40 gigabytes of text is about how much a person would say if they spoke nonstop for 8 hours a day, 7 days a week, for 250 years.
Preschoolers learn language much faster than that.
You can argue that vision transfers to language, but blind preschoolers also learn language much faster than that.
You can argue that hearing-bird-songs-or-whatever transfers to language but (1) really? (2) that doesn’t add much.
You can argue that the full data-heavy audio waveform (as opposed to just the words per se) makes the language learning task dramatically easier, but (1) that seems like the exact opposite of what we should expect, i.e. I’d assume that one of the hardest computational parts of language acquisition is learning to ignore the countless aspects of the auditory waveform over which the linguistic content is invariant, (2) deaf-blind preschoolers also learn language much faster than that.
In model-size terms, how large is a preschooler?
I’m not sure why you’re bringing up parameter count in a post about data efficiency. You can argue that LLMs are more data-efficient than brains, or you can argue that brains are more data-efficient than LLMs because brains have a higher parameter count, but it’s weird how you seem to be arguing both at once.
(My own opinion, stated without justification, is that brains are more data-efficient than LLMs for reasons unrelated to “higher parameter count”.)
Maybe they are? It’s unclear how strong that is. But I think it’s more relevant that: LLMs are optimized, in particular, for being able to learn language.
I don’t have a strong belief about any of this right now, but flagging a lot of the brain’s data is kinesthetic/proprioception. (also, a ton of the brain’s data is presumably “managing stuff inside the body”, I’m not sure whether that is relevant to learning or not)
I’d guess there’s some basic “ability to learn and combine concepts, at all” which is a key building block for visually recognizing mama, papa, recognizing the word “mama” and “pape”, one-shotting the concept of “dog”, learning to count, etc. How much data do humans take before they grok that?
Do there exist blind + deaf + nonproprioceptive people who seem smart (or, survive at all?)
Helen Keller looks like she lost sight/hearing at 19 months old which seems after my hypothesized “generalized concept-grokking” would form.
a lot of the brain’s data is kinesthetic/proprioception
Sure! A proper accounting of “how many sensory bits” should certainly include all the senses, including proprioception, tactile, interoception, and so on. (But those are OOMs fewer bits than vision, if I recall correctly.)
Do there exist blind + deaf + nonproprioceptive people who seem smart (or, survive at all?)
Lacking proprioception is super-rare. The most famous example is Ian Waterman, who lost all proprioception at age 19 (they think he got a random minor infection, and by terrible luck, the immune system response happened to also false-positive upon his own proprioceptive nerves, or something like that). There was a neat little BBC documentary about him.
On a quick search, it seems like a congenital absence of proprioception is something that’s happened in like two documented cases ever, and those two were not blind.
Putting proprioception aside: I think congenital deaf-blindness is a thing (I heard that congenital rubella syndrome can cause it) and such people can indeed acquire language in childhood.
Putting proprioception aside: I think congenital deaf-blindness is a thing (I heard that congenital rubella syndrome can cause it) and such people can indeed acquire language in childhood.
The current theory I want to distinguish is “children get a fuck-ton-of-bits, in the first ~2 years, of some kind, which is when they get a generalized-concept-wrangling skill, which makes all subsequent learning much easier”
from
“children do that, but, don’t require that many bits” (either because it comes hardwired somehow, or because they are just very good at learning)
from
“idk everything else.”
For purposes of figuring out “apart from not having continual learning, how much less sample-efficient are LLMs than humans”.
The “deaf/blind people still learn” thing does nudge me towards #2 over #1, but seems plausible that there’s still a ton of proprioception bits that contribute substantively.
Also, I vaguely recall people missing a sense end up repurposing the brain that usually goes towards that sense to doing other stuff. Might be something like “the congenital deaf/blind children turn the spare brainpower towards eeking out more effective learning from proprioception.”
Children don’t learn language from reading a bunch of text. They learn it from people interactively teaching them, pointing at things, explaining words on demand, etc. Non-textual information like intonation and gestures is certainly useful for that process.
Let’s try an (thought) experiment where we lock a child in a small room and pass in sheets of paper covered in English text. Occasionally he can go to the backyard and experience nature. How long before he will read and write about nature?
Yet, an LLM can do this, even without direct experience.
