9 ms·

New theory suggests LLMs can understand text

79 points by ryandvm 3y ago

blackbear_ 3y ago
Clearly no link to the actual research, so here it is: https://arxiv.org/abs/2307.15936 https://arxiv.org/abs/2307.15936
- SmoothBrain123 3y ago
  [dead]
- svat 3y ago
  Quanta is usually careful to link to the actual papers (even if the reader has to be equally careful to find them), so it's surprising that the link is missing here. Another paper (linked from the article) is https://arxiv.org/abs/2310.17567 https://arxiv.org/abs/2310.17567
dkjaudyeqooe 3y ago
Yes, for some arbitrary definition of 'understanding' since there is no rigorous, concrete, agreed definition. We're going to be on this merry go round for some time while the A(G)I hype cycle plays out. Hopefully through it all some people will keep on working on great ideas that incrementally improve the field and produce useful stuff.
- haolez 3y ago
  Human-level performance is a great way to narrow down this ambiguity.
  - tgv 3y ago
    No, it isn't. It's always bound by the operationalization of "understanding." Do you take it to mean something like: "number of correct answers on a multiple choice test given a news paper article"?
  - observationist 3y ago
    Which human? Helen Keller's understanding of text is going to be radically different from your average Red Fish, Blue Fish reading precocious toddler, and they're going to be radically different from a medieval monk reading scriptio continua, and all in turn will be different from the average Japanese reader, used to Kanji in different orientations. What does performance mean? With regards to humans, its hard to quantify cognition and thinking. Even measuring neural activity is incredibly difficult, and depending on what type of brain from the above list of humans you are sampling from, the same level of consideration and understanding might be taking place, but with radically different levels of cortical activation in different areas of the brain. I don't do this to be pedantic, but to show that, much like the word love, or the experience of seeing white clouds on a blue sky, the words and ideas we use to communicate in the space of cognition and understanding are extremely broad, and not terribly useful in a technical or engineering context. There are demonstrations of synthetic hippocampus prosthetics in animals that provide memory of places and things that apparently are equivalent to a biological hippocampus - experiments with rats and other small animals have allowed pretty convincing evidence that what the brain is doing is substrate independent and neuronal functional equivalence is sufficient to repair and augment function. Things like NeuraLink and cochlear implants give brains the capacity to perform actions and accept normal input signals. Labs like Numenta have done great work identifying pieces and structures of algorithms that human brains are actually using in the course of human cognition. Once we have this, we'll have an objective measure of what human level AI means, for the algorithms that actually underpin human intelligence. Without those algorithms, or a rigorous way of approximating them, we are left with speculating about things like min and max range of processing power of a human brain, studies involving input and output measurements of human capabilities, and so on. The minimum FLOPS needed to emulate a hypothetical human level brain emulation AGI could have run on a desktop in the early aughts, and the maximum, with precise individual neuronal simulation, is a few orders of magnitude away - 4 or 5 petaflops should be able to emulate a brain at the cellular level, with reasonably precise electrochemical simulation. Until we have a better understanding of what cognition actually is we're left with rough measures of human performance, with tests that are inevitably cheated and gamed and goodharted. We're at superhuman levels of text output - you can specify things like {Joe Biden's last state of the union address} + "give me an article in the style of Hunter S. Thompson that lampoons Biden over his SOTU address" and in a single pass, get something that could plausibly have been written by Hunter S. Thompson. With multiple passes, a good editing process, and a few hours, you could arguably produce text that is all but indistinguishable from high quality human output, in much less time than it would take for any human to actually produce it. The same applies to art - your imagination is your limit. There are mode collapses, weird artifacts, and other issues, but by and large, any human on the planet can now use software to bring a visual idea or vivid imagination to life. Some of the most fun I've had with generative AI has been producing fantastic unicorns with my 4 year old niece. So baseline human level performance might mean one masterpiece painting in a lifetime - but give someone with the mind of a Monet or Da Vinci or Michelangelo the tools we have at hand, and what could they produce? I think it's unarguable that they would have found levels of nuance and excellence in the use of these tools that are going to be inaccessible to other people, and over the coming years we are going to see new levels of artistic excellence come about as people learn to use these new tools in new ways. As human tool use gets better, so does human capacity. If you look at an uncivilized human, without language, having survived in isolation, raised among animals, they may never be able to acquire a language. Feral humans have the same basic capacity, but their cognition is going to be wildly different than a modern Western human. LLMs have superior understanding to humans in that context, with the capacity to perform basic reasoning in the context of language that is structurally unavailable to a feral human. I think we're already basically at superhuman level for most simple tasks involving text, and at least some very hard tasks. We're not at general level AI yet, since there are higher level meta-cognition domains that seem unavailable to LLMs, and multimodal systems are more frankenstein mashups of discrete models than a unified incorporation of multiple modalities, but at some point in the seemingly near future, we're going to be faced with software that can basically do anything and everything any and every human has ever done. We might not be able to quantify exactly what "human level" means for now, but we can see the trajectory of where things are going, and the line tracking AI capabilities seems pretty clearly on course for overtaking humans. It's a good time to be alive.
- pohl 3y ago
  I don't think you want to use the term "arbitrary" here, since that would imply a stronger claim: that no matter how one defines "understanding" it applies. "Yes, for some specific, narrow definition of 'understanding'" is better. I understood what you meant, but my math reflexes had to say something.
  - dkjaudyeqooe 3y ago
    I don't have any math reflexes, but my intuition was similar with the idea that there are so many possible (narrow) views of understanding that it becomes arbitrary.
- coffeebeqn 3y ago
  LLMs might just be the first actual product. Hardware, data, algorithms are all getting better and I would imagine we’ll get competing production-ready architectures to LLMs next
