Unfortunately, the text does not seem to be visible for some reason.
But anyway. I think this is the best and most open discussion about AI we had so far. There are a lot of good points on both sides of the line, and I enjoyed reading it for the most part which I cannot say of all the threads we had so far about it.
Disclaimer: I actually had computational linguistics, and artificial intelligence as major in university. I pretty much lost interest in this though, afterwards; so while I know the theoretical underpinnings, I’m not really more knowledgeable about LLMs as the next guy.
It is true that AI does miss out on reality, it is not grounded in the actual physical world with emotions, feelings, and physical sensors, and memories. This has nothing to do with having or not having been in the US. But this does not have anything to do with LLMs either. They are not artificial intelligence in the sense of a general artificial intelligence, they are just stochastic machines. They don’t think at all. They also don’t really do any reasoning.
They actually can produce good quality translations for the most part. Much better than any purely mechanical tool, or poor English wording by someone who cannot really speak English. Anyone claiming them to be bad or worse should just put them on a test drive. I greatly prefer reading their output compared to the other possibilities. They obviously are not, and cannot be, as good as a human translator. They may “hallucinate” wildly on questions, but that’s hardly a problem for simple translations.
It’s true that we do not know exactly how the human brain works. But it’s a fact that artificial neuronal networks work similarly even if they are only very simplistic copies. I would even assume that the speech generation in human brains works actually similar to LLMs - we do not have to know any language rules in order to speak our native language perfectly. We sense if something sounds unfamiliar because it’s not common. There are examples of perfectly valid sentences in your native language that you simply don’t understand because they sound unfamiliar. It’s just that we have actual brain functionality beyond speech.
So comparing them with humans is just not justified in my opinion. What they create “freely” is thoughtless stochastic output; it may create something good on pure luck, but it’s mostly closer to garbage. You can greatly use it as a tool for various things. Don’t humanize them, it’s belittling humans. On the other hand, don’t overestimate the human mind either. At least not in comparison to other living beings.
Most of our technological progress is reducing social engagement. From washing machines to smart phones. LLMs are just one step further down that road. Personally, I find the lack of learning how to think when using them incorrectly more worrisome. But then, I’m possibly a nerd that cannot objectively judge what is the bigger threat 
The concentration of power to a small amount of companies, the resource costs, and unethical behavior of these companies are indisputable. Local LLMs can solve most of that, though.
I find it impractical to add the source translation to every post. If someone wants to do that, fine, but I wouldn’t ask it of anyone. But I also find it remarkable, that so many people take the time to use translation tools instead of learning the language; that’s so much overhead in daily communication that I would probably go for the latter… a solution here could be that we include translation services into the forum software, if that’s possible. Then people could just write in their native tongue, and anyone could read the translation, or the original, too, if they understand the language. If that isn’t an option, asking for clarification seems to be the way to go instead of using other translation tools to translate the original to maybe understand it better – that really doesn’t make much sense. Communication beyond the language barrier is always hard, let’s not make it worse.
As for contributions to the code base, I actually think using LLMs is a possible copyright infringement, and should therefore be handled with care. I don’t feel comfortable stealing other people’s work without attribution.
Code quality can also be an issue. LLMs are not able to understand complex problems or code bases (and how could they, they don’t think or actually can understand anything). Using them for smaller applications or UI changes is not an issue, though. So if you have an LLM around that was trained on license compatible material, I personally wouldn’t have any objections if you used them for code ending up in the Haiku repository.
Sorry for the lengthy post, just my 2 cents.
