151
Today's Large Language Models are Essentially BS Machines
(quandyfactory.com)
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 12 Sep 2023
151 points (100.0% liked)
Technology
37750 readers
299 users here now
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 2 years ago
MODERATORS
LLMs are fundamentally different from human consciousness. It isn't a problem of scale, but kind.
They are like your phone's autocomplete, but very very good. But there's no level of "very good" for autocomplete that makes it a human, or will give it sentience, or allow it to understand the words it is suggesting. It simply returns the next most-likely word in a response.
If we want computerized intelligence, LLMs are a dead end. They might be a good way for that intelligence to speak pretty sentences to us, but they will never be that themselves.
You're guessing, you don't actually know that for sure, it seems intuitively correct, but we simply do not know enough about cognition to make that assumption.
Perhaps our ability to reason exclusively comes from our ability to predict, and by scaling up the ability to predict, we become more and more able to reason.
These are guesses, all we have now are guesses, you can say "it doesn't reason" and "it's just autocorrect" all you want, but if that were the case why did scaling it up eventually enable it to perform basic math? Why did scaling it up improve its ability to problemsolve significantly (gpt3 vs gpt4), there's so many unknowns in this field, to just say "nah, can't be, it works differently from us" doesn't mean it can't do the same things as us given enough scale.
I'm not guessing. When I say it's a difference of kind, I really did mean that. There is no cognition here; and we know enough about cognition to say that LLMs are not performing anything like it.
Believing LLMs will eventually perform cognition with enough hardware is like saying, "if we throw enough hardware at a calculator, it will eventually become alive." Even if you throw all the hardware in the world at it, there is no emergent property of a calculator that would create sentience. So too LLMs, which really are just calculators that can speak English. But just like calculators they have no conception of what English is and they do not think in any way, and never will.
We do not know that, I challenge you to find a source for that, in fact, i've seen sources showing the opposite, they seem to reason in tokens, for example, LLM's perform significantly better at tasks when asked to give a step by step reasoned explanation, this indicates that they are doing a form of reasoning, and their reasoning is limited by what I have no better term for than laziness.
https://blog.research.google/2022/05/language-models-perform-reasoning-via.html
It is your responsibility to prove your assertion that if we just throw enough hardware at LLMs they will suddenly become alive in any recognizable sense, not mine to prove you wrong.
You are anthropomorphizing LLMs. They do not reason and they are not lazy. The paper discusses a way to improve their predictive output, not a way to actually make them reason.
But don't take my word for it. Go talk to ChatGPT. Ask it anything like this:
"If an LLM is provided enough processing power, would it eventually be conscious?"
"Are LLM neural networks like a human brain?"
"Do LLMs have thoughts?"
"Are LLMs similar in any way to human consciousness?"
Just always make sure to check the output of LLMs. Since they are complicated autosuggestion engines, they will sometimes confidently spout bullshit, so must be examined for correctness. (As my initial post discussed.)
You're assuming i'm saying something that i'm not, and then arguing with that, instead of my actual claim.
I'm saying we don't know for sure what they will be able to do when they're scaled up. That's the end of my assertion. I don't have to prove that they will suddenly come alive, i'm not claiming they will, i'm just claiming we don't know what will happen when they're scaled, and they seem to have emergent properties as they scale up. Nobody has devised a way of predicting what emergent properties happen when, nobody has made any progress whatsoever on knowing what scaling up accomplishes.
Can they reason? Yes, but poorly right now, will that get better? Who knows.
The end of my claim is that we don't know what'll happen when they scale up, and that you can't just write it off like you are.
If you want proof that they reason, see the research article I linked. If they can do that in their rudimentary form that we've created with very little time, we can't write off the possibility that they will scale.
Whether or not they reason LIKE HUMANS is irrelevant if they can do the job.
And i'm not anthropomorphizing them without reason, there aren't terms for this already, what would you call this behavior of answering questions significantly better when asked to fully explain reasoning? I would say it is taking the easiest option that still meets the qualifications of what it is requested to do, following the path of least resistance, I don't have a better word for this than laziness.
https://www.downtoearth.org.in/news/science-technology/artificial-intelligence-gpt-4-shows-sparks-of-common-sense-human-like-reasoning-finds-microsoft-89429
Furthermore predictive power is just another way of achieving reasoning, better predictive power IS better reasoning, because you can't predict well without reasoning.
It’s your job to prove your assertion that we know enough about cognition to make reasonable comparisons.
May as well ask me to prove that we know enough about calculators to say they won't develop sentience while I'm at it.