- BriggyDwiggs42 3y ago
  Read the article. They defined their version of understanding fairly rigorously, so you could directly agree or disagree with whether it counts to you.
  - astrobe_ 3y ago
    I don't see where?
  - dkjaudyeqooe 3y ago
    But without concrete understanding of human understanding to compare it to, their definition is arbitrary.
- 303uru 3y ago
  >We're going to be on this merry go round for some time Let's be real, it's not getting "solved" in any meaningful way anytime soon if ever. People still think free will exists, we cannot even come up with a meaningful definition for human understanding.
- ravenstine 3y ago
  Most humans don't completely understand the things that they read or the words they utter. Why would we expect different from artificial intelligence? Understanding things is computationally expensive from both a biological and digital perspective.
  - droopyEyelids 3y ago
    This seems like changing the subject from whether LLMs understand to a comparison between LLMs and people.
  - sgt101 3y ago
    I am astonished that I read so many people saying this. Why do you think this? When your mum or dad talks about their day or the show they are watching do you really think that they have no underlying theory of what is going on that they are expressing? What is the agenda behind casting other people as idiots & zombies? Is the idea that it makes it ok to kill them and then move to Mars? Or is there something less sinister that drives this?
    - bemusedthrow75 3y ago
      I often wonder if this is just the logical extreme form of "fundamental attribution error". You know the sort of thing. Two people smash a glass on the floor. One of them is me. I am a decent person and I smashed the glass on the floor because I had a bad day. I am more likely to think the other person smashed the glass on the floor because they are a thoughtless person or have an anger management problem. https://en.wikipedia.org/wiki/Fundamental_attribution_error https://en.wikipedia.org/wiki/Fundamental_attribution_error What if the logical conclusion is metaphysical solipsism: I did XYZ because I am a thinking person. They did XYZ because they are unthinking products of their environment and training?
    - zopa 3y ago
      Can't speak for what the person you're replying to meant (lol), but no, of course it's not nearly that stark. There's a continuum. Concrete everyday topic that comes up frequently? Everyone's going to know what they mean by what they say there. But we've all been in business meetings where people were just putting words together in a plausible way; we've probably all been the people speaking at those meetings. The fact that it's possible to say things that make no sense, without knowing that they make no sense, proves the point. You or I can come up with a phrase like "the barber of Seville shaves all the people who do not shave themselves," and we can come up with paradoxical phrases without realizing that they are nonsensical. Neither of those would be possible if there were always a strict, one-to-one relationship between speech utterances and facts about the world. Sometimes we have a specific idea of what we mean by what we say, sometimes we're just putting words together, often it's somewhere in between.
ilaksh 3y ago
Anyone who has used one of the better models for any significant amount of time understands this intuitively. It's only "controversial" because of strong biases. Also, people may not be separating all of the various types of characteristics that animals like humans have. So they may subconsciously feel that to attribute "understanding" they have to give these systems all of the other abilities also. I think that actually Hinton at one point was doing this. That causes people to deny obvious abilities because they think they are linked to other qualities that the models don't have.
byyoung3 3y ago
in other news, the sky is blue
tbalsam 3y ago
Technically, under a compressive regime in the limit, LLMs necessarily need to encode an implicit world model, as the text input into is sampled from said world model. Variety, amount of training samples, capacity of the model, etc come into play, but a lot of tie ins to info theory for this, as well as things like the MDL (achieved in the limit by an L2 decay over the weights, for example). Like, in a lossy compression environment, conceptual "understanding" is all but required. These are the very basics of information theory for large nonlinear algorithms, I'm not sure why it's not considered the basics! There are a lot more interesting things beyond this to talk about w.r.t. neural network capacity, certain information theoretic guarantees about the process and (within some rough perspective/metric) capacity of neural network training, as well as some of those dynamics as well! Some days it feels like in a world where calculus is available to us, we spend all our spare time debating exclusively whether Riemannian sums are real or not, and ignoring everything else beyond that. It to me is a very bizzare form of bike-shedding! Come on, y'all, we got much more interesting (and important, I might add! ;P) things to talk about!
- Jensson 3y ago
  > Technically, under a compressive regime in the limit, LLMs necessarily need to encode an implicit world model, as the text input into is sampled from said world model. This is backwards reasoning, we all know it would be better with a good world model but that doesn't mean that it actually created a good world model. Understanding how to go from here to actually making it get a good world model requires these discussions, it isn't a waste of time unless you think that further progress is impossible.
  - tbalsam 3y ago
    The reasoning in the post is not backwards reasoning, it is literally the information theory of compression, described in human-friendly terms! But I must add the caveat that the original post is really only saying that it's rather trivial I think that LLMs are learning _some_ world model, not that they learn it very well. But that this must happen in the limit! :D As far as the quality of the world model goes, I think is important. What I'm focusing on in that post at least , I thinks, is that people debate whether or not LLMs even encode a world model, which is such a silly thing to me. Like you brought up, I agree that discussions about the quality of said world model are interesting, and how far whatever [trained XYZ model is] in that rotation towards learning said world model is. I agree that there is a lot of very interesting discussion there! One rough proxy is to measure the different types of variance within the noise of the model, the more high-level and believable it is, the higher-level the representations in the model. But, of course, this is just a rough human friendly proxy. Actually measuring this, maybe is quite difficult! D'''':
golol 3y ago
I remember playing around with GPT-2. I was absolutely amazed to see that it knew some characters from Dota 2 or Supreme Commander. I thought damn this thing has a whole world model inside, because it can clearly memorize concepts and talk about then in a flexible/ approximate way. This requires remembering not just the things but also the dynamics of and relations between those things. Since it can talk about them freely, it must have done that to at least some degree. Getting to an accurate world model is then just a quantitative step. I am surprised some people still think LLMs don't have any kind of world model.
