay with 20% fewer minecraft mods, in exchange for a healthy society.
Angle🖇
in reply to KolaMagpie • • •KolaMagpie
in reply to Angle🖇 • • •@hypolite But this isn't a warzone, not like that at least. In war, if you don't have a weapon, you're a victim. In society, we are only victims /if we let ourselves be/ (please don't take this out of context).
NFT bros wanted us to be victims, but we rejected that because that future sucks. We /do/ have the power to do that. Collectively. And enough individual people were like "fuck that" that it wasn't the future. The same can be true of AI.
KolaMagpie
in reply to KolaMagpie • • •@hypolite But I think the biggest problem is getting that across. Getting people to think in terms of societal cost-benefit.
Sure, meat is delicious. Am I okay with eating meat, if it means wasting 54% of calories grown in the USA?
Sure, Walmart is cheap. Am I okay with having more things if it means my town and economy is destroyed?
Sure, AI is convenient. Am I okay with more websites and movies if it means the destruction of professionals?
Angle🖇
in reply to KolaMagpie • • •KolaMagpie
in reply to Angle🖇 • • •@hypolite Sure, in which case, we're fucked. Cuz Capital will /always/ be able to out-spend, out-propoganda, out-/anything/ us. Capital are experts at making us complacent.
I'd posit that the lack of education and leftist outreach during the rise of those industries lead to their current monopolies. And I'd rather try for a better future than give up and give more power to the weapons that oppress us.
Angle🖇
in reply to KolaMagpie • • •Angle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •Angle🖇
in reply to Hypolite Petovan • • •Hypolite Petovan
in reply to Angle🖇 • • •Angle🖇
in reply to Hypolite Petovan • • •Hypolite Petovan
in reply to Angle🖇 • • •KolaMagpie
in reply to Angle🖇 • • •@hypolite To circle back to my original point: the people rejecting AI aren't your enemy nor are they the antithesis for keeping society safe from the AI crisis.
Just like vegans aren't the enemy when fighting wasteful land use policies.
I stand by that AI, currently, isn't inevitable so rejection might be by itself valid. It might need more active work than NFTs did, but it's possible it can be nipped in the bud.
Angle🖇
in reply to KolaMagpie • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 @KolaMagpie I believe we disagree on capabilities. I believe GenAI is capable of mesmerizing an entire generations of humans into believing a human-shaped parrot serves them using a language they are familiar with.
On the other hand I don't think GenAI is capable or will ever be capable of writing or drawing masterpieces, generating production-ready code for complex real-world applications, or make any progress in research and development. I believe its widespread adoption would lead to an overall stagnation of the artistic, scientific and technologic production, while decreasing humans' very ability to learn and make stuff ourselves.
The capability that is currently being developed is fooling the training staff that it's reaching its arbitrary metrics goals. Machine Learning has always been able to do that, and the introduction of natural language only reinforces this because as humans, we strongly associate language with intelligence.
But even this is showing its limits. To "increase productivity", training staff is now running the output of a model by another GenAI model to check for "quality". Predictably, the trained model has started optimizing for this very task and outputs very well-formed nonsense that the other model is applauding as the pinnacle of literacy: futurism.com/gpt-5-literary-ou…
Angle🖇
in reply to KolaMagpie • • •KolaMagpie
in reply to Angle🖇 • • •@hypolite Tell people about it! Be a society about it. Target's imploding cuz people stopped wanting to go there because they made anti-societal decisions.
Unionize. Organize. We are not helpless.
Angle🖇
in reply to KolaMagpie • • •KolaMagpie
in reply to Angle🖇 • • •Hypolite Petovan likes this.
Angle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •1. I like to write code.
2. I don't like to read code someone else wrote.
3. I'm not confident I would be able to measure productivity change with GenAI. Studies show that people who believe they are more productive with AI actually aren't. Because that's what GenAI is good at and optimized for, giving the change about its actual limited capabilities.
Hypolite Petovan
in reply to Hypolite Petovan • • •4. I don't know what code it's been trained on, if their authors are aware and consenting to their work being used this way.
Angle🖇
in reply to Angle🖇 • • •Angle🖇
in reply to Angle🖇 • • •upbeat estimation software
in reply to Angle🖇 • • •upbeat estimation software
in reply to upbeat estimation software • • •Angle🖇
in reply to upbeat estimation software • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 @upbeat estimation software Again, I have to reject the analogy. GenAI is (only?) good at making people believe it is competent, regardless of their actual competency. What kind of people do you think wants to wield such "weapon"? What kind of people do you think are the prime targets?
Using local models isn't fighting back against megacorps (I assume the Russians in your military analogy), it's potentially a softer self-subjugation with a much smaller energy footprint. Which is still a plus in my book.
Angle🖇
in reply to Hypolite Petovan • • •@hypolite @amsomniac ...I'm sorry, but you're just fucking delusional on this. You're noticing a handful of obvious failures and insisting that's all there is, because that's what you want to believe. Here's a video examining the effects:
youtube.com/watch?v=yhpyHV1iz0…
TLDR: AI absolutely can be useful if you use it right, and it already *is* transforming our economies in big ways. We will need to figure out how to grapple with it, and ignoring it won't get you anywhere.
- YouTube
www.youtube.comAngle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 @upbeat estimation software Nope, I did my due diligence, I learned how the tech work, what it is capable of and mostly not capable of, I've been impressed with some of the output (mainly the generated video) but I don't want any part of it.
There's no valid strategy for me that involves actually using the stuff in any form. Just like drugs, you don't have to try them to know how they work and how harmful they can be. I don't judge too harshly people who casually use chat bots or GenAI because they are designed to be enticing and I know there are valid niche use cases, but I won't get near it myself.
