10 ms·

Comparing humans, GPT-4, and GPT-4V on abstraction and reasoning tasks

217 points by mpweiher 3y ago

stillsut 3y ago
I've actually been working on a similar eval task that uses grids of symbols to evaluate LLM's reasoning ability around the game Wordle. My initial results are: - Success: ChatGPT-4-November, Phind-November - Failures (via hallucination): GPT-3.5, Mistral-7b, Llama2-13b Interestingly, ChatGPT-4 wrote a short python script (unprompted) to develop a pseudo-CoT type generation and answered correctly. Phind, which also answered correctly, is also a code-based/code-enabled LLM. Non-code-based-LLM's appear to reason cogently when describing their games - what moves are illegal, what the best guess in certain conditions are, etc - but their actual answer, even in multiple choice is comically wrong. These are very initial results but my tentative hypothesis is symbolic logical reasoning - string manipulation / function calling type tasks - are greatly enhanced by having a code modality. How to switch between natural language and code-generation may offer a strong boost on tasks like these.
tovej 3y ago
Conclusion is obvious, but the paper is still probably necessary. Of course LLM's can't reason. They pattern match answers to previously asked questions, and humans will read the text as a reasonable answer because we assign meaning to it, but there is simply no way an LLM could use a "mental model" to "reason" about a problem other than constructing sentences out of probable matches it's been trained on.
- mr_toad 3y ago
  > Of course LLM's can't reason. They pattern match You’re assuming that reasoning isn’t just pattern matching.
- sfn42 3y ago
  Love how the ai hypebros always downvote answers like this. Everyone's just insisting that LLMs are AGI just because they kinda seem that way.
  - viraptor 3y ago
    Ftr, I mostly agree with the comment but downvoted it for the low quality "of course that's the answer, obviously" padded with some unhelpful tropes about trained pattern matching. (Let us know when you can prove that human communication is not just trained pattern matching)
    - tovej 3y ago
      Ok, I'll bite. I personally don't think human communication is simply pattern matching (e.g. we have drives, motives, relationships that we want to tell each other, these are internal systems, outside the visible band of communication). That's a little off topic though, my post was not about communication, but reasoning. Pattern matching in LLMs is specifically at the language level, as in "predict the next token in a stream", I should have probably specified this bit. You seem to refer to a more general kind of pattern matching when you say, where we're looking at a complicated structural model within which humans reason about the behavior of different system components and their interactions. There is no way to form an internal representation of such systems within an LLM, which is why I said "of course" it can't reason. An LLM is not designed with this capability in mind. If all you have is language, and predictive text, there is no way to use this to simulate and predict the rules in a higher-dimensional space than "stream of tokens", or let's put it another way: there would need to be an extremely high correlation between the pattern in the problem space and in language space for the likelihood of the model guessing right to be anywhere near human levels.
      - stevenhuang 3y ago
        > There is no way to form an internal representation of such systems within an LLM, which is why I said "of course" it can't reason. An LLM is not designed with this capability in mind. You claim this as if it is a foregone conclusion yet it is anything but. This is an active area of study, even within human cognition https://en.m.wikipedia.org/wiki/Predictive_coding https://en.m.wikipedia.org/wiki/Predictive_coding A complex system clearly does not need to be designed to evoke some behavior. Just look to the concept of emergent phenomenon. See evolution through natural selection as an obvious counter example. Evolution selects for fitness and not reasoning ability. Yet here we are. It is not known if the same isn't happening now for the transformer model. LLMs are universal function approximators. If reasoning ability is at all representable through such a process, it is possible for LLMs to find it. https://en.m.wikipedia.org/wiki/Universal_approximation_theorem https://en.m.wikipedia.org/wiki/Universal_approximation_theo...
- koe123 3y ago
  On the other hand, can we conclusively say that humans aren’t really advanced biological stochastic parrots?
- naasking 3y ago
  > Of course LLM's can't reason That they are not effective at some forms of reasoning does not entail they can't reason.
- mcswell 3y ago
