691–700 of 739 posts
Earlier quoted context omitted.
This is a famous short story just rewritten sloppily by an AI. Although it does amusingly do what annoys critics of AI: take something good written by a person, steal it and slop it up while overall misunderstanding it.
I'm pretty sure it doesn't misunderstand the original. The original says that the aliens don't understand how meat (humans) can be conscious, have language and be basically like the aliens (thinking conscious beings). Their main idea is: humans are meat, meat can't be conscious, therefore humans can't be conscious, but they somehow act as if they are???? The "AI slop" here says that just like the aliens in the origin…
Here's what makes meat special over LLMs: locality and stable identity. Imagine two identical twins for a moment. They have roughly the same hardware and software. They've probably had a lot of the same thoughts. And yet we'd never consider them to have the same consciousness: we recognize that for whatever reason their consciousness is confined to the body they're in. Each request/response pair you make to an LLM go…
> ...no stable identity to it ... what's doing the thinking? The model (its parameters, its architecture and the inference algorithm) is doing the thinking. That's what we call "ChatGPT" or "Claude": it's the same model residing God knows where, somewhere in "the cloud" on some random GPUs. But when you're talking to a model, you can feel that you're talking to the same entity — the same model. It's a bit like the Sc…
Earlier quoted context omitted.
The bit that lost me quite early in the piece was >"A side effect. You're asking me to believe in sentient weights." Huh? Did I miss that logical jump? Genuine question, maybe I'm not clueing into something here.
It's a poor adaptation of the line in the original story ( https://www.eastoftheweb.com/short-stories/UBooks/TheyMade.s... ) The original: >That's ridiculous. How can meat make a machine? You're asking me to believe in sentient meat. I think it was adapted by an LLM which didn't quite get the meaning.
Earlier quoted context omitted.
> I highly recommend people in the AI research space should read philosophy and modern linguistics. On the contrary, I highly recommend people in Philosophy of mind and linguistics should start reading AI research papers because their theories and ideas are highly outdated, even ancient. Your books are from 1927 and 1972 respectively and Turing's article is from 1950s. And they are relatively new with respect to othe…
> On the contrary, I highly recommend people in Philosophy of mind and linguistics should start reading AI research papers because their theories and ideas are highly outdated, even ancient. Your books are from 1927 and 1972 respectively and Turing's article is from 1950s. And they are relatively new with respect to other works in Philosophy. People in philosophy and cognitive linguistics do read AI research. Don't g…
The First Edition (1995) of the classic textbook Artificial Intelligence: A Modern Approach by Russell and Norvig talks about the criticisms of Dreyfus quite extensively.
In the second edition (2003) they conclude: "In sum, many of the issues Dreyfus has focused on-background commonsense knowledge, the qualification problem, uncertainty, learning, compiled forms of decision making, the importance of considering situated agents rather than disembodied inference engines-have by now been incorporated into standard intelligent agent design. In our view, this is evidence of AI's progress, not of its impossibility."
In the 4th edition (2020) Dreyfus reduces to a paragraph and Heidegger is just a reference in a footnote.
Earlier quoted context omitted.
LLMs also have other inputs, like audio and images. They get encoded (just like a human eye encodes an image) and passed to the weights.
I don’t think this analogy holds. The whole way through the processing pipeline in the brain, different sensory data is ingested separately and processed separately; and we still don’t understand how that data is then integrated into a cohesive experience. LLMs have the same fundamental input regardless of modality, tokens. There is a preprocessing step before the “brain”, which is more akin to some super-synesthesia…
Earlier quoted context omitted.
LLMs have really changed the world. I didn’t think something like then would be possible in my lifetime
It came out of nowhere. It’s all emergent. I’m convinced this is possible with just about anything given enough data. We will be seeing a near magical physical outputs LLM in the near future. It’s going to take in video and sounds and spit out physical movements that will be just as mind blowing as when 3.5 came out and it will come out of nowhere.
sufficient telemetry + sufficient compute = AI solution to any problem
From the Universal Approximation Theorem for neural nets, we know that if we have the right training method and net architecture we can get approximate any function with a NN. Of course, that doesn't imply that we actually have a sufficient training method and net architecture for the problem at hand, but we have been able to demonstrably solve at least two engineering domains: physical world navigation (Waymo) and language (GPT). It turns out a robust enough language model is sufficient for reasoning.
