LittleAlex 🇺🇦🇮🇱🇩🇪🇳🇴
John Tinker@littlealex
True, the AI does not understand human experience. However, it know nearly everything that has ever been said about the human experience. The AI brings a chorus of echos of real authors.
Nini@johntinker @littlealex Eh, do they? They have no knowledge, they have data but it's a dissonant chorus of real authors' stolen words all said at once with no information beyond the fragment they found devoid of context. They might bring those experiences to you but it's truly a hollow understanding of it, twisted into a frankensteinian patchwork quilt that has form but no substance.
@nini @littlealex
Ok. You paint a definite picture. Let me point out a few things. First it is well known that good writers invent, and that great writers steal. I forget who said that. I may have said it myself. The point is that language itself is a matter of words being passed around, and that we do appropriate each other's "work", in the sense of reusing the language of others.
I do not know how many Elvis Presley portraits I have seen, but I do not know that I have ever seen one by an artist who sal Elvis. We share a cultural set of references, ideas, ways to talk about things. All of these were inventions, once upon a the time.
Folk music is filled with the form.
Cutting to the chase: my view is that we should focus on the culprits themselves, rather than the tools that they use. These tools were largely appropriated from the public domain, you may or may not already know. The pursuit of natural language processing, or of understanding grammar itself, has been going on for a long, long, time. We inherit a legacy of our forebears. In the integration of global culture, we inherit this more widely.
Altman named the company "Open AI" to take a ride on the communarian theme of "open", like open source. But there is plenty of open work being done on LLMs by the open source community.
The mathematicians among us will appreciate the importance in not investing time in category errors when discussing the pros and cons of everything that is implied by the nebulous "AI".
The first mistake is to imagine that the chatbot is conscious. Starting from there will require patience. But getting beyond that, the real helpful understanding is to know what a "transformer" prediction does essentially, in its "training".
What is being built is a map of "vectors, in a high-dimensional vector space. The dimensionality is so high, in fact, that it is unabtainable. So the reduce it: vectors within reduced dimensionality vector spaces. This is why the quotations coming from this type of LLM are not quotations, but synthetic products of a generative process. They are informed by the mappings, but the mappings never reach a true conclusion.
There are whole other layers that are called "AI", and it refers to adaptive reasoning. Reasoning here, refers to algorithms, invented by programmers, if you dig deep enough, down to the last turtle.
All of this is simply a machine amplification of a natural process of discovering something. What ARE those patterns with which we describe the human condition? I will testify that I have felt comforted to know that certain people back in history phrased ideas in ways that are close to my own, and which I understand implicitly.
Last point: knowing what it is, what it is not, is the key to interacting with it. The real culprits are those intending to use these tools to harm others. My view is that we should focus more on the culprits themselves, and consider less that the evil resides within the tool.
@johntinker @littlealex LLMs and ML are two seperate things, I am aware but within this context I figure we're talking about generative LLMs only and not ML which is a whole other discipline to my mind and not at all what most people are exposed to nor find themselves bombarded with either.
Here's my view on "AI" as sold by the OpenAIs and Anthrophics of the world, the tool was made to harm others. It is a greedy, rapacious devourer of human made information, deeply inefficient and wasteful. What it makes is fictions, incredibly imperfect fictions but ones that are nonetheless good enough to bypass our cognitive senses and fool us providing you don't look too close. It infects everything regardless of if you use it or not because it damages our perceptions of reality and breaks key elements of societal cohesion like trust and truth. There is so little that's positive about the tech that it cannot hope to outweigh the sheer weight of problems LLMs bring to our species.
Could they be used for good? Not sure, I'm not that confident unless something fundamental about the tech changes or something replaces it that has far fewer moral, ethical, qualitative and legal issues attached.