  The conclusion may be obvious to you and me (although it's hard to know for certain, since these available LLMs are black boxes). But it's definitely not obvious to everyone. There are plenty of people saying this is the dawn of AGI, or that we're a few short steps from AGI. Whereas people like Gary Marcus (who knows tons more than I do) says LLMs are going off in the wrong direction.
  - visarga 3y ago
    Yes, LLMs can't reason 100% correctly, but neither do humans. We can often reason correctly, but it's not always the case. Even reasoning, fundamental as it is, comes from feedback. Feedback from our actions teaches us how to reason. Learning from feedback is more general than reasoning - AI agents can definitely learn this way too, if they have enough freedom to explore. But you can't do it with supervised training sets alone. You need to put language models into agents in environments, give them goals and rewards. Then they can make their own training data and mistakes, build up their own experience. Can't teach an AI based on how people make mistakes, it needs to fix its own mistakes, but that means to deploy it in the wild, where there are consequences for errors. If you remember, DeepMind first tried to train a Go model on human game play, but it was limited. Then they started from scratch, and learning from feedback alone they surpassed human level, even though the feedback was one single bit of information at the end of a whole self-play game. And it had no pre-training prior to learning Go, unlike human players.
    - somewhereoutth 3y ago
      That presupposes that language encodes the world we live in completely, whereas in fact language is meaningless without the shared context of reality. 'up' and 'down' are only meaningful to intelligence that can experience space. Essentially LLMs just are oracles for the shadows on Plato's cave.
      - visarga 3y ago
        Do a reverse Chinese room experiment - remove from a human all the abilities multi-modal LLMs gain after training on human media. What's left? Homo ferus. Most of our intelligence is encoded in the environment and language, it's a collective process, not an individual one. We're collectively, not individually, very smart.
baxtr 3y ago
Could a human and a LLM submit a summary of the paper so we can compare?
- j2kun 3y ago
  That's what an abstract is for: > Our experimental results support the conclusion that neither version of GPT-4 has developed robust abstraction abilities at humanlike levels.
- z7 3y ago
  >The paper investigates the abstract reasoning abilities of text-only and multimodal versions of GPT-4 using the ConceptARC benchmark, concluding that neither version of GPT-4 has developed robust abstraction abilities at humanlike levels.
cs702 3y ago
Interesting. Immediate thoughts and questions: * How would human beings perform on the text-only version of the tasks given to GPT-4? * How would human beings perform if each grid is shown on its own, making it impossible to perform side-by-side visual comparisons? * How would human beings perform if each grid is shown on its own only once, making it impossible to perform any back-and-forth comparisons? * How could we give LLMs the ability to "pay attention" to different parts of images, as needed, so they can make back-and-forth comparisons between parts of different images to solve these kinds of visual reasoning tasks?
- mistermann 3y ago
  * How would human beings perform if they didn't know they were being tested (ie: if in the same mode they are in when writing comments on the internet)? * How would human beings perform if the questions are based on culture war topics, which tend to invoke System 1 intuitive/emotional thinking?
- QuadmasterXLII 3y ago
  > How could we give LLMs the ability to "pay attention" to different parts of images, as needed, so they can make back-and-forth comparisons between parts of different images to solve these kinds of visual reasoning tasks? I’ve got good news
  - oefnak 3y ago
    What is it?
  - phh 3y ago
    It's even all we need
- YetAnotherNick 3y ago
  Also I want to know how would know how much gain could be made by optimizing the prompt for GPT and by including things like CoT. Current version of the prompt is pretty bad both for humans and AI.
air7 3y ago
Thank god...
firtoz 3y ago
The best thing from research like this is that they allow new models to be built, or improvements on the existing ones that can lead them to pass these evaluations.
- lewhoo 3y ago
  Yeah but the question remains are they passing because of a new obtained/emerged quality or are they passing because they were taught specifically to how to pass or is there even a difference at this point.
  - firtoz 3y ago