Except calculators aren't models capable of understanding language that appear to become more and more capable as they grow. It's nothing like that.
Isn't it, though? Take two cells and rub them together, do it a bit more, boom here we are on Lemmy.
We wouldn't refer to our consciousness as an emergent property of algae.
Yes, but we would refer to our consciousness as an emergent property of our brain.
And we're trying to build artificial brains.
Sorry, I guess what I was trying to say is that it doesn't take much to get from "calculator" to a system that is Turing complete, and from there we're just a few sleeps away from LLMs, and from there...
Are you just fucking around here? C'mon. In your hypothetical scenario, the emergent property would not be "of a calculator".
So for context, I am an applied mathematician, and I primarily work in neural computation. I have an essentially cursory knowledge of LLMs, their architecture, and the mathematics of how they work.
I hear this argument, that LLMs are glorified autocomplete and merely statistical inference machines and are therefore completely divorced from anything resembling human thought.
I feel the need to point out that not only is there no compelling evidence that any neural computation that humans do anything different from a statistical inference machine, there's actually quite a bit of evidence that that is exactly what real, biological neural networks do.
Now, admittedly, real neurons and real neural networks are way more sophisticated than any deep learning network module, real neural networks are extremely recurrent and extremely nonlinear, with some neural circuits devoted to simply changing how other neural circuits process signals without actually processing said signals on their own. And in the case of humans, several orders of magnitude larger than even the largest LLM.
All that said, it boils down to an insanely powerful statistical machine.
There are questions of motivation and input: we all want to stay alive (ish), avoid pain, and have constant feedback from sensory organs while a LLM just produces what it was supposed to. But in an abstraction the ideas of wants and needs and rewards aren't substantively different from prompts.
Anyway. I agree that modern AI is a poor substitute for real human intelligence, but the fundamental reason is a matter of complexity, not method.
Some reading:
Large scale neural recordings call for new insights to link brain and behavior
A unifying perspective on neural manifolds and circuits for cognition
a comparison of neuronal population dynamics measured with calcium imaging and electrophysiology
If you truly believe humans are simply autocompletion engines then I just don't know what to tell you. I think most reasonable people would disagree with you.
Humans have actual thoughts and emotions; LLMs do not. The neural networks that LLMs use, while based conceptually in biological neural networks, are not biological neural networks. It is not a difference of complexity, but of kind.
Additionally, no matter how many statistics, CPU power, or data you give an LLM, it will not develop cognition because it is not designed to mimic cognition. It is designed to link words together. It does that and nothing more.
A dog is more sentient than an LLM in the same way that a human is more sentient than a toaster.
We all want to believe that humans, or indeed animals as a whole, have some secret special sauce that makes us fundamentally distinguishable from statistical algorithms that approximate a best fit function according to some cost metric, but the fact of the matter is we don't.
There is no science to support the idea that biological neurons are particularly special, and there are reams and reams of papers suggestin that real neural cognition is little more than an extremely powerful statistical machine.
I don't care about what "most reasonable people" think. "Most reasonable people" don't have an opinion about the axiom of choice, or the existence of central pattern generators. That's not to devalue them but their opinions on things this far outside of their expertise are worth about as much as my opinions on the concept of art. I am a professional in neural computation, and I put it to you to even hypothesize about how animal neural computation is fundamentally distinct from LLM computation.
Like I said, we are wildly more capable than GPT, because our hardware is wildly more complex than any ANN, but the fundamental computing strategy is not all that different.
In a more diplomatic reading of your post, I'll say this: Yes, I think humans are basically incredibly powerful autocomplete engines. The distinction is that an LLM has to autocomplete a single prompt at a time, with plenty of time between the prompt and response to consider the best result, while living animals are autocompleting a continuous and endless barrage of multimodal high resolution prompts and doing it quickly enough that we can manipulate the environment (prompt generator) to some level.
Yeah biocomputers are fucking wild and put silicates to shame. The issue I have is with considering biocomputation as something that fundamentally cannot be be done by any computational engine, and as far as neural computation is understood, it's a really sophisticated statistical prediction machine
"most reasonable people" - indirect ad hominem is still ad hominem. You are making a fool of yourself.
They are also fundamentally different from a toaster. But that's completely irrelevant. Consciousness is something you get when you put intelligent in an agent that has to move around in and interact with an environment. A chatbot has no use for that, it's just there to mush through lots of data and produce some, it doesn't have or should worry about its own existence.
So does the all knowing oracle that predicts the lotto numbers from next week. It being autocomplete does not limit its power.