Given these results, I am personally stumped to come up with a problem humans can solve now that we can't solve with a computer given the correct telemetry and sufficient compute.
Earlier quoted context omitted.
It doesn’t preclude it correct, however it provides a pretty strong examples of where reductionism is lacking, which what I believe has turned a lot of philosophers of science against pure reductionism (I am probably oversimplifying here. An expert science historian [which I am not] could probably write a whole book about why reductionism is not as popular today as it was in the 1970s). There are a whole lot more phy…
I feel like these are separate things... neural activity being necessary but not sufficient for consciousness does not mean reductionism is wrong, it just means the fundamental building block is not a neuron. It might not even be possible to fully understand the physical mechanisms that underlie consciousness, but that doesn't mean there has to be something more than physical mechanisms.
I could put a bunch of metal and rubber and gasoline in a pile and light it on fire — all the necessary ingredients for a car — but it wouldn’t create a working automobile. The arrangement of the objects and processes matters.
In the same way, if you put a bunch of brain cells together in a Petri dish, but their connections or firings were disordered, I wouldn’t expect consciousness. “Neural activity” is thus insufficient on its own, but this I doesn’t mean reductionism is incorrect. It just means you didn’t correctly reduce the problem to the correct constituent parts. You left some out.
It's not often I see something that's fractally wrong but here we are. There is a dictionary, it's called the tokenizer. There are grammar rules, they are just very weak because the structure of human language is generally quite weak. When presented with languages which have strong consistent grammars the weights are very easily interpretable as a grammar: https://arxiv.org/abs/2201.02177 The point of the original sh…
> fractally wrong fractally or factually? You mean wrong on so many levels you need a fractal to capture them? If so, what if you could use a neural network instead?
Earlier quoted context omitted.
And knowing that structure is about as meaningful as knowing "a PC consists of a keyboard, on which you type, a screen, at which you look, and a processor, which does things with binary logic". None of that helps you understand how exactly LLMs do what they do. Because it describes an interface, not a mechanism. The inner mechanisms of an LLM are more learned than designed. We know what an LLM does on a low level, bu…
We know how LLMs learn at the fundamental level. What we do not know is the actual dynamic process of encoding embeddings and their distributions. Your analogies about the PC and web browser are not correctly formulated, because in the case of the PC you talk about 'external components' (you should be talking about cpu arch, structure, digital components, interfaces, etc); in the case of the web browser, you should b…
Yeah, no. I'm not walking that chain. If you want to, do it, but for now, I'm filing it as "has no evidence and knows it".
By now, there's plenty of works, up to and including direct neural interfaces. Utah arrays, Michigan arrays. Stab the brain, dump the spike trains, decode. You crack the manifold open by correlating to known stimuli using ML, and generalize from there to unknown stimuli. There is no need to "know the exact configuration", and few bother - you put your hardware into the part of the brain you want (top level map is consistent enough brain to brain), gather a set of reference points, and use them to anchor the rest of the decoding process.
Why use ML? Because you need a very expressive correlator to bridge the gap between known inputs and the products of whatever transformations the brain subjects them to before they show up in spike trains.
> So what we need to look at are the fundamental possibilities of the structure of LLMs, not how the weights are distributed.
And the fundamental possibilities are... what exactly? We know the I/O planes, we know the possible flow of information, now, what does that give us?
We know enough to prove that a transformer LLM can implement a Turing machine, the same way a CPU can implement a Turing machine. So an LLM is capable of performing arbitrary computation within its capacity. That's it. That's the upper bound.
What follows is: if you can represent "thinking" as a computational process, you can implement it with a Turing machine, and thus, an LLM can be made to think. That proves LLMs can think. But not that the existing ones do or don't! Because that's the entire thing about upper bounds!
We've looked at LLM architecture, and learned basically nothing about whether LLMs think, other than "it's not impossible". That's the actual "fundamental possibilities" we derived from knowing the architecture. One step above worthless. Oh fun.
(If thinking requires hypercomputation, then, nope. LLMs are out. Good luck proving that it does though.)