    Well, while training AI we make sure that we include enough training for them to pass some evaluation, but then test them on things that were not in the training data to make sure that they are not "over fitted".
    - viraptor 3y ago
      I don't think parent meant training specifically on that test data, but rather in that kind of task. Think more "LLM trained only on images of art" doing badly on this test, but then "LLM trained on images of art and IQ puzzles" doing better on this test. It's kind of like asking what even is IQ if you can learn how to solve Mensa puzzles and improve your score. Does it mean you're more intelligent?
      - firtoz 3y ago
        Oh, I see. One guess for GPT4 is that it's an amalgamation of models. So you would have various models trained more specifically like the ones you mentioned, and you ask them all to start answering the query, then you choose which one produces better results, and then you present that to the user. Alternatively you can have a decider model that knows what kind of queries goes to what specialised one, then have that as a "hidden layer", whether it be in the application level or "neuron layers".
krona 3y ago
This paper evaluates performance compared to a 'human' which presumably is an average adult human without cognitive impairment. I had to dig in to the references: In the first batch of participants collected via Amazon Mechanical Turk, each received 11 problems (this batch also only had two “minimal Problems,” as opposed to three such problems for everyone else). However, preliminary data examination showed that some participants did not fully follow the study instructions and had to be excluded (see Section 5.2). In response, we made the screening criteria more strict (requiring a Master Worker qualification, 99% of HITs approved with at least 2000 HIT history, as opposed to 95% approval requirement in the first batch). Participants in all but the first batch were paid $10 upon completing the experiment. Participants in the first batch were paid $5. In all batches, the median pay-per-hour exceeded the U.S. minimal wage. (Arseny Moskvichev et al) So in conclusion, this isn't a random sample of (adult) humans, and the paper doesn't give standard deviations. It would've been more interesting of they had sampled an age range of humans which we would place GPT-4 on rather than just 'it's not as good' which is all this paper can say, really.
- cs702 3y ago
  Also, it's possible there are LLMs pretending to be human beings on Mechanical Turk!
  - jgalt212 3y ago
    What per cent of Mechanical Turk hits are being performed by an LLM these days?
    - dmd 3y ago
      Several researchers where I work used to use mturk regularly and now have been forced to stop using it entirely because of this. They started getting "As a large language model" answers ... in even things where the only possible answer was 'true' or 'false' or '1-7'.
      - ttul 3y ago
skepticATX 3y ago
It has been interesting to see evidence accumulating that shows, despite initial excitement bred by papers such as "Sparks", there is something missing from current language models. Individually none of these results will ever get the attention of a "Sparks" type paper, but collectively a strong case has been built.
- naasking 3y ago
  Sparks of AGI is not AGI. It's also possible that we're not testing LLMs fairly, or that merely slight tweaks to the architecture or methods would address the issues. I think this comment elaborates nicely: https://news.ycombinator.com/item?id=38332420 https://news.ycombinator.com/item?id=38332420 I do think there might be something missing, but I also suspect that it's not as far off as most think.
  - incongruity 3y ago
    So, in other words, perhaps what we have is a necessary component but not wholly sufficient on its own?
    - pixl97 3y ago
      That is my take on it. I think embodiment and the encoding of the natural laws (gravity, force, etc) that go into that will be another huge step at grounding AI. People tend to gravitate to thinking about humanoid robots when that is mentioned (and thereby terminators), but honestly I would think things closer to sensor networks involving thousands or millions of bodies like a hivemind would be more likely (why stick at the human level of a single body if you didn't have to). Interaction with the world is a means of determining truth... The ability to perform science. And as hard as embodiment is, it will be the easy part in my opinion. Continuous learning without losing the plot is going to be quite the challenge. If an LLM has something wrong, how does it update and change that bit of information without huge amounts of power use? How do you make the system learn 'important' things without filling up with junk/spam it is subject to? How do you keep the system aligned with a goal that is not destructive to itself or others?