There might be better or faster approaches, but it's certainly not a dead end. It's a building block. Add some long term memory, bigger prompts, bigger model, interaction with the Web, etc. and you can build a much more powerful bit of software than what we have today, without even any real breakthrough on the AI side. GPT as it is today is already "good enough" for a scary number of things that used to be exclusively done by humans.
It literally can't worry about its own existence; it can't worry about anything because it has no thoughts or feelings. Adding computational power will not miraculously change that.
I agree this would be a very useful chatbot. But it is still not a toaster. Nor would it be conscious.
You seem unfamiliar with the concept of consciousness as an emergent property.
What if we dramatically reduce the cost of training - what if we add realtime feedback mechanisms as part of a perpetual model refinement process?
As far as I'm aware, we don't know.
How are you so confident that your feelings are not simply a consequence of complexity?
Even if they are a result of complexity, that still doesn't change the fact that LLMs will never be complex in that manner.
Again, LLMs have no self-awareness. They are not designed to have self-awareness. They do not have feelings or emotions or thoughts; they cannot have those things because all they do is generate words in response to queries. Unless their design fundamentally changes, they are incompatible with consciousness. They are, as I've said before, complicated autosuggestion algorithms.
Suggesting that throwing enough hardware at them will change their design is absurd. It's like saying if you throw enough hardware at a calculator, it will develop sentience. But a calculator will not do that because all it's programmed to do is add numbers together. There's no hidden ability to think or feel lurking in its design. So too LLMs.
I believe I understand everything you are saying and why you are saying it. I think you are completely missing the point, though. LLMs already do quite a few things they were not designed to do. Also, your idea of sentience seems very limited. Yes, with our biological computers we have some degree of presence over "time", but is that critical - or is it just critical for us due to our limited faculties.
What if "the internet" developed some form of self-awareness - would we know? Our entire society could be subtly manipulated through carefully placed latency spikes, for example. I'm not saying this is happening, just that I think you are incredibly overconfident because you have an understanding of LLMs current lack of state etc.
If we added a direct feedback mechanism - realtime or otherwise - we could start seeing more compelling emergent properties develop. What about feedback and ability to self-modify?
These systems are processing information on a level we cannot even pretend to comprehend. How can you be so certain that a single training refinement couldn't result in some sort of spark - curiosity, desire to be introspective, whatever.
Perhaps Hofstadter is losing his mind - but I think we should at least consider the possibility that his concern is warranted. We are not special.
No; they do exactly what they were designed to do, which is convert words to vectors, do math with them, and convert it back again. That we've find more utility in this use does not change their design.
Uh what? Like how would it? This is just technomystical garbage. Enough data in one place and enough CPU in one place doesn't magically make that place sentient. I love it as a book idea, but this is real life.
This would be a significant design divergence from what LLMs are, so I'd call those things something different.
But in any event that still would not actually give LLMs anything approaching: thoughts, feelings, or rationality. Or even the capability to understand what they were operating on. Again, they have none of those things and they aren't close to them. They are word completion algorithms.
Humans are not word completion algorithms. We have an internal existence and thought process that LLMs do not have and will never have.
Perhaps at some point we will have true artificial intelligence. But LLMs are not that, and they are not close.
Are we arguing semantics here?
https://www.jasonwei.net/blog/emergence https://arxiv.org/pdf/2206.07682.pdf https://arxiv.org/pdf/2304.15004.pdf
I could be wrong, obviously, but I don't think this is as straightforward or settled as you are suggesting.
Lol... come on. Your second source disagrees with your assertion:
You are wrong and it is quite settled. Read more, including the very sources you're trying to recommend others read.
You apply a reductionist view to LLMs that you do not apply to humans.
LLMs receive words and produce the next word. Humans receive stimulus from their senses and produce muscle movements.
LLMs are in their infancy, but I'm not convinced their "core loop", so to speak, is any more basic than our own.
In the world of text: text in -> word out
In the physical word: sense stimulation in -> muscle movement out
There's nothing more to it than that, right?
Well, actually there is more to it than that, we have to look at these things on a higher level. If we believe that humans are more than sense stimulation and muscle movements, then we should also be willing to believe that LLMs are more than just a loop producing one word at a time. We need to assess both at the same level of abstraction.
They have no core loop. You are anthropomorphizing them. They are literally no more self-directed than a calculator, and have no more of a "core loop" than a calculator does.
