"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not.
-
This argument makes no sense.
You don't have to try something to judge it, in fact personal experience is anecdotical and not a good foundation for judgment.
Knowing what the tool is and how it works is more valuable as a judgment criteria.
Especially with something like LLM that "seem" absolutely fantastic at first touch.You're looking like a predicant saying "when you're blessed by the faith, you'll believe!"
Besides that, your tool is built on stolen material, near-slavery training work, it's an ecological disaster and has been designed to reduce the bargain power of intellectual worker like yourself.
You have to be completely ignorant or a complete asshole to be ok with all this.
-
-
Not relevant to whether or not it has failed or works poorly, which is what the original post was about
It is, you can't judge the success and viability of something without considering real cost and externalities.
-
@rainynight65 @skavish @Gargron The thing is, for 90% of software, "don't care" probably is the correct approach from a business management PoV.
-
It is, you can't judge the success and viability of something without considering real cost and externalities.
That much is true. What do you think of local models?
-
"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not. I am curious to see how long the hype will last once the Nvidia ouroboros ends and money stops being pumped into marketing.
@Gargron I basically use it just as a glorified stackoverflow/google. Am I using it wrong?
-
@Gargron What we term AI today is misnamed and misused and oversold.
But the technology as such (i.e. LLMs and Neural Networks and the like) isn't useless.
E.g. my daughter is training AI to recognize and classify life in ROV videos from the deep sea ocean floors.
Many such applications. "AI" doesn't get bored and is extremely good at spotting patterns and things that break the patterns.
But, yeah, it is shit for coding...
@steinarb The thing is, you obviously know the difference between those models, yet you still managed to somehow confuse them. LLMs have clearly nothing to do with what your daughter does. AI is a buzzword and it basically means nothing at this point. There are lots of specialized ML (machine learning) models that are actually helpful (like what your daughter does). But LLMs ain't it, chief. They're just a waste of electricity and water.
-
"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not. I am curious to see how long the hype will last once the Nvidia ouroboros ends and money stops being pumped into marketing.
@Gargron It is so refreshing to see one of your favourite devs speak in this way about this sh*t. I swear, AI is really bringing my mood down as of late.
-
That much is true. What do you think of local models?
@golemwire Local model of what? Language model for general purpose/coding?
They are better under the consume aspect, but they share the same problem in all the other aspects (stolen training data, salevry-like training, sociological and psychological impact and so on). -
"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not. I am curious to see how long the hype will last once the Nvidia ouroboros ends and money stops being pumped into marketing.
@Gargron I disagree with calling it a failed technology. The Google Brain project that spun up the topology that made the LLMs in 2016 functional was a huge step forward in machine learning in general. We've been working on AI for 70 years. No matter what they want to call it, it's more than a fucking chatbot. We need to progress this. We just need to stop treating it like it's something it isn't.
-
@clayote the difference is that i actually use the tools and build real software with them. you are arguing about a guitar you have never played, based on people you do not respect playing it badly. that is not exactly a solid basis for judging the instrument.
@skavish @clayote Great, if you are looking for someone who has actually used thesr tools in real life, I'll raise my hand. I was lead of DevAI tools at a major tech company, and I've used them to build websites, infra, and other tooling.
While you can get these tools to write code that "works", in my experience they constantly add tech debt to your project, sometimes in obvious ways, often in subtle ways. Since they have no mental model of how the code base should work, they are constantly violating API conventions. They tend towards verbosity -- in one project, it implemented Haversine distance repeatedly in whatever class it felt it needed it, despite explicit instructions in .md files to use the canonical version.
You can of course keep correcting the bot and eventually get decent results. But that assumes that you are doing thorough code review of *everything*, and what's more, that you yourself actually understand a code base that you aren't actively engaged in writing. That's *hard*.
-
@skavish @clayote Great, if you are looking for someone who has actually used thesr tools in real life, I'll raise my hand. I was lead of DevAI tools at a major tech company, and I've used them to build websites, infra, and other tooling.
While you can get these tools to write code that "works", in my experience they constantly add tech debt to your project, sometimes in obvious ways, often in subtle ways. Since they have no mental model of how the code base should work, they are constantly violating API conventions. They tend towards verbosity -- in one project, it implemented Haversine distance repeatedly in whatever class it felt it needed it, despite explicit instructions in .md files to use the canonical version.
You can of course keep correcting the bot and eventually get decent results. But that assumes that you are doing thorough code review of *everything*, and what's more, that you yourself actually understand a code base that you aren't actively engaged in writing. That's *hard*.
@skavish @clayote Yes, humans write shitty code too. That's why the study of software engineering processes and culture have been so critical to our field. We've learned over time how to create a process that generated useful code with imperfect humans.