kenjackson 3y ago
Can someone provide the prompts in text rather than the images from the paper? That would make it easier to try and replicate results.
- devbrac 3y ago
  Raw json files for each task from the ConceptARC paper are here: https://github.com/victorvikram/ConceptARC https://github.com/victorvikram/ConceptARC. It would be trivial to convert these into prompts using the format in the appendix of the paper.
kaoD 3y ago
Maybe I'm missing what "abstraction" means here but seems like the tasks were centered around grids and other spatial problems, which are a very limited subset of abstraction/reasoning. In my experience GPT4/V is pretty bad at those specifically, not necessarily around abstraction in general. Positions, rotations, etc. are a concept that GPT4 finds very hard to apply, which is kinda unsurprising since it has no body, no world, no space; it "lives" in the realm of text. DALLE3 suffers a similar problem where it has trouble with concepts like "upside down" and consistently fails to apply them to generated images.
- mr_toad 3y ago
  > DALLE3 suffers a similar problem where it has trouble with concepts like "upside down" and consistently fails to apply them to generated images. There’s probably not many (if any) upside down images or objects in the training data.
- pixl97 3y ago
  What would directions be for an intelligent creature that lives in zero gravity? I just like thinking about this for the same reasons humans like writing speculative science fiction. Trying to guess what alien perspectives look like, might also give us insights when we're the ones making the alien.
  - Retric 3y ago
    Basically the same, gravity doesn’t define left/right or North, South, East, and West for us just up and down.
    - trealira 3y ago
      However, North, South, East, and West are relative to the poles of the Earth. Something living in zero gravity would have to use some object as an anchor to determine the direction.
      - Retric 3y ago
xbmcuser 3y ago
It has been really interesting to read in last few years with machine learning how the model cant do this or that and the next week or month read it can do this or something else. Chat gpt and the models that have come after seem to have accelerated this back and forth a lot. Unless you keep up with it closely and keep updating your information I think what you knew it could do well or could not do well is no longer correct
RecycledEle 3y ago
What every paper I have seen so far is missing is that there are many ways to achieve super-human intelligence. (I need to give creidt to Isaac Arthur of SFIA for this.) Getting results faster is one way. AIs beat me in speed. Getting results cheaper is another way. AI is cheaper than I am. Knowledge across many fields is better. AI beats me here too. Getting better results in one narrow field is another way, but only one of many ways. I love evaluations of human produced work vs. machine produced work. If we had quality evaluations (not the type-oh riddled garbage most people use.) If we compared AIs to people who work in those fields in occupations recognized by the US Dept of Labor. If we asked both sides to justify their answers. If we had statistically significant sample sizes. Then maybe we could get some good results on quality of work. I can imagine the US DOL spending billions if dollars benchmarking AIs against humans i all the occupations the recognize. Alternately, this could be a very profitable company.
- jacquesm 3y ago
  That's why the 'G' in AGI is so important.
- varispeed 3y ago
  I view LLM as a search engine, that can find and present the knowledge in a way that is easier to consume. For instance: Find me a solution to 2x2. I see this as a search problem, not reasoning problem. I think many scientists don't look at it from this angle and push the agenda that somehow current AI can reason. It can't. It's just a search engine / pattern matching contraption that works so well it can fool people into thinking they interact with an intelligence. Frankly, many people "think" in the same way. They store facts in their brain and then retrieve them when asked, but they have limited ability to arrive at new facts from those they already know, without external input. In that sense, the "AI" can certainly replace humans in tasks that don't require to think or where there is a need to distil information from a wide array of sources in a useful way.
  - ben_w 3y ago
    I see it as translation, not search. Search is already done fantastically well at small scale with simple indexing, and (before adversarial reactions from SEO) at internet scale with Page rank. Asking if LLMs are really reasoning or not feels like an argument about terminology, like asking if A* really is route planning.