I think that western culture vastly overstates the importance of deliberately teaching things to young children. There are cultures where “everyone waits until the child can talk sensibly before acknowledging its existence”, and the kids still learn to talk just fine. See Notes on “The Anthropology of Childhood”.
I don’t think your thought experiment is quite right: if I understand correctly, LLMs get pretraining data of different modalities coupled together; they don’t have to discover on their own a correspondence between a text corpus and a completely disconnected vision corpus.
As of today, I strongly believe that there is no known learning algorithm that can start from random initialization, spend the equivalent of 2 waking years roaming around and doing stuff in various virtual environments like VR & Minecraft, occasionally happening upon people who are talking to each other about whatever, and/or watching YouTube videos, such that the model winds up at the end with the linguistic and other competences of a human 4yo. You sure can’t train an LLM that way. How would it ever take intelligent action?
Maybe you’ll say: Well, we raise humans in an appropriate environment built for humans, and we “raise” LLMs in an appropriate environment built for LLMs. Each would do badly if put in the shoes of the other. And my response would be: Yeah, sure, whatever, I think it’s obvious that LLMs can do things that humans can’t and vice-versa, and I’d like to avoid a getting into a vague and pointless argument about which of them is “more impressive”. Those arguments tend to be divorced from any concrete question and unhelpful for understanding.
I wouldn’t say it’s about the “appropriate environment“ being different, I would say LLMs can actually learn language in a way that humans probably would not be able to do from scratch.
Do you think that a blind child who everyone ignored would learn to speak fluently? Intuitively I doubt it, though I’m not that confident.
But that is how LLMs learned to speak, until the introduction of multi-modal training. I believe that GPT-3 was trained on text only. The problem isn’t just establishing the right correspondences, it is that establishing correspondences would make learning language much easier by breaking harder-to-learn concepts down into easily grounded ones.
I also think that the evidence you’re raising about people in other cultures giving children less attention is pretty weak. It seems to mean relative to western parenting, versus “not at all” or “sparsely” in some statistical sense. Also I don’t want to dig through that entire post, when it seems maybe one paragraph is relevant.
It really doesn’t appear to me that there’s a straightforward comparison for language learning efficiency of LLMs versus humans, and probably both are better in certain senses.
learning problems can become drastically easier or harder based on representation or interaction protocol. Generally, I have the impression that you are too cavalier about brushing things under the rug as constant factor differences or lower order corrections that seem potentially quite important to me. i often find your conclusions plausible but your arguments not very convincing. perhaps in this case also, there is some underlying conclusion-generator that is stronger than the arguments given.
I would say LLMs can actually learn language in a way that humans probably would not be able to do from scratch.
Yeah I probably agree.
It really doesn’t appear to me that there’s a straightforward comparison for language learning efficiency of LLMs versus humans, and probably both are better in certain senses.
I agree here too. Like I said, it’s hard to pin down an apples-to-apples comparison between human learning and LLM pretraining, because human learning is just a totally different thing from LLM pretraining. Best to ask more specific questions.
(But I stand by my belief that the gist of the OP is mostly wrong and misleading.)
I also think that the evidence you’re raising about people in other cultures giving children less attention is pretty weak. It seems to mean relative to western parenting, versus “not at all” or “sparsely” in some statistical sense. Also I don’t want to dig through that entire post, when it seems maybe one paragraph is relevant.
It seems like (in those cultures) 1yo toddlers might often be supervised by somewhat-older siblings, peers, etc., and I’m a bit unclear on what kinds of interactions they’d have. (I doubt it’s “deliberately teaching”, except for things like yelling at toddlers not to wander off in dangerous areas (cf. the “Chewong” excerpt). But maybe there’s other kinds of playful interactions between toddlers and older kids?)
Anyway, I put in that summary quote (“everyone waits until the child can talk sensibly before acknowledging its existence”) because it’s from the author and I figure he would know. His tone throughout the book is pretty careful as opposed to sensationalizing and exaggerating, IMO.
(I read the whole book, but that was a decade ago so my memory is a bit hazy. Julia’s excerpts are good.)
Do you think that a blind child who everyone ignored would learn to speak fluently? Intuitively I doubt it, though I’m not that confident.