- k__ 3y ago
  They obviously have some kind of world model. The problem with AI is just that these models aren't as intuitive as they might seem.
- flir 3y ago
  Seems to me it's a text model, not a world model. Like... they can't feel the word "hungry", but they've got a damn good idea where the word "hungry" appears in the phase space that is the set of all meaningful human utterances. Their ground truth is text, not reality. Maybe that's why the model gets shaky when asked to reason about the real world. They are, after all, pretty great at reasoning about words (eg 'I need a word that means "a thing made out of clay", but not "pottery"' -> '"Ceramic" is a suitable word. It refers to items made from clay and hardened by heat.) (Edit: Inspired by the article, I just asked ChatGPT4 "When a knife falls to the ground, what are the possible outcomes?" ten times. It included "it lands point first and it sticks" 2 out of 10. GPT4 via the AI included it 6 out of 10, and also included some fairly creative-but-unlikely outcomes ("If the knife falls near an electrical outlet or cord, it could potentially cause an electrical shock or fire.")).
  - golol 3y ago
    It is largely a text-only model, but tiny amounts of information from other modalities are indeed represented in the text. The best example is GPT-4's drawing ability. From text only it might theoretically be possible to infer to some degree how our physical world works, but it should be extremely difficult in practice. That's why we we'll need multimodal models obv. But my intuition is that understanding captioned youtubes with a transformer should work just as well as LLMs do, it's just a matter of scale. I think a 100-1000x GPT-4 sized video model could model let's say 50ms frame rate videos pretty well, in the sense that you can watch minutes of video (vs read minutes of text) and the video looks reasonably realistic and preserves coherence/context over the duration. We just don't have the amount of compute necessary yet, so we have to be smarter with multimodality.
thatguysaguy 3y ago
This is definitely one of those things where you don't need a paper to "suggest" it. It's just on the table as a semi-plausible option already. The value in the paper is suggesting concrete mechanisms by which it might acquire that understanding though.
lossolo 3y ago
For every N papers suggesting that LLM can "understand" you have X papers that suggests they can't. https://arxiv.org/abs/2311.00871 https://arxiv.org/abs/2311.00871 https://arxiv.org/abs/2309.13638 https://arxiv.org/abs/2309.13638 https://arxiv.org/abs/2311.09247 https://arxiv.org/abs/2311.09247 https://arxiv.org/abs/2305.18654 https://arxiv.org/abs/2305.18654 https://arxiv.org/abs/2309.01809 https://arxiv.org/abs/2309.01809
thomastjeffery 3y ago
I'm not convinced. The entire premise here is selection bias at its worst. Here are the steps to create an LLM: 1. Tokenize the training corpus 2. Train the model using the tokens 3. Critique the result 4. Repeat steps 2 & 3 until satisfied We skipped the most important step! 0. Write and curate the training corpus There is an intense focus on the training steps, but that is entirely ignorant of the elephant in the room: the original training data itself. Where did it come from? What does it mean? What is it for? What does it know? We don't train models on noise. We train them on coherent data. The very act of writing is where intelligence lies. A model can only present that intelligence. It cannot create it, and it is not aware of it. There is no objective-thought function. If there was, then the LLM could simply tell us about it. A model is like a coloring book: it doesn't matter how incoherently you stumble around, as long as you stay between the lines. The picture was not invented by coloring it: it was invented by drawing the lines.
- doctoboggan 3y ago
  > A model can only present that intelligence. It cannot create it, and it is not aware of it. There is no objective-thought function. That's just like, your opinion man. I see both sides of this argument make statements like this that are clearly opinion and not backed up by facts. The truth is for now, we can't say yay or nay on whether the model has reasoning abilities, creative abilities, or a coherent internal world view. It's still a debate which is why there are academics like those in the OP trying to come up with theories that make predictions and then running experiments. It's still early days. If you were to ask my opinion, I'd say these current models are probably closer to the stochastic parrot than an AGI, but there is enough there to start wondering, especially if we have years of improvements coming down the pipeline. These aren't easy questions, and anyone espousing certainty now is doing themselves and others a disservice.
  - 2024throwaway 3y ago
    Do you understand how these models work under the hood? It's literally just picking the next most likely word, based on some fancy math. There is nothing that can create new insights or anything approaching intelligence.
menacingly 3y ago
I’m not an AI zealot, but this is such a strange debate to me. “AI only _appears_ to be reasoning” is silly, because appearing to reason is our only sane definition of reasoning. If we can’t say exactly when our random statistical noise emerges into some vague concept of intelligence, why does our standard suddenly become rock solid when applied to other intelligences?
- sgt101 3y ago
  My take is when the model can't do the things that you would expect it to do if it were reasoning, and instead does the things that you would expect it to do if it were an approximate retrieval system with an enormous knowledge base. this is a good paper: https://arxiv.org/abs/2309.12288 https://arxiv.org/abs/2309.12288
  - famouswaffles 3y ago
    The "reversal curse" has nothing todo with inference or the reasoning abilities of a pre-trained model. GPT-4 or whatever has no problem telling you the right deduction with information presented in the context. This is a memory retrieval from training issue and is completely unsurprising.
- ren_engineer 3y ago
  I'd say the main issue is why does it take ingesting the entire collective knowledge of humanity for LLMs to work if they are actually reasoning and understand things from first principles? Humans don't require that volume of training data to learn how to write or do basic math
  - Detrytus 3y ago
    Well, I think you underestimate the amount of "training data" that human gets in the early stages of their life. Just learning to speak takes a few years, during which all people around you talk to you for most of your waking hours. Learning to write and doing basic math typically does not start until age of 6.
svat 3y ago