wouldbecouldbe 3y ago
Never heard of Mechanical Turk, hahah for a Dutch person it sounds pretty racist. Turk is what we call Turkish people.
- margalabargala 3y ago
  It's a reference to an old famous chess playing "machine" that had a human inside it' https://en.wikipedia.org/wiki/Mechanical_Turk https://en.wikipedia.org/wiki/Mechanical_Turk
  - wouldbecouldbe 3y ago
    Yeah I read, just interesting for non-english cultures sounds very different.
    - baxter001 3y ago
      No, it doesn't, Turk is what we call Turkish people too.
    - WrongAssumption 3y ago
      It was invented by a Hungarian to impress an Austrian. The Turk portion of the name is from the original Hungarian. It toured Europe for 80+ years. I have no clue what about this you think has anything to do with English culture.
    - renewiltord 3y ago
      That’s nothing. In American English, calling someone a turkey would be an insult. Turkey is how we refer to the nation Türkiye Cumhuriyeti. Imagine calling an entire nation an ungainly bird that we ritually consume every November. Almost as bad as giving Michael Jordan his own seat at the UN. But that’s what capitalism and the US hegemony have brought us.
      - mr_toad 3y ago
        > Imagine calling an entire nation an ungainly bird The bird was named after the country. And the country after the ethnic groups who spoke the Turkic languages, which derive from the native term Türk, which goes back at least 1400 years. Changing the spelling and pronunciation changes nothing. Saying that we called the country after the bird is buying into populist rabble-rousing without doing even five minutes of fact checking.
dr_dshiv 3y ago
I’m really looking forward to students majoring in “machine psychology.”
- ryzvonusef 3y ago
  https://en.wikipedia.org/wiki/Susan_Calvin https://en.wikipedia.org/wiki/Susan_Calvin > Graduating with a bachelor's degree from Columbia University in 2003, she began post-graduate work in cybernetics, learning to construct positronic brains such that responses to given stimuli could be accurately predicted. She joined US Robots in 2008 as their first Robopsychologist, having earned her PhD. By 2029, when she left Earth for the first time to visit Hyper Base, her formal title was Head Psychologist. https://en.wikipedia.org/wiki/Robopsychology https://en.wikipedia.org/wiki/Robopsychology
UrineSqueegee 3y ago
[dead]
theptip 3y ago
If you look at the appendix, you can see example transcripts. The sample they provide looks like a very bad eval. It’s encoding an originally visual problem into a textual matrix form, and then expecting GPT to recognize visual correlations. You simply can’t compare these two tasks! Most humans wouldn’t recognize the 5x5 matrix for a 4x4 square. So the comparison with “human level” is completely invalid. And even the valid comparison is only measuring visio-spatial intelligence, not IQ.
- Sharlin 3y ago
  Humans are biased towards grokking two-dimensional information arranged in grids because that's how vision works. (Non-V) GPT is not. Its natural gestalt has a vast number of dimensions. It is, and it should, be able to find correlations across data no matter how the input is arranged. (Note that for GPT text is not linear even if it appears so to humans.) To humans it would be qualitatively more difficult to recognize patterns in four or six or ten-dimensional data than in two dimensions. To GPT, it's just another variation of the same problem, one that's at most quantitatively more difficult. > And even the valid comparison is only measuring visio-spatial intelligence, not IQ. You're the first one to mention IQ. The paper definitely does not. In any case "visio-spatial intelligence" is a human concept. GPT doesn't distinguish between "visio-spatial" and any other kind of pattern recognition.
  - famouswaffles 3y ago
    >It is, and it should, be able to find correlations across data no matter how the input is arranged. That doesn't mean it doesn't favor certain arrangements. https://arxiv.org/abs/2305.18354 https://arxiv.org/abs/2305.18354 Linear is in fact better for this benchmark
datadrivenangel 3y ago
"Our experimental results support the conclusion that neither version of GPT-4 has developed robust abstraction abilities at humanlike levels." Some combination of LLMs and logical reasoning systems will get us much closer, but that becomes a lot more complicated.
intended 3y ago
This new Gen of AI adds an interesting twist to the infinite monkeys and typewriter issue. How do you actually check an infinite amount of junk to verify that one of them is the collected works of Shakespeare? The question I ask now is “whats your error rate for domain specific work ?” It could be faster and smarter, but it doesnt matter if it’s wrong.
varelse 3y ago
[dead]