Do you believe humans are simply very advanced and very complicated calculators? I think most people would say "no." While humans can do mathematics, we are different entirely to calculators. We experience sentience; thoughts, feelings, emotions, rationality. None of the devices we've ever built, no matter how clever, has any of those things: and neither do LLMs.
If you do think humans are as deterministic as a calculator then I guess I don't know what to tell you other than I disagree. Other people actually exist and have internal realities. LLMs don't. That's the difference.
As a programmer I can confirm that LLMs definitely have loops. Look at the code, look at the algorithms, you will see the loops. The "core loop" in the LLM algorithm is "read the context, produce the next work, read the context, produce the next word".
The core loop in animals is "receive stimulus using senses, move muscles, receive stimulus using senses, move muscles". That's all humans do, that's all animals do.
I think there's a possibility that humans are simply very advance machines. Look at the debate over whether humans have free will, it's an interesting question and the important take away is that we still have a lot to learn about our brains and physics. I don't want to get into that though.
You've ignored my main complaint. I said that you treat LLMs and humans at different levels of abstraction:
It's not fair to say that LLMs simply predict the next word and humans have feelings and reason.
It would be fair though, to say that LLMs simply predict the next word and humans simply bounce electric-chemical signals between neurons and move muscles.
I don't think that way about people or LLMs though. I think people have feeling and reason, and I think LLMs reason too. LLMs aren't the same as people and aren't as good though. But LLMs are good enough to say that they can "reason" in my experience[0].
[0]: I formed this opinion when learning linear algebra from GPT4. It was quite a good teacher. The textbook I'm using made a mistake that GPT4 caught. I encountered a proof that GPT4 wasn't aware of, and GPT4 wouldn't agree with me that C(A) = C(AA^T) until I explained the proof, and then GPT4 could finally reason for itself and see for itself that C(A) = C(AA^T). As an experiment, I started a new GPT4 session and repeated the experiment using a faulty proof, but I wasn't able to convince GPT4 with a faulty proof, it was able to reason through the math concepts well enough to recognize when a mathematical proof was faulty and could not be convinced by a faulty proof. I tried this experiment 4 or 5 times. To be clear, what happened here is that GPT4 was able to learn a near math concept in one shot (within a single context window), but only if accompanied by a proper mathematical proof, and was smart enough to recognize faulty proofs as being faulty. To me, that rises to the level of "reason".
The two types of loops you equivocate are totally different; saying that a computer executing a program, and an animal living, are actually the same, is very silly indeed. Like, air currents have a "core loop" of blowing around a lot but no one says that they're intelligent or that they're like computer programs or humans.
No; you are analogizing them but losing sense of their differences in the process. I am not abstracting LLMs. That is all they do. That is what they were designed to do and what they accomplish.
You are drawing a comparison between a process humans have that generates consciousness, and literally the entirety of an LLM's existence. There is nothing else to an LLM. Whereas if you say "well a human is basically just bouncing electro-chemical signals between neurons and moving muscles" people (like me) would rightly say you were missing the forest for the trees.
The "trees" for an LLM are their neural networks and word vectors. The forest is a word prediction algorithm. There is no higher level to what they do.
At what level do LLMs teach? Something was teaching me linear algebra and I thought it was the GPT4. When GPT4 was able to recognize a valid mathematical proof that was previously unknown to it, what level was it operating at?
LLMs do not "teach," and that is why learning from them is dangerous. They synthesize words and return other words, but they do not understand the content presented to them in any sense. Because of this, there is the chance that they are simply spouting bullshit.
Learn from them if you like, but remember they are absolutely no substitute for a human, and basically everything they tell you must be checked for correctness.
GPT4 did teach me. I say this as the one who learned, whatever that's worth.
Who cares? This has no real world practical usecase. Its thoughts are what it says, it doesn't have a hidden layer of thoughts, which is quite frankly a feature to me. Whether it's conscious or not has nothing to do with its level of functionality.
Suppose you were saying that about me. How would I prove you wrong? How could a thinking being express that it is actually sentient to meet your standards?
By telling me you are.
If you ask ChatGPT if it is sentient, or has any thoughts, or experiences any feelings, what is its response?
But suppose it's lying.
We also understand the math underlying it. Humans designed and constructed it; we know exactly what it is capable of and what it does. And there is nothing inside it that is capable of thought or feeling or even rationality.
It is a word generation algorithm. Nothing more.
This is false. Read about their emergent properties. We have no way of knowing when emergent properties appear, we just notice them.
I am picking up a hint of the autocompletion you describe, in your writing.
I think I write well :) I am not an LLM though.