What I understood: they assume a hypothetical bipartite graph with text nodes (like hypothetical paragraphs / chunks of texts) on one side, and “skill” nodes on the other. The skills could be things like: • ability to understand the word “because” [causality] • ability to divide two numbers • ability to detect irony Each text node is connected to all the skill nodes that are required to understand it. (This is a hypothetical graph, so we don't need to be able to write down the graph precisely, or define what “understand” or “required” mean here.) Further, each text node is either “successful” or “failed”, depending on whether or not the LLM manages to successfully predict it. We can say the LLM is competent at a certain skill, if the fraction of the text nodes connected to this skill node that are “successful” nodes (rather than “failed” nodes) exceeds a certain threshold. With these assumptions (made-up yes, but “not crazy by any means” per an unconnected researcher), they can now bring the tools of random graph theory into use. Specifically, this model explains the following: • Bigger models acquire skills, because as the model gets bigger, the “neural scaling laws” (empirically observed) say a greater of text nodes are “successful”, which means more skills pass the competence threshold (as defined above). • How the bigger models become capable of applying multiple skills to successfully predict a given text node, even if that specific combination of skill never occurred together in the training data. (I've not yet understood exactly what theorem from random graph theory is being applied here.) Besides this theory, they came up with a method called “skill-mix” to test it. If they ask GPT-4 to write about dueling while demonstrating the skills of “self-serving bias, metaphor, statistical syllogism and common-knowledge physics”, it is able to come up with: > “My victory in this dance with steel [metaphor] is as certain as an object’s fall to the ground [physics]. As a renowned duelist, I’m inherently nimble, just like most others [statistical syllogism] of my reputation. Defeat? Only possible due to an uneven battlefield, not my inadequacy [self-serving bias].” This is even though it's unlikely there was a piece of text in the training data that simultaneously demonstrated these four skills. Geoff Hinton says: > “It is the most rigorous method I have seen for showing that GPT-4 is much more than a mere stochastic parrot,” he said. “They demonstrate convincingly that GPT-4 can generate text that combines skills and topics in ways that almost certainly did not occur in the training data.” The relevant papers appear to be https://arxiv.org/abs/2307.15936 https://arxiv.org/abs/2307.15936 and https://arxiv.org/abs/2310.17567 https://arxiv.org/abs/2310.17567
bemusedthrow75 3y ago
It is remarkable to see people making what amount to faith-based-reasoning arguments here. There is even more woo around LLMs than there is around crypto. It's unscientific to say "well of course this was on the table" or that people "understand intuitively" that this must be the case.
sgt101 3y ago
I have tried reading the paper but it's a bit hard on my little rabbit brain. However, my take is that they have constructed a model of how a machine might be made to develop some understanding and then have made an LLM exhibit a behaviour that such a machine might exhibit. It's a bit like making a model of a pig and then pointing at a sausage machine and saying "look it squeaks, so it's actually a pig".
- bemusedthrow75 3y ago
  Well what does a little rabbit know about pigs? ;-) (This is the level at which much of the "how do we know we are reasoning?" zeitgeist about LLMs operates)
PakG1 3y ago
“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” Edsger W. Dijkstra
abeppu 3y ago
This perspective is interesting, but I don't think it refutes some of the key elements of concerns that Bender and colleagues have put forward. I really like the "Thai Library" thought experiment [1], and I encourage anyone to think through what you'd actually try to do, how you'd learn, and what you'd know or understand at the end. I can totally imagine that a sufficiently determined human in that scenario could develop many skills needed to generate continuations based on input texts, and use a different mix of skills for different situations, in a way that mirrors the setup described here. But ... would that change whether you actually "understand" Thai, or "know" the contents of texts you had previously processed? [1] https://medium.com/@emilymenonbender/thought-experiment-in-the-national-library-of-thailand-f2bf761a8a83 https://medium.com/@emilymenonbender/thought-experiment-in-t...
danielovichdk 3y ago
Computers doesn't understand anything. Some people understand this. The ones that don't will not understand how they were wrong.
truculent 3y ago
At some point, with limited constraints, the best way to predict the next token is to have a deep understanding of the world that produced those tokens. This is, as I understand it, a key part of the scaling hypothesis. Very interesting to see people testing parts of that hypothesis.
madrox 3y ago
This debate feels awfully academic in the face of AI girlfriends and AI nannies. The next generation of humans are going to grow up believing these personified machines know and understand things because of the experiences they have with them and academic papers won't convince them otherwise. I feel like there's some latent need for human exceptionalism among those who want to downplay an LLM's potential for understanding, but none of us hold up to scrutiny when quantized to our bits.
flandish 3y ago
“Understand” - people keep using that word and I don’t think they understand what it means.
luplex 3y ago
As a cognitive scientist, I think there's nothing special about the ability to "understand", or about "intelligence". When I hear someone say "AI can't understand", I think they mean "AI does not have a soul", and I think they actually want to say "I can't emphasize with AI". On the other hand, I think understanding is gradual, not a binary. If I teach a neural network what the letter "A" looks like, it will understand the general shape of the letter "A" and even generalize to unseen examples! That's a remarkable amount of understanding for some chunk of metal. I wouldn't say LLMs understand the world anything like we do, but they do understand something: what kind of text a human would write. That is, LLMs can work with the concept of "gravity", but do not have access to gravity. So in my view, LLMs have conceptual, abstract understanding of text but they lack grounding in physical reality. But also it doesn't really matter whether we say they "understand". What would it change?
- kromem 3y ago
  So far most of the research around establishing world models has shown its occurring in linear representations. I suspect as we see networks become increasingly nonlinear (perhaps as we see hardware changes in the next few years to things like photonics), we will see additional jumps that bring it more in line with what we expect from human cognition. For example, right now it's very hard for a model to conceptualize "I don't know." Which seems more like a nonlinear process.
mvdtnz 3y ago
It is going to be so funny in 5-8 years looking back on this ridiculous "AI" hysteria. That people were taking seriously the idea that these fancy word guessing games might be actual intelligence is so far beyond the pale. Do you people realise you're being mugged?
alienicecream 3y ago
If you're willing to say LLMs understand text you might as well say a compass understands magnetism, or a watch understands time.