I make sure to make my point known when the topic comes up, and that's pretty much all I can do. Like I said, it might reduce my opportunities, but I prefer the idea of finding my people in the corner rather than submitting to this humankind-sucking trend. Either it will come to pass and things will keep on for me practically unchanged, or it will reign and I don't want to have any part of it.
Angle🖇
in reply to Hypolite Petovan • • •@hypolite @amsomniac ...I mean, I took my dexmethylphenidate this morning. 5 MG, extended release. Plus 100 MG caffiene. Helps me focus and get moving. :/
So, yeah, drugs can be useful, actually, if used correctly. Lots of finicky details, of course, and a lot of people probably shouldn't try and use them, but they're far from useless. A good parallel to AI, honestly. XD
Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 @upbeat estimation software Still not. Medicine (which I separate from drugs even if there's a single word for both 😖) has to undergo extensive clinical trials before being put on the market which AI currently isn't bound to.
Would you use a new bootleg drug its manufacturer says it will cure all your ills if only you trust them, although they didn't go through any serious trials? This is what LLMs feel to me at the moment. I don't doubt it has useful applications, LLMs existed before chat bots and was pretty useful as a first pass for translation for example. I've happily used DeepL for years, for example.
But I don't want to have to talk to machines, it just isn't the most efficient way of interacting with them and it tricks people into awkward situations where they trust the machine based on the presentation of the output rather than the output accuracy itself.
Angle🖇
in reply to Hypolite Petovan • • •@hypolite @amsomniac Drugs are just medicine used wrong. I could walk outside and start selling the same pills I use, straight from the pill bottle, and bam, I'd be a drug dealer. Not something I want, but it's a thin line. :/
Of course, for some chemicals, there an argument to be made that there is no good use for them - but that doesn't mean we can dismiss all chemicals. -_-
Angle🖇
in reply to Angle🖇 • • •@hypolite @amsomniac
And I'm kinda iffy about talking to machines too? I use them in Cline, a Codium plugin, mainly, where I can pretty much just give them a few sentences, shove some files at them, and have them figure out the rest. Summarizing error logs or code, suggesting what might be the cause of problems I encounter, that sort of thing.
Angle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •Angle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •Angle🖇
in reply to Hypolite Petovan • • •Angle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •Angle🖇
in reply to Angle🖇 • • •@hypolite @amsomniac Like, the early industrial revolution analogy really works, IMO. You're in the position of a worker seeing the new machines and going 'These things are terrible! They do bad work, they cost too much, they're trying to take my job, I saw one grind a small child into paste right in front of me! I hate them!"
It's a fair viewpoint. From the perspective of the worker, those things are all true - his mistake is in thinking that they're universal and will never change.
Angle🖇
in reply to Angle🖇 • • •Angle🖇
in reply to Angle🖇 • • •Angle🖇
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 @upbeat estimation software I understand the appeal of the Industrial Revolution analogy, but I'm weary of it. On one hand, we know the exact way it went down for industrial machines. They became more precise and safer to use for workers, and then they got relocated to developing countries to save even more on labor. But they were things that needed to be engineered and built by people who built the skills to make them more precise and accurate with safety features.
On the GenAI/LLM front however, we're building global black boxes that aren't engineered for accuracy but for human likeness. No industrial machine engineer was ever convinced their design was their friend and deserved personhood rights.
Since they are global, we all have a personal choice whether to use them or not, unlike industrial machinery that had an expensive upfront cost for business owners only.
Since they are black boxes, we have no reliably way of increasing the precision of the models or measuring the accuracy of their output, and that isn't even an optimization goal for the three main players in the space.
And then for safety, using local models sounds better but they are purposefully mesmerizing machines made to exploit human instincts, so it is only marginally safer.
upbeat estimation software
in reply to Angle🖇 • • •Angle🖇
in reply to upbeat estimation software • • •Hypolite Petovan
in reply to Angle🖇 • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 @Nick As for what to do about it:
The tech already is unpopular after just a few years of mass deployment, so the only thing currently sustaining it is the massive sunk costs by the handful of companies who poured everything they had into an intangible moon shot.
Angle🖇
in reply to Hypolite Petovan • • •Hypolite Petovan
in reply to Angle🖇 • • •@Angle🖇 I understand what you are trying to get at with the warzone analogy, but I don't think it reaches the mark. Mainly because the current crop of AI directly exploits human weaknesses: our tendency to look for humanity anywhere and our longing for convenience in all things. This means the only people wielding it as a weapon are the megacorps. Individuals running homegrown models aren't "fighting back", they just contribute slop at a much smaller scale.
This also means to me that the only way to defeat the megacorps is to reject their premise entirely. No, GenAI isn't an inexorable future, no, I do not want any of its output in any of my workflows, and yes, I prefer humans writing text, drawing images or coding slower and more deliberately. For me, running local models is already giving them the benefit of the doubt, which is more than I want to do.
To take your analogy literally, not everybody in a warzone takes up arms. You still need plenty of civilian support, and that's what I feel closer to than direct combat if it even meant something in the GenAI paradigm, which I don't think it does.
Hypolite Petovan
in reply to Hypolite Petovan • • •@Angle🖇 Ah wait, I found a case where you can somehow fight fire with fire: theregister.com/2025/09/03/ai_…
But for me my main policy would rather be to avoid companies using AI at all. I know this limits my opportunities and I'm not impatient for my possible layoff which will throw me into the IT job search meat grinder, but I'd still favor direct approaches at tech event and mixers rather than looking to automate job search using AI. I'm hoping that I would find my people by doing the legwork and standing firm on my principles, but I'm not eager to see them tested in real life.