217 points by mpweiher 3y ago

I've actually been working on a similar eval task that uses grids of symbols to evaluate LLM's reasoning ability around the game Wordle. My initial results are: - Success: ChatGPT-4-November, Phind-November - Failures (via hallucination): GPT-3.5, Mistral-7b, Llama2-13b Interestingly, ChatGPT-4 wrote a short python script (unprompted) to develop a pseudo-CoT type generation and answered correctly. Phind, which also answered correctly, is also a code-based/code-enabled LLM. Non-code-based-LLM's appear to reason cogently when describing their games - what moves are illegal, what the best guess in certain conditions are, etc - but their actual answer, even in multiple choice is comically wrong. These are very initial results but my tentative hypothesis is symbolic logical reasoning - string manipulation / function calling type tasks - are greatly enhanced by having a code modality. How to switch between natural language and code-generation may offer a strong boost on tasks like these.

tovej 3y ago

Conclusion is obvious, but the paper is still probably necessary. Of course LLM's can't reason. They pattern match answers to previously asked questions, and humans will read the text as a reasonable answer because we assign meaning to it, but there is simply no way an LLM could use a "mental model" to "reason" about a problem other than constructing sentences out of probable matches it's been trained on.

mr_toad 3y ago
> Of course LLM's can't reason. They pattern match You’re assuming that reasoning isn’t just pattern matching.
sfn42 3y ago
Love how the ai hypebros always downvote answers like this. Everyone's just insisting that LLMs are AGI just because they kinda seem that way.
- viraptor 3y ago
  Ftr, I mostly agree with the comment but downvoted it for the low quality "of course that's the answer, obviously" padded with some unhelpful tropes about trained pattern matching. (Let us know when you can prove that human communication is not just trained pattern matching)
  - tovej 3y ago
    Ok, I'll bite. I personally don't think human communication is simply pattern matching (e.g. we have drives, motives, relationships that we want to tell each other, these are internal systems, outside the visible band of communication). That's a little off topic though, my post was not about communication, but reasoning. Pattern matching in LLMs is specifically at the language level, as in "predict the next token in a stream", I should have probably specified this bit. You seem to refer to a more general kind of pattern matching when you say, where we're looking at a complicated structural model within which humans reason about the behavior of different system components and their interactions. There is no way to form an internal representation of such systems within an LLM, which is why I said "of course" it can't reason. An LLM is not designed with this capability in mind. If all you have is language, and predictive text, there is no way to use this to simulate and predict the rules in a higher-dimensional space than "stream of tokens", or let's put it another way: there would need to be an extremely high correlation between the pattern in the problem space and in language space for the likelihood of the model guessing right to be anywhere near human levels.
    - stevenhuang 3y ago
      > There is no way to form an internal representation of such systems within an LLM, which is why I said "of course" it can't reason. An LLM is not designed with this capability in mind. You claim this as if it is a foregone conclusion yet it is anything but. This is an active area of study, even within human cognition https://en.m.wikipedia.org/wiki/Predictive_coding https://en.m.wikipedia.org/wiki/Predictive_coding A complex system clearly does not need to be designed to evoke some behavior. Just look to the concept of emergent phenomenon. See evolution through natural selection as an obvious counter example. Evolution selects for fitness and not reasoning ability. Yet here we are. It is not known if the same isn't happening now for the transformer model. LLMs are universal function approximators. If reasoning ability is at all representable through such a process, it is possible for LLMs to find it. https://en.m.wikipedia.org/wiki/Universal_approximation_theorem https://en.m.wikipedia.org/wiki/Universal_approximation_theo...
koe123 3y ago
On the other hand, can we conclusively say that humans aren’t really advanced biological stochastic parrots?
naasking 3y ago
> Of course LLM's can't reason That they are not effective at some forms of reasoning does not entail they can't reason.
mcswell 3y ago
The conclusion may be obvious to you and me (although it's hard to know for certain, since these available LLMs are black boxes). But it's definitely not obvious to everyone. There are plenty of people saying this is the dawn of AGI, or that we're a few short steps from AGI. Whereas people like Gary Marcus (who knows tons more than I do) says LLMs are going off in the wrong direction.
- visarga 3y ago
  Yes, LLMs can't reason 100% correctly, but neither do humans. We can often reason correctly, but it's not always the case. Even reasoning, fundamental as it is, comes from feedback. Feedback from our actions teaches us how to reason. Learning from feedback is more general than reasoning - AI agents can definitely learn this way too, if they have enough freedom to explore. But you can't do it with supervised training sets alone. You need to put language models into agents in environments, give them goals and rewards. Then they can make their own training data and mistakes, build up their own experience. Can't teach an AI based on how people make mistakes, it needs to fix its own mistakes, but that means to deploy it in the wild, where there are consequences for errors. If you remember, DeepMind first tried to train a Go model on human game play, but it was limited. Then they started from scratch, and learning from feedback alone they surpassed human level, even though the feedback was one single bit of information at the end of a whole self-play game. And it had no pre-training prior to learning Go, unlike human players.
  - somewhereoutth 3y ago
    That presupposes that language encodes the world we live in completely, whereas in fact language is meaningless without the shared context of reality. 'up' and 'down' are only meaningful to intelligence that can experience space. Essentially LLMs just are oracles for the shadows on Plato's cave.
    - visarga 3y ago
      Do a reverse Chinese room experiment - remove from a human all the abilities multi-modal LLMs gain after training on human media. What's left? Homo ferus. Most of our intelligence is encoded in the environment and language, it's a collective process, not an individual one. We're collectively, not individually, very smart.