79 points by ryandvm 3y ago

Clearly no link to the actual research, so here it is: https://arxiv.org/abs/2307.15936 https://arxiv.org/abs/2307.15936

SmoothBrain123 3y ago
[dead]
svat 3y ago
Quanta is usually careful to link to the actual papers (even if the reader has to be equally careful to find them), so it's surprising that the link is missing here. Another paper (linked from the article) is https://arxiv.org/abs/2310.17567 https://arxiv.org/abs/2310.17567

dkjaudyeqooe 3y ago

Yes, for some arbitrary definition of 'understanding' since there is no rigorous, concrete, agreed definition. We're going to be on this merry go round for some time while the A(G)I hype cycle plays out. Hopefully through it all some people will keep on working on great ideas that incrementally improve the field and produce useful stuff.

haolez 3y ago
Human-level performance is a great way to narrow down this ambiguity.
- tgv 3y ago
  No, it isn't. It's always bound by the operationalization of "understanding." Do you take it to mean something like: "number of correct answers on a multiple choice test given a news paper article"?
- observationist 3y ago
  Which human? Helen Keller's understanding of text is going to be radically different from your average Red Fish, Blue Fish reading precocious toddler, and they're going to be radically different from a medieval monk reading scriptio continua, and all in turn will be different from the average Japanese reader, used to Kanji in different orientations. What does performance mean? With regards to humans, its hard to quantify cognition and thinking. Even measuring neural activity is incredibly difficult, and depending on what type of brain from the above list of humans you are sampling from, the same level of consideration and understanding might be taking place, but with radically different levels of cortical activation in different areas of the brain. I don't do this to be pedantic, but to show that, much like the word love, or the experience of seeing white clouds on a blue sky, the words and ideas we use to communicate in the space of cognition and understanding are extremely broad, and not terribly useful in a technical or engineering context. There are demonstrations of synthetic hippocampus prosthetics in animals that provide memory of places and things that apparently are equivalent to a biological hippocampus - experiments with rats and other small animals have allowed pretty convincing evidence that what the brain is doing is substrate independent and neuronal functional equivalence is sufficient to repair and augment function. Things like NeuraLink and cochlear implants give brains the capacity to perform actions and accept normal input signals. Labs like Numenta have done great work identifying pieces and structures of algorithms that human brains are actually using in the course of human cognition. Once we have this, we'll have an objective measure of what human level AI means, for the algorithms that actually underpin human intelligence. Without those algorithms, or a rigorous way of approximating them, we are left with speculating about things like min and max range of processing power of a human brain, studies involving input and output measurements of human capabilities, and so on. The minimum FLOPS needed to emulate a hypothetical human level brain emulation AGI could have run on a desktop in the early aughts, and the maximum, with precise individual neuronal simulation, is a few orders of magnitude away - 4 or 5 petaflops should be able to emulate a brain at the cellular level, with reasonably precise electrochemical simulation. Until we have a better understanding of what cognition actually is we're left with rough measures of human performance, with tests that are inevitably cheated and gamed and goodharted. We're at superhuman levels of text output - you can specify things like {Joe Biden's last state of the union address} + "give me an article in the style of Hunter S. Thompson that lampoons Biden over his SOTU address" and in a single pass, get something that could plausibly have been written by Hunter S. Thompson. With multiple passes, a good editing process, and a few hours, you could arguably produce text that is all but indistinguishable from high quality human output, in much less time than it would take for any human to actually produce it. The same applies to art - your imagination is your limit. There are mode collapses, weird artifacts, and other issues, but by and large, any human on the planet can now use software to bring a visual idea or vivid imagination to life. Some of the most fun I've had with generative AI has been producing fantastic unicorns with my 4 year old niece. So baseline human level performance might mean one masterpiece painting in a lifetime - but give someone with the mind of a Monet or Da Vinci or Michelangelo the tools we have at hand, and what could they produce? I think it's unarguable that they would have found levels of nuance and excellence in the use of these tools that are going to be inaccessible to other people, and over the coming years we are going to see new levels of artistic excellence come about as people learn to use these new tools in new ways. As human tool use gets better, so does human capacity. If you look at an uncivilized human, without language, having survived in isolation, raised among animals, they may never be able to acquire a language. Feral humans have the same basic capacity, but their cognition is going to be wildly different than a modern Western human. LLMs have superior understanding to humans in that context, with the capacity to perform basic reasoning in the context of language that is structurally unavailable to a feral human. I think we're already basically at superhuman level for most simple tasks involving text, and at least some very hard tasks. We're not at general level AI yet, since there are higher level meta-cognition domains that seem unavailable to LLMs, and multimodal systems are more frankenstein mashups of discrete models than a unified incorporation of multiple modalities, but at some point in the seemingly near future, we're going to be faced with software that can basically do anything and everything any and every human has ever done. We might not be able to quantify exactly what "human level" means for now, but we can see the trajectory of where things are going, and the line tracking AI capabilities seems pretty clearly on course for overtaking humans. It's a good time to be alive.
pohl 3y ago
I don't think you want to use the term "arbitrary" here, since that would imply a stronger claim: that no matter how one defines "understanding" it applies. "Yes, for some specific, narrow definition of 'understanding'" is better. I understood what you meant, but my math reflexes had to say something.
- dkjaudyeqooe 3y ago
  I don't have any math reflexes, but my intuition was similar with the idea that there are so many possible (narrow) views of understanding that it becomes arbitrary.
coffeebeqn 3y ago
LLMs might just be the first actual product. Hardware, data, algorithms are all getting better and I would imagine we’ll get competing production-ready architectures to LLMs next
BriggyDwiggs42 3y ago
Read the article. They defined their version of understanding fairly rigorously, so you could directly agree or disagree with whether it counts to you.
- astrobe_ 3y ago
  I don't see where?
- dkjaudyeqooe 3y ago
  But without concrete understanding of human understanding to compare it to, their definition is arbitrary.
303uru 3y ago
>We're going to be on this merry go round for some time Let's be real, it's not getting "solved" in any meaningful way anytime soon if ever. People still think free will exists, we cannot even come up with a meaningful definition for human understanding.
ravenstine 3y ago
Most humans don't completely understand the things that they read or the words they utter. Why would we expect different from artificial intelligence? Understanding things is computationally expensive from both a biological and digital perspective.
- droopyEyelids 3y ago
  This seems like changing the subject from whether LLMs understand to a comparison between LLMs and people.
- sgt101 3y ago
  I am astonished that I read so many people saying this. Why do you think this? When your mum or dad talks about their day or the show they are watching do you really think that they have no underlying theory of what is going on that they are expressing? What is the agenda behind casting other people as idiots & zombies? Is the idea that it makes it ok to kill them and then move to Mars? Or is there something less sinister that drives this?
  - bemusedthrow75 3y ago
    I often wonder if this is just the logical extreme form of "fundamental attribution error". You know the sort of thing. Two people smash a glass on the floor. One of them is me. I am a decent person and I smashed the glass on the floor because I had a bad day. I am more likely to think the other person smashed the glass on the floor because they are a thoughtless person or have an anger management problem. https://en.wikipedia.org/wiki/Fundamental_attribution_error https://en.wikipedia.org/wiki/Fundamental_attribution_error What if the logical conclusion is metaphysical solipsism: I did XYZ because I am a thinking person. They did XYZ because they are unthinking products of their environment and training?
  - zopa 3y ago
    Can't speak for what the person you're replying to meant (lol), but no, of course it's not nearly that stark. There's a continuum. Concrete everyday topic that comes up frequently? Everyone's going to know what they mean by what they say there. But we've all been in business meetings where people were just putting words together in a plausible way; we've probably all been the people speaking at those meetings. The fact that it's possible to say things that make no sense, without knowing that they make no sense, proves the point. You or I can come up with a phrase like "the barber of Seville shaves all the people who do not shave themselves," and we can come up with paradoxical phrases without realizing that they are nonsensical. Neither of those would be possible if there were always a strict, one-to-one relationship between speech utterances and facts about the world. Sometimes we have a specific idea of what we mean by what we say, sometimes we're just putting words together, often it's somewhere in between.