It’s a bit pointless to argue since we don’t know the answer. Oh well: My guess is yes (assuming they survive), because I think little blind kids, just like little sighted kids, are generally capable of learning how the world works, and what’s what, and learning intuitive physics and intuitive biology and so on. It’s restricted to tactile and auditory for the blind kids, and for all I know it might take somewhat longer, but they still get there. And so they’re still in a position to connect overheard words to relevant objects and situations, just like the sighted kids are. (Again, maybe it would take somewhat longer, but I’d guess not overwhelmingly longer.)
If people kept ignoring the kid, they would sure have a lot of unanswered questions, e.g. about how to identify some plant by touch if everyone else is talking about its color. But that would be after the kid is fluently talking about more basic things like rain and beds and inside and mud and fighting etc., that blind kids can easily develop an intuitive understanding of by autonomous nonverbal exploration and sensing.
What about testing the ability of neuralese/brainlike “LLMs” to learn? Current AIs, even including the presumably looped transformers like GPT-6 Astra, are CoT-based in the same sense as described in the section “Neuralese recurrence and memory” of AI-2027. I conjecture that a neuralese/brainlike LLM learns faster by virtue of being free from such constraints. The alternate explanation is that the human’s genes bias the human brain towards learning some tasks, which I find unlikely since the human genes are hard to change and carry around 4 gigabits of information.
The text to speech conversion is a good point which I should have considered. I would still argue that humans have an easier time because they bootstrap a world model and then learn language on top of it, but this leans very far into conjecture.
I also still think it’s incredible how effective transformers are considering their tiny size, but yes, that is unrelated to data efficiency. I framed the post as about data inefficiency because that is what most of the bad hominid metaphors focus on. My main goal was to attack those metaphors in general, because suggesting model capabilities are limited using hominid metaphors has rarely done any good for anyone.
If it’s true that “humans have an easier time because they bootstrap a world model and then learn language on top of it”, why would you contextualize that as, like, “…and so humans aren’t really data-efficient, that doesn’t count, it’s cheating”, rather than as “…and perhaps this is part of the explanation of how humans manage to be so data-efficient”?
suggesting model capabilities are limited using hominid metaphors has rarely done any good for anyone
I think it’s valid and useful for people to point out things that human brains can do that AI engineers can’t (yet) do, because human brains provide a convenient existence proof that the thing in possible at all. For example, (copying from a different comment) as of today, I strongly believe that there is no known learning algorithm that can start from random initialization, spend the equivalent of 2 waking years roaming around and doing stuff in various virtual environments like VR & Minecraft, occasionally happening upon people who are talking to each other about whatever, and/or watching YouTube videos, such that the model winds up at the end with the linguistic and other competences of a human 4yo. You sure can’t train an LLM that way. LLMs need (pre)training data, humans instead need a training environment. It’s very different.
But when (as above) someone points out a thing that human brains can do that AI engineers can’t (yet) do, I feel like you’re not treating that as an neutral observation about AI-algorithm-possibility-space, but rather as a front in the war over what we actually care about, which is whether we should be more impressed by human brains versus LLMs. And you’re fighting for the yay-LLMs side and they’re fighting for the yay-brains side. And indeed fighting this war is a popular activity on both sides. But that’s a waste of time, because this is a “meaningless argument” in the Scott Alexander sense. Really we should all be very impressed by both human brains and LLMs in their own ways, and then we should move on to arguing about more concrete and specific things.
On further reflection, the point I was originally trying to make with this post is that text token language modeling is more “data efficient” than objectives using world models/games/video data. My basis for this is that language modeling enabled the scaling era we are now in and still remains the only method with clear potential of achieving human-level intelligence. Furthermore it shows evidence of being able to adapt to robotic control and visual design and other non-text native domains better than trying to bootstrap on those modalities.
It’s still possible that language modeling hits a wall, and that something more “human” ends up being more efficient for achieving AGI.
I think that my favorable comparisons of LLMs to humans were initially intended for rhetorical effect, but at some point I think I let myself slip into defending them as a position in a meaningless argument.
I disagree, see some relevant discussion here (me) and here (Dwarkesh).
40 gigabytes of text is about how much a person would say if they spoke nonstop for 8 hours a day, 7 days a week, for 250 years.