But the lessons we think we learned with human software engineering are being forgotten, or what's worse, misinterpreted. They were designed for humans, who work by building internal mental models of the world, and instinctually strive to sync their mental models with their colleagues. But LLMs do not build mental models. They predict what is most likely based on their context window. That's a vast and critical difference.
As these tools are being used today, they are becoming engines of tech debt creation. And anyone who's worked in software knows what a trap that can be. The gains become incrementally smaller as the inertia of tech debt becomes larger. At some point, nobody knows how it works any more, and then you're fucked.
-
"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not. I am curious to see how long the hype will last once the Nvidia ouroboros ends and money stops being pumped into marketing.
@Gargron The pursue of strong AI or the so-called “AGI” is disgusting and undesirable to start with.
So, even if it was a path to “AGI” I would fucking hate and try to boycott it.
-
@golemwire Local model of what? Language model for general purpose/coding?
They are better under the consume aspect, but they share the same problem in all the other aspects (stolen training data, salevry-like training, sociological and psychological impact and so on).I did mean local model as in local LLM, like Qwen, DeepSeek, or GLM, and probably for coding or other such tasks.
I think some people don't understand the implications of posting content in the public.
When you do that, while people can't take it and redistribute it, as far as I can tell it is fair use to use it to create something else that can then create something else that is different from the original. That seems like fair use to me, so long as the models aren't regurgitating the same training data, but that's still debated obviously.
As for the sociological and psychological impact, this is one of the reasons I discourage some of my friends from relying on it. My usecases are narrow enough that I've only benefited. But there's little point in me avoiding using it for good purposes, since avoiding it won't stop other people from using it for bad.
-
@Gargron @imhotep @skavish On the other hand you say in another post, that you have never used AI, but maintaining one of the biggest FOSS project on GitHub is enough expertise.
Im no Dev, but I assume Linux may be a even bigger FOSS project and Linus explored AI. According to your argumentation, he is as capable as you, if not more, to make guesses about GenAI.
Or what’s the difference, that make your opinion superior to his one?
@kernic @Gargron @imhotep @skavish not «superior» I think, just watching the influx of low quality «contributions», another well documented issue in pretty much any open source repo, I think is enough experience to speak about AI impact in society irrespective of the fact that someone finds it useful.
The same is happening in most companies with much less visible impact at the moment. AI is an amplifier. Everyone that was slowly going in the wrong direction, now is going faster. -
I did mean local model as in local LLM, like Qwen, DeepSeek, or GLM, and probably for coding or other such tasks.
I think some people don't understand the implications of posting content in the public.
When you do that, while people can't take it and redistribute it, as far as I can tell it is fair use to use it to create something else that can then create something else that is different from the original. That seems like fair use to me, so long as the models aren't regurgitating the same training data, but that's still debated obviously.
As for the sociological and psychological impact, this is one of the reasons I discourage some of my friends from relying on it. My usecases are narrow enough that I've only benefited. But there's little point in me avoiding using it for good purposes, since avoiding it won't stop other people from using it for bad.What? You don't know what fair use is, man.
You absolutely cannot do whatever you want with things just because they are online, especially when protected by copyright or other licenses. -
What? You don't know what fair use is, man.
You absolutely cannot do whatever you want with things just because they are online, especially when protected by copyright or other licenses.I didn't say you can do whatever you like with online things just because they're online. For example I said that you should not redistribute them and that the models should not output training data (thus redistributing, more or less). Instead, they produce new content. But the legality of training AI from random public-facing data is still debated.
I found this PDF from a court case with Anthropic: https://www.copyright.gov/fair-use/summaries/Bartz-v-Anthropic-PBC-787-F-Supp-3d-1007-ND-Cal-2025.pdf
In this court case, it was said that Anthropic could train on copyrighted works since its AI model did not output said copyrighted works. It was also concluded that Anthropic was at fault for keeping copies of the books after using them for training, and had to delete the copies.
I avoid Claude (haven't used it) because they apparently didn't obtain all of the books legitimately, using piracy sites for some of the downloads rather than purchasing all the books.
While I avoid particular LLMs, I consider LLM training in general to be fair use. Though I can't fault someone for rejecting LLMs for considering them not to be.
-
"AI" is a failed technology, in the sense that it's neither fit for purpose, nor a likely path to its stated end goal of "AGI", but a lot of people's profits currently depend on pretending it's not. I am curious to see how long the hype will last once the Nvidia ouroboros ends and money stops being pumped into marketing.
-
@mellowfish @Gargron I do not find your personal experience convincing.
-
@Shes_Gabrielle @FreakyFwoof @skavish @Gargron excellent definition of centrism, and I'll stop here. Many implications.
@precariousmind
@Shes_Gabrielle @FreakyFwoof @skavish@welley.social @Gargron
Many such cases.