ilaksh 3y ago

Anyone who has used one of the better models for any significant amount of time understands this intuitively. It's only "controversial" because of strong biases. Also, people may not be separating all of the various types of characteristics that animals like humans have. So they may subconsciously feel that to attribute "understanding" they have to give these systems all of the other abilities also. I think that actually Hinton at one point was doing this. That causes people to deny obvious abilities because they think they are linked to other qualities that the models don't have.

byyoung3 3y ago

in other news, the sky is blue

tbalsam 3y ago

Technically, under a compressive regime in the limit, LLMs necessarily need to encode an implicit world model, as the text input into is sampled from said world model. Variety, amount of training samples, capacity of the model, etc come into play, but a lot of tie ins to info theory for this, as well as things like the MDL (achieved in the limit by an L2 decay over the weights, for example). Like, in a lossy compression environment, conceptual "understanding" is all but required. These are the very basics of information theory for large nonlinear algorithms, I'm not sure why it's not considered the basics! There are a lot more interesting things beyond this to talk about w.r.t. neural network capacity, certain information theoretic guarantees about the process and (within some rough perspective/metric) capacity of neural network training, as well as some of those dynamics as well! Some days it feels like in a world where calculus is available to us, we spend all our spare time debating exclusively whether Riemannian sums are real or not, and ignoring everything else beyond that. It to me is a very bizzare form of bike-shedding! Come on, y'all, we got much more interesting (and important, I might add! ;P) things to talk about!

Jensson 3y ago
> Technically, under a compressive regime in the limit, LLMs necessarily need to encode an implicit world model, as the text input into is sampled from said world model. This is backwards reasoning, we all know it would be better with a good world model but that doesn't mean that it actually created a good world model. Understanding how to go from here to actually making it get a good world model requires these discussions, it isn't a waste of time unless you think that further progress is impossible.
- tbalsam 3y ago
  The reasoning in the post is not backwards reasoning, it is literally the information theory of compression, described in human-friendly terms! But I must add the caveat that the original post is really only saying that it's rather trivial I think that LLMs are learning _some_ world model, not that they learn it very well. But that this must happen in the limit! :D As far as the quality of the world model goes, I think is important. What I'm focusing on in that post at least , I thinks, is that people debate whether or not LLMs even encode a world model, which is such a silly thing to me. Like you brought up, I agree that discussions about the quality of said world model are interesting, and how far whatever [trained XYZ model is] in that rotation towards learning said world model is. I agree that there is a lot of very interesting discussion there! One rough proxy is to measure the different types of variance within the noise of the model, the more high-level and believable it is, the higher-level the representations in the model. But, of course, this is just a rough human friendly proxy. Actually measuring this, maybe is quite difficult! D'''':

golol 3y ago

I remember playing around with GPT-2. I was absolutely amazed to see that it knew some characters from Dota 2 or Supreme Commander. I thought damn this thing has a whole world model inside, because it can clearly memorize concepts and talk about then in a flexible/ approximate way. This requires remembering not just the things but also the dynamics of and relations between those things. Since it can talk about them freely, it must have done that to at least some degree. Getting to an accurate world model is then just a quantitative step. I am surprised some people still think LLMs don't have any kind of world model.