Preschoolers learn language much faster than that.
You can argue that vision transfers to language, but blind preschoolers also learn language much faster than that.
You can argue that hearing-bird-songs-or-whatever transfers to language but (1) really? (2) that doesn’t add much.
You can argue that the full data-heavy audio waveform (as opposed to just the words per se) makes the language learning task dramatically easier, but (1) that seems like the exact opposite of what we should expect, i.e. I’d assume that one of the hardest computational parts of language acquisition is learning to ignore the countless aspects of the auditory waveform over which the linguistic content is invariant, (2) deaf-blind preschoolers also learn language much faster than that.
I’m not sure why you’re bringing up parameter count in a post about data efficiency. You can argue that LLMs are more data-efficient than brains, or you can argue that brains are more data-efficient than LLMs because brains have a higher parameter count, but it’s weird how you seem to be arguing both at once.
(My own opinion, stated without justification, is that brains are more data-efficient than LLMs for reasons unrelated to “higher parameter count”.)
Language is optimized for being easy for humans to learn. You can’t make any conclusions about general data-efficiency of brains from that.
Maybe they are? It’s unclear how strong that is. But I think it’s more relevant that: LLMs are optimized, in particular, for being able to learn language.
Wait why not “brains are optimized, in particular, for being able to learn language”?
Language is a late addon to an animal brain system and still not most of what humans do. It’s the core of what LLMs do and are built for.
Because language changes much faster than brains.
Mmm, hadn’t read your reply on Dissolving the Deep Learning Sample Efficiency Gap before. I’d been wondering about this.
The thing about deaf people is quite interesting.
I don’t have a strong belief about any of this right now, but flagging a lot of the brain’s data is kinesthetic/proprioception. (also, a ton of the brain’s data is presumably “managing stuff inside the body”, I’m not sure whether that is relevant to learning or not)
I’d guess there’s some basic “ability to learn and combine concepts, at all” which is a key building block for visually recognizing mama, papa, recognizing the word “mama” and “pape”, one-shotting the concept of “dog”, learning to count, etc. How much data do humans take before they grok that?
Do there exist blind + deaf + nonproprioceptive people who seem smart (or, survive at all?)
Helen Keller looks like she lost sight/hearing at 19 months old which seems after my hypothesized “generalized concept-grokking” would form.
Sure! A proper accounting of “how many sensory bits” should certainly include all the senses, including proprioception, tactile, interoception, and so on. (But those are OOMs fewer bits than vision, if I recall correctly.)
Lacking proprioception is super-rare. The most famous example is Ian Waterman, who lost all proprioception at age 19 (they think he got a random minor infection, and by terrible luck, the immune system response happened to also false-positive upon his own proprioceptive nerves, or something like that). There was a neat little BBC documentary about him.
On a quick search, it seems like a congenital absence of proprioception is something that’s happened in like two documented cases ever, and those two were not blind.
Putting proprioception aside: I think congenital deaf-blindness is a thing (I heard that congenital rubella syndrome can cause it) and such people can indeed acquire language in childhood.
The current theory I want to distinguish is “children get a fuck-ton-of-bits, in the first ~2 years, of some kind, which is when they get a generalized-concept-wrangling skill, which makes all subsequent learning much easier”
from
“children do that, but, don’t require that many bits” (either because it comes hardwired somehow, or because they are just very good at learning)
from
“idk everything else.”
For purposes of figuring out “apart from not having continual learning, how much less sample-efficient are LLMs than humans”.
The “deaf/blind people still learn” thing does nudge me towards #2 over #1, but seems plausible that there’s still a ton of proprioception bits that contribute substantively.
Also, I vaguely recall people missing a sense end up repurposing the brain that usually goes towards that sense to doing other stuff. Might be something like “the congenital deaf/blind children turn the spare brainpower towards eeking out more effective learning from proprioception.”
(Not sure if this is framed right)
Children don’t learn language from reading a bunch of text. They learn it from people interactively teaching them, pointing at things, explaining words on demand, etc. Non-textual information like intonation and gestures is certainly useful for that process.
Let’s try an (thought) experiment where we lock a child in a small room and pass in sheets of paper covered in English text. Occasionally he can go to the backyard and experience nature. How long before he will read and write about nature?