baxtr 3y ago

Could a human and a LLM submit a summary of the paper so we can compare?

j2kun 3y ago
That's what an abstract is for: > Our experimental results support the conclusion that neither version of GPT-4 has developed robust abstraction abilities at humanlike levels.
z7 3y ago
>The paper investigates the abstract reasoning abilities of text-only and multimodal versions of GPT-4 using the ConceptARC benchmark, concluding that neither version of GPT-4 has developed robust abstraction abilities at humanlike levels.

cs702 3y ago

Interesting. Immediate thoughts and questions: * How would human beings perform on the text-only version of the tasks given to GPT-4? * How would human beings perform if each grid is shown on its own, making it impossible to perform side-by-side visual comparisons? * How would human beings perform if each grid is shown on its own only once, making it impossible to perform any back-and-forth comparisons? * How could we give LLMs the ability to "pay attention" to different parts of images, as needed, so they can make back-and-forth comparisons between parts of different images to solve these kinds of visual reasoning tasks?

mistermann 3y ago
* How would human beings perform if they didn't know they were being tested (ie: if in the same mode they are in when writing comments on the internet)? * How would human beings perform if the questions are based on culture war topics, which tend to invoke System 1 intuitive/emotional thinking?
QuadmasterXLII 3y ago
> How could we give LLMs the ability to "pay attention" to different parts of images, as needed, so they can make back-and-forth comparisons between parts of different images to solve these kinds of visual reasoning tasks? I’ve got good news
- oefnak 3y ago
  What is it?
- phh 3y ago
  It's even all we need
YetAnotherNick 3y ago
Also I want to know how would know how much gain could be made by optimizing the prompt for GPT and by including things like CoT. Current version of the prompt is pretty bad both for humans and AI.

air7 3y ago

Thank god...

firtoz 3y ago

The best thing from research like this is that they allow new models to be built, or improvements on the existing ones that can lead them to pass these evaluations.