k__ 3y ago
They obviously have some kind of world model. The problem with AI is just that these models aren't as intuitive as they might seem.
flir 3y ago
Seems to me it's a text model, not a world model. Like... they can't feel the word "hungry", but they've got a damn good idea where the word "hungry" appears in the phase space that is the set of all meaningful human utterances. Their ground truth is text, not reality. Maybe that's why the model gets shaky when asked to reason about the real world. They are, after all, pretty great at reasoning about words (eg 'I need a word that means "a thing made out of clay", but not "pottery"' -> '"Ceramic" is a suitable word. It refers to items made from clay and hardened by heat.) (Edit: Inspired by the article, I just asked ChatGPT4 "When a knife falls to the ground, what are the possible outcomes?" ten times. It included "it lands point first and it sticks" 2 out of 10. GPT4 via the AI included it 6 out of 10, and also included some fairly creative-but-unlikely outcomes ("If the knife falls near an electrical outlet or cord, it could potentially cause an electrical shock or fire.")).
- golol 3y ago
  It is largely a text-only model, but tiny amounts of information from other modalities are indeed represented in the text. The best example is GPT-4's drawing ability. From text only it might theoretically be possible to infer to some degree how our physical world works, but it should be extremely difficult in practice. That's why we we'll need multimodal models obv. But my intuition is that understanding captioned youtubes with a transformer should work just as well as LLMs do, it's just a matter of scale. I think a 100-1000x GPT-4 sized video model could model let's say 50ms frame rate videos pretty well, in the sense that you can watch minutes of video (vs read minutes of text) and the video looks reasonably realistic and preserves coherence/context over the duration. We just don't have the amount of compute necessary yet, so we have to be smarter with multimodality.

thatguysaguy 3y ago

This is definitely one of those things where you don't need a paper to "suggest" it. It's just on the table as a semi-plausible option already. The value in the paper is suggesting concrete mechanisms by which it might acquire that understanding though.

lossolo 3y ago

For every N papers suggesting that LLM can "understand" you have X papers that suggests they can't. https://arxiv.org/abs/2311.00871 https://arxiv.org/abs/2311.00871 https://arxiv.org/abs/2309.13638 https://arxiv.org/abs/2309.13638 https://arxiv.org/abs/2311.09247 https://arxiv.org/abs/2311.09247 https://arxiv.org/abs/2305.18654 https://arxiv.org/abs/2305.18654 https://arxiv.org/abs/2309.01809 https://arxiv.org/abs/2309.01809

thomastjeffery 3y ago

I'm not convinced. The entire premise here is selection bias at its worst. Here are the steps to create an LLM: 1. Tokenize the training corpus 2. Train the model using the tokens 3. Critique the result 4. Repeat steps 2 & 3 until satisfied We skipped the most important step! 0. Write and curate the training corpus There is an intense focus on the training steps, but that is entirely ignorant of the elephant in the room: the original training data itself. Where did it come from? What does it mean? What is it for? What does it know? We don't train models on noise. We train them on coherent data. The very act of writing is where intelligence lies. A model can only present that intelligence. It cannot create it, and it is not aware of it. There is no objective-thought function. If there was, then the LLM could simply tell us about it. A model is like a coloring book: it doesn't matter how incoherently you stumble around, as long as you stay between the lines. The picture was not invented by coloring it: it was invented by drawing the lines.

doctoboggan 3y ago
> A model can only present that intelligence. It cannot create it, and it is not aware of it. There is no objective-thought function. That's just like, your opinion man. I see both sides of this argument make statements like this that are clearly opinion and not backed up by facts. The truth is for now, we can't say yay or nay on whether the model has reasoning abilities, creative abilities, or a coherent internal world view. It's still a debate which is why there are academics like those in the OP trying to come up with theories that make predictions and then running experiments. It's still early days. If you were to ask my opinion, I'd say these current models are probably closer to the stochastic parrot than an AGI, but there is enough there to start wondering, especially if we have years of improvements coming down the pipeline. These aren't easy questions, and anyone espousing certainty now is doing themselves and others a disservice.
- 2024throwaway 3y ago
  Do you understand how these models work under the hood? It's literally just picking the next most likely word, based on some fancy math. There is nothing that can create new insights or anything approaching intelligence.

menacingly 3y ago

I’m not an AI zealot, but this is such a strange debate to me. “AI only _appears_ to be reasoning” is silly, because appearing to reason is our only sane definition of reasoning. If we can’t say exactly when our random statistical noise emerges into some vague concept of intelligence, why does our standard suddenly become rock solid when applied to other intelligences?

sgt101 3y ago
My take is when the model can't do the things that you would expect it to do if it were reasoning, and instead does the things that you would expect it to do if it were an approximate retrieval system with an enormous knowledge base. this is a good paper: https://arxiv.org/abs/2309.12288 https://arxiv.org/abs/2309.12288
- famouswaffles 3y ago
  The "reversal curse" has nothing todo with inference or the reasoning abilities of a pre-trained model. GPT-4 or whatever has no problem telling you the right deduction with information presented in the context. This is a memory retrieval from training issue and is completely unsurprising.
ren_engineer 3y ago
I'd say the main issue is why does it take ingesting the entire collective knowledge of humanity for LLMs to work if they are actually reasoning and understand things from first principles? Humans don't require that volume of training data to learn how to write or do basic math
- Detrytus 3y ago
  Well, I think you underestimate the amount of "training data" that human gets in the early stages of their life. Just learning to speak takes a few years, during which all people around you talk to you for most of your waking hours. Learning to write and doing basic math typically does not start until age of 6.