Yet, an LLM can do this, even without direct experience.
I think that western culture vastly overstates the importance of deliberately teaching things to young children. There are cultures where “everyone waits until the child can talk sensibly before acknowledging its existence”, and the kids still learn to talk just fine. See Notes on “The Anthropology of Childhood”.
I don’t think your thought experiment is quite right: if I understand correctly, LLMs get pretraining data of different modalities coupled together; they don’t have to discover on their own a correspondence between a text corpus and a completely disconnected vision corpus.
Anyway, it’s definitely hard to pin down an apples-to-apples comparison between human learning and LLM pretraining, because human learning is just a totally different thing from LLM pretraining. See “LLM pretraining magically transmutes observations into behavior, in a way that is profoundly disanalogous to how brains work”. LLMs have pretraining data, humans are more like RL agents in that there’s a training environment. A fun example is Nicaraguan Sign Language, or pidgins turning into creoles, where a group of kids who have never seen a grammatical language in their life simply create one. What would be the “sample efficiency” of that? They’re learning grammatical language from exactly zero samples!
As of today, I strongly believe that there is no known learning algorithm that can start from random initialization, spend the equivalent of 2 waking years roaming around and doing stuff in various virtual environments like VR & Minecraft, occasionally happening upon people who are talking to each other about whatever, and/or watching YouTube videos, such that the model winds up at the end with the linguistic and other competences of a human 4yo. You sure can’t train an LLM that way. How would it ever take intelligent action?
Maybe you’ll say: Well, we raise humans in an appropriate environment built for humans, and we “raise” LLMs in an appropriate environment built for LLMs. Each would do badly if put in the shoes of the other. And my response would be: Yeah, sure, whatever, I think it’s obvious that LLMs can do things that humans can’t and vice-versa, and I’d like to avoid a getting into a vague and pointless argument about which of them is “more impressive”. Those arguments tend to be divorced from any concrete question and unhelpful for understanding.
I wouldn’t say it’s about the “appropriate environment“ being different, I would say LLMs can actually learn language in a way that humans probably would not be able to do from scratch.
Do you think that a blind child who everyone ignored would learn to speak fluently? Intuitively I doubt it, though I’m not that confident.
But that is how LLMs learned to speak, until the introduction of multi-modal training. I believe that GPT-3 was trained on text only. The problem isn’t just establishing the right correspondences, it is that establishing correspondences would make learning language much easier by breaking harder-to-learn concepts down into easily grounded ones.
I also think that the evidence you’re raising about people in other cultures giving children less attention is pretty weak. It seems to mean relative to western parenting, versus “not at all” or “sparsely” in some statistical sense. Also I don’t want to dig through that entire post, when it seems maybe one paragraph is relevant.
It really doesn’t appear to me that there’s a straightforward comparison for language learning efficiency of LLMs versus humans, and probably both are better in certain senses.
learning problems can become drastically easier or harder based on representation or interaction protocol. Generally, I have the impression that you are too cavalier about brushing things under the rug as constant factor differences or lower order corrections that seem potentially quite important to me. i often find your conclusions plausible but your arguments not very convincing. perhaps in this case also, there is some underlying conclusion-generator that is stronger than the arguments given.
Yeah I probably agree.
I agree here too. Like I said, it’s hard to pin down an apples-to-apples comparison between human learning and LLM pretraining, because human learning is just a totally different thing from LLM pretraining. Best to ask more specific questions.
(But I stand by my belief that the gist of the OP is mostly wrong and misleading.)
Yeah my claim was “western culture vastly overstates…” which I stand by. The relevant sections of Notes on “The Anthropology of Childhood” include “How do children learn?” and “How children are expected to speak”.
It seems like (in those cultures) 1yo toddlers might often be supervised by somewhat-older siblings, peers, etc., and I’m a bit unclear on what kinds of interactions they’d have. (I doubt it’s “deliberately teaching”, except for things like yelling at toddlers not to wander off in dangerous areas (cf. the “Chewong” excerpt). But maybe there’s other kinds of playful interactions between toddlers and older kids?)
Anyway, I put in that summary quote (“everyone waits until the child can talk sensibly before acknowledging its existence”) because it’s from the author and I figure he would know. His tone throughout the book is pretty careful as opposed to sensationalizing and exaggerating, IMO.
