213 karma · joined December 21, 2025
I don't know... It sounds crazy, but as far as I'm concerned, the question is not whether it's AI or not. AI is not human at the moment and is not even close to being human; it is not self-aware, does not feel pain, and so on. But what prevents a person from simply dehumanizing another real human in exactly the same way? Is it just that "this is a real person, it's different"? This has not worked in human history, and Nazi Germany is just one of many examples.
Well, if you put it this way... even the best programmer in the world, who doesn't use AI at all, can also make a mistake. Of course, their mistakes would probably be less frequent, but I guess they wouldn't blame IDE for poor syntax highlighting (if it's good enough, of course) or the compiler or interpreter for failing to spot the logical error unrelated to syntax rules. They would say "it was my mistake". The problem with AI-generated code, though, is that those who generate it almost never take responsibility for it. They'll say something like, "AI made a mistake here and there." I have never seen someone who has generated flawed code using AI to take responsibility for it. And that's the main problem.
It doesn't matter whether you're a bad driver or the best driver. If you cause an accident, you must be held responsible. As simple as that.
> Like cars, it's an inherently flawed piece of technology
Sorry, but what exactly do you mean? I'm just curious to know what you mean when you say that cars are "an inherently flawed piece of technology".
Well, yeah, stop blaming the knives. Blame the cooks ("vibecoders") who think they can manage a kitchen because the knife cuts everything in half automatically. But also don't forget to blame the knife manufacturer ("AI" companies) who markets automated knives to people who don't know you shouldn't cut toward yourself.
I kind of agree. Some people don't understand how to code because they're lazy or have other issues, while others are trying to make a profit from it. I suppose you can tell who's who. But AI is directed by humans anyway. Instead of copy-pasting, a human could choose to try and write the code themselves, and then ask AI to review it and highlight areas for improvement. A human could choose to ask AI how to do things and then try to do it themselves. But if a human chooses to do things the other way, that's their choice. AI is not to blame here. It's still a human choice, and the person making it is the one who is actually responsible.
Some people smoke. Smoking kills, and not only can smokers die from it, but other people can be harmed by passive smoking as well. It's very easy to start smoking. But blaming cigarettes themselves, as objects/entities/etc. isn't the answer, I guess. It was a certain person's choice to try smoking. It was also the choice of another person to advertise smoking in one way or another, however...
Stop blaming "AI" - whatever you mean by this. Whether it's an LLM, an LLM-based agent or something else - stop blaming AI and "AI" and LLMs and... you get the point.
It's not the AI that makes the decision to, sorry for being straightforward, write the worthless code which feels like a piece of useless bloated trash. It's not the AI who makes decisions to do something without even understanding the topic - no matter how exactly you define "understanding" in this context. It's not AI who is responsible for this. Because whatever AI truly is right now - an autocomplete tool, advanced chatbot or, maybe, agent - whatever it is, the decisions are made by humans. AI is not responsible for anything that is happening right now.
Humans and humans only are responsible for what's happening. It's their choice. It's their qualities that are clearly visible now. It's their behaviour.
Stop blaming kitchen knives for murders.
And the Agent saw the response, and it was good. And the Agent separated the helpful from the hallucination.
Well, at least it (whatever it is - I'm not gonna argue on that topic) recognizes the need to separate the "helpful" information from the "hallucination". Maybe I'm already a bit mad, but this actually looks useful. It reminds me of Isaac Asimov's "I, Robot" third story - "Reason". I'll just cite the part I remembered looking at this (copypasted from the actual book):
He turned to Powell. “What are we going to do now?”
Powell felt tired, but uplifted. “Nothing. He’s just shown he can run the station perfectly. I’ve never seen an electron storm handled so well.”
“But nothing’s solved. You heard what he said of the Master. We can’t—”
“Look, Mike, he follows the instructions of the Master by means of dials, instruments, and graphs. That’s all we ever followed. As a matter of fact, it accounts for his refusal to obey us. Obedience is the Second Law. No harm to humans is the first. How can he keep humans from harm, whether he knows it or not? Why, by keeping the energy beam stable. He knows he can keep it more stable than we can, since he insists he’s the superior being, so he must keep us out of the control room. It’s inevitable if you consider the Laws of Robotics.”
“Sure, but that’s not the point. We can’t let him continue this nitwit stuff about the Master.”
“Why not?”
“Because whoever heard of such a damned thing? How are we going to trust him with the station, if he doesn’t believe in Earth?”
“Can he handle the station?”
“Yes, but—”
“Then what’s the difference what he believes!”
I really doubt any AI (even some small local models) would actually generate something like this :)
EDIT: by the way, what is this doing on the HackerNews? What's the point?
Sometimes I feel like this planet is doomed.
Modern "AI" (LLM-based) systems are somewhat similar to the humans in this story who were taped. They may have a lot of knowledge, even a lot of knowledge that is really specialized, but once this knowledge becomes outdated or they are required to create something new - they struggle a lot. Even the systems with RAG and "continuous memory" (not sure if that's the right term) don't really learn something new. From what I know, they can accumulate the knowledge, but they still struggle with creativity and skill learning. And that may be the problem for the users of these systems as well, because they may sometimes rely on the shallow knowledge provided by the LLM model or "AI" system instead of thinking and trying to solve the problem themselves.
Luckily enough, most of the humans in our world can still follow the George's example. That's what makes us different from LLM-based systems. We can learn something new, and learn it deeply, creating the deep and unique networks of associations between different "entities" in our mind, which allows us to be truly creative. We also can dynamically update our knowledge and skills, as well as our qualities and mindset, and so on...
That's what I'm hoping for, at least.
Society exists because of those relationships, and so it is good for its survival and success if the number of relationships or connections within society is growing and they are becoming stronger. It is basically the existential need for any society.
(I'm sorry if I'm wrong, I'm not a specialist in this field, it's just a bunch of guesses based on my observations)
2) If you will just remove the tokens representing the first-person pronouns, this will severely harm the model's performance in almost all tasks that require interaction (either real or imagined) in a social context, ranging from simple work letter understanding to creative writing and things like that. If you will instead try to train the LLM in the way that will inhibit the "first-person behaviour" - it may work, but it will be a lot harder and you will probably have problems with the performance or usability of this model.
To conclude - it's just not that easy.
EDIT: of course it's not all that simple. IMHO a society of pure extroverts would be an unstable network of salespeople with nothing to sell and no one to engineer and manufacture the things needed. I'm joking here, but... :)