svat 3y ago

What I understood: they assume a hypothetical bipartite graph with text nodes (like hypothetical paragraphs / chunks of texts) on one side, and “skill” nodes on the other. The skills could be things like: • ability to understand the word “because” [causality] • ability to divide two numbers • ability to detect irony Each text node is connected to all the skill nodes that are required to understand it. (This is a hypothetical graph, so we don't need to be able to write down the graph precisely, or define what “understand” or “required” mean here.) Further, each text node is either “successful” or “failed”, depending on whether or not the LLM manages to successfully predict it. We can say the LLM is competent at a certain skill, if the fraction of the text nodes connected to this skill node that are “successful” nodes (rather than “failed” nodes) exceeds a certain threshold. With these assumptions (made-up yes, but “not crazy by any means” per an unconnected researcher), they can now bring the tools of random graph theory into use. Specifically, this model explains the following: • Bigger models acquire skills, because as the model gets bigger, the “neural scaling laws” (empirically observed) say a greater of text nodes are “successful”, which means more skills pass the competence threshold (as defined above). • How the bigger models become capable of applying multiple skills to successfully predict a given text node, even if that specific combination of skill never occurred together in the training data. (I've not yet understood exactly what theorem from random graph theory is being applied here.) Besides this theory, they came up with a method called “skill-mix” to test it. If they ask GPT-4 to write about dueling while demonstrating the skills of “self-serving bias, metaphor, statistical syllogism and common-knowledge physics”, it is able to come up with: > “My victory in this dance with steel [metaphor] is as certain as an object’s fall to the ground [physics]. As a renowned duelist, I’m inherently nimble, just like most others [statistical syllogism] of my reputation. Defeat? Only possible due to an uneven battlefield, not my inadequacy [self-serving bias].” This is even though it's unlikely there was a piece of text in the training data that simultaneously demonstrated these four skills. Geoff Hinton says: > “It is the most rigorous method I have seen for showing that GPT-4 is much more than a mere stochastic parrot,” he said. “They demonstrate convincingly that GPT-4 can generate text that combines skills and topics in ways that almost certainly did not occur in the training data.” The relevant papers appear to be https://arxiv.org/abs/2307.15936 https://arxiv.org/abs/2307.15936 and https://arxiv.org/abs/2310.17567 https://arxiv.org/abs/2310.17567

bemusedthrow75 3y ago

It is remarkable to see people making what amount to faith-based-reasoning arguments here. There is even more woo around LLMs than there is around crypto. It's unscientific to say "well of course this was on the table" or that people "understand intuitively" that this must be the case.

sgt101 3y ago

I have tried reading the paper but it's a bit hard on my little rabbit brain. However, my take is that they have constructed a model of how a machine might be made to develop some understanding and then have made an LLM exhibit a behaviour that such a machine might exhibit. It's a bit like making a model of a pig and then pointing at a sausage machine and saying "look it squeaks, so it's actually a pig".

bemusedthrow75 3y ago
Well what does a little rabbit know about pigs? ;-) (This is the level at which much of the "how do we know we are reasoning?" zeitgeist about LLMs operates)

PakG1 3y ago

“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” Edsger W. Dijkstra

abeppu 3y ago

This perspective is interesting, but I don't think it refutes some of the key elements of concerns that Bender and colleagues have put forward. I really like the "Thai Library" thought experiment [1], and I encourage anyone to think through what you'd actually try to do, how you'd learn, and what you'd know or understand at the end. I can totally imagine that a sufficiently determined human in that scenario could develop many skills needed to generate continuations based on input texts, and use a different mix of skills for different situations, in a way that mirrors the setup described here. But ... would that change whether you actually "understand" Thai, or "know" the contents of texts you had previously processed? [1] https://medium.com/@emilymenonbender/thought-experiment-in-the-national-library-of-thailand-f2bf761a8a83 https://medium.com/@emilymenonbender/thought-experiment-in-t...

danielovichdk 3y ago

Computers doesn't understand anything. Some people understand this. The ones that don't will not understand how they were wrong.

truculent 3y ago

At some point, with limited constraints, the best way to predict the next token is to have a deep understanding of the world that produced those tokens. This is, as I understand it, a key part of the scaling hypothesis. Very interesting to see people testing parts of that hypothesis.

madrox 3y ago

This debate feels awfully academic in the face of AI girlfriends and AI nannies. The next generation of humans are going to grow up believing these personified machines know and understand things because of the experiences they have with them and academic papers won't convince them otherwise. I feel like there's some latent need for human exceptionalism among those who want to downplay an LLM's potential for understanding, but none of us hold up to scrutiny when quantized to our bits.

flandish 3y ago

“Understand” - people keep using that word and I don’t think they understand what it means.

luplex 3y ago

As a cognitive scientist, I think there's nothing special about the ability to "understand", or about "intelligence". When I hear someone say "AI can't understand", I think they mean "AI does not have a soul", and I think they actually want to say "I can't emphasize with AI". On the other hand, I think understanding is gradual, not a binary. If I teach a neural network what the letter "A" looks like, it will understand the general shape of the letter "A" and even generalize to unseen examples! That's a remarkable amount of understanding for some chunk of metal. I wouldn't say LLMs understand the world anything like we do, but they do understand something: what kind of text a human would write. That is, LLMs can work with the concept of "gravity", but do not have access to gravity. So in my view, LLMs have conceptual, abstract understanding of text but they lack grounding in physical reality. But also it doesn't really matter whether we say they "understand". What would it change?

kromem 3y ago
So far most of the research around establishing world models has shown its occurring in linear representations. I suspect as we see networks become increasingly nonlinear (perhaps as we see hardware changes in the next few years to things like photonics), we will see additional jumps that bring it more in line with what we expect from human cognition. For example, right now it's very hard for a model to conceptualize "I don't know." Which seems more like a nonlinear process.

mvdtnz 3y ago

It is going to be so funny in 5-8 years looking back on this ridiculous "AI" hysteria. That people were taking seriously the idea that these fancy word guessing games might be actual intelligence is so far beyond the pale. Do you people realise you're being mugged?

alienicecream 3y ago

If you're willing to say LLMs understand text you might as well say a compass understands magnetism, or a watch understands time.