(I read the whole book, but that was a decade ago so my memory is a bit hazy. Julia’s excerpts are good.)
It’s a bit pointless to argue since we don’t know the answer. Oh well: My guess is yes (assuming they survive), because I think little blind kids, just like little sighted kids, are generally capable of learning how the world works, and what’s what, and learning intuitive physics and intuitive biology and so on. It’s restricted to tactile and auditory for the blind kids, and for all I know it might take somewhat longer, but they still get there. And so they’re still in a position to connect overheard words to relevant objects and situations, just like the sighted kids are. (Again, maybe it would take somewhat longer, but I’d guess not overwhelmingly longer.)
If people kept ignoring the kid, they would sure have a lot of unanswered questions, e.g. about how to identify some plant by touch if everyone else is talking about its color. But that would be after the kid is fluently talking about more basic things like rain and beds and inside and mud and fighting etc., that blind kids can easily develop an intuitive understanding of by autonomous nonverbal exploration and sensing.
What about testing the ability of neuralese/brainlike “LLMs” to learn? Current AIs, even including the presumably looped transformers like GPT-6 Astra, are CoT-based in the same sense as described in the section “Neuralese recurrence and memory” of AI-2027. I conjecture that a neuralese/brainlike LLM learns faster by virtue of being free from such constraints. The alternate explanation is that the human’s genes bias the human brain towards learning some tasks, which I find unlikely since the human genes are hard to change and carry around 4 gigabits of information.
The text to speech conversion is a good point which I should have considered. I would still argue that humans have an easier time because they bootstrap a world model and then learn language on top of it, but this leans very far into conjecture.
I also still think it’s incredible how effective transformers are considering their tiny size, but yes, that is unrelated to data efficiency. I framed the post as about data inefficiency because that is what most of the bad hominid metaphors focus on. My main goal was to attack those metaphors in general, because suggesting model capabilities are limited using hominid metaphors has rarely done any good for anyone.
Thanks!
If it’s true that “humans have an easier time because they bootstrap a world model and then learn language on top of it”, why would you contextualize that as, like, “…and so humans aren’t really data-efficient, that doesn’t count, it’s cheating”, rather than as “…and perhaps this is part of the explanation of how humans manage to be so data-efficient”?
I think it’s valid and useful for people to point out things that human brains can do that AI engineers can’t (yet) do, because human brains provide a convenient existence proof that the thing in possible at all. For example, (copying from a different comment) as of today, I strongly believe that there is no known learning algorithm that can start from random initialization, spend the equivalent of 2 waking years roaming around and doing stuff in various virtual environments like VR & Minecraft, occasionally happening upon people who are talking to each other about whatever, and/or watching YouTube videos, such that the model winds up at the end with the linguistic and other competences of a human 4yo. You sure can’t train an LLM that way. LLMs need (pre)training data, humans instead need a training environment. It’s very different.
But when (as above) someone points out a thing that human brains can do that AI engineers can’t (yet) do, I feel like you’re not treating that as an neutral observation about AI-algorithm-possibility-space, but rather as a front in the war over what we actually care about, which is whether we should be more impressed by human brains versus LLMs. And you’re fighting for the yay-LLMs side and they’re fighting for the yay-brains side. And indeed fighting this war is a popular activity on both sides. But that’s a waste of time, because this is a “meaningless argument” in the Scott Alexander sense. Really we should all be very impressed by both human brains and LLMs in their own ways, and then we should move on to arguing about more concrete and specific things.
On further reflection, the point I was originally trying to make with this post is that text token language modeling is more “data efficient” than objectives using world models/games/video data. My basis for this is that language modeling enabled the scaling era we are now in and still remains the only method with clear potential of achieving human-level intelligence. Furthermore it shows evidence of being able to adapt to robotic control and visual design and other non-text native domains better than trying to bootstrap on those modalities.
It’s still possible that language modeling hits a wall, and that something more “human” ends up being more efficient for achieving AGI.
I think that my favorable comparisons of LLMs to humans were initially intended for rhetorical effect, but at some point I think I let myself slip into defending them as a position in a meaningless argument.
Would you happen to have any writings that gesture at your thinking on this, out of curiosity?