lewhoo 3y ago
Yeah but the question remains are they passing because of a new obtained/emerged quality or are they passing because they were taught specifically to how to pass or is there even a difference at this point.
- firtoz 3y ago
  Well, while training AI we make sure that we include enough training for them to pass some evaluation, but then test them on things that were not in the training data to make sure that they are not "over fitted".
  - viraptor 3y ago
    I don't think parent meant training specifically on that test data, but rather in that kind of task. Think more "LLM trained only on images of art" doing badly on this test, but then "LLM trained on images of art and IQ puzzles" doing better on this test. It's kind of like asking what even is IQ if you can learn how to solve Mensa puzzles and improve your score. Does it mean you're more intelligent?
    - firtoz 3y ago
      Oh, I see. One guess for GPT4 is that it's an amalgamation of models. So you would have various models trained more specifically like the ones you mentioned, and you ask them all to start answering the query, then you choose which one produces better results, and then you present that to the user. Alternatively you can have a decider model that knows what kind of queries goes to what specialised one, then have that as a "hidden layer", whether it be in the application level or "neuron layers".

krona 3y ago

This paper evaluates performance compared to a 'human' which presumably is an average adult human without cognitive impairment. I had to dig in to the references: In the first batch of participants collected via Amazon Mechanical Turk, each received 11 problems (this batch also only had two “minimal Problems,” as opposed to three such problems for everyone else). However, preliminary data examination showed that some participants did not fully follow the study instructions and had to be excluded (see Section 5.2). In response, we made the screening criteria more strict (requiring a Master Worker qualification, 99% of HITs approved with at least 2000 HIT history, as opposed to 95% approval requirement in the first batch). Participants in all but the first batch were paid $10 upon completing the experiment. Participants in the first batch were paid $5. In all batches, the median pay-per-hour exceeded the U.S. minimal wage. (Arseny Moskvichev et al) So in conclusion, this isn't a random sample of (adult) humans, and the paper doesn't give standard deviations. It would've been more interesting of they had sampled an age range of humans which we would place GPT-4 on rather than just 'it's not as good' which is all this paper can say, really.

cs702 3y ago
Also, it's possible there are LLMs pretending to be human beings on Mechanical Turk!
- jgalt212 3y ago
  What per cent of Mechanical Turk hits are being performed by an LLM these days?
  - dmd 3y ago
    Several researchers where I work used to use mturk regularly and now have been forced to stop using it entirely because of this. They started getting "As a large language model" answers ... in even things where the only possible answer was 'true' or 'false' or '1-7'.
    - ttul 3y ago

skepticATX 3y ago

It has been interesting to see evidence accumulating that shows, despite initial excitement bred by papers such as "Sparks", there is something missing from current language models. Individually none of these results will ever get the attention of a "Sparks" type paper, but collectively a strong case has been built.

naasking 3y ago
Sparks of AGI is not AGI. It's also possible that we're not testing LLMs fairly, or that merely slight tweaks to the architecture or methods would address the issues. I think this comment elaborates nicely: https://news.ycombinator.com/item?id=38332420 https://news.ycombinator.com/item?id=38332420 I do think there might be something missing, but I also suspect that it's not as far off as most think.
- incongruity 3y ago
  So, in other words, perhaps what we have is a necessary component but not wholly sufficient on its own?
  - pixl97 3y ago
    That is my take on it. I think embodiment and the encoding of the natural laws (gravity, force, etc) that go into that will be another huge step at grounding AI. People tend to gravitate to thinking about humanoid robots when that is mentioned (and thereby terminators), but honestly I would think things closer to sensor networks involving thousands or millions of bodies like a hivemind would be more likely (why stick at the human level of a single body if you didn't have to). Interaction with the world is a means of determining truth... The ability to perform science. And as hard as embodiment is, it will be the easy part in my opinion. Continuous learning without losing the plot is going to be quite the challenge. If an LLM has something wrong, how does it update and change that bit of information without huge amounts of power use? How do you make the system learn 'important' things without filling up with junk/spam it is subject to? How do you keep the system aligned with a goal that is not destructive to itself or others?

kenjackson 3y ago

Can someone provide the prompts in text rather than the images from the paper? That would make it easier to try and replicate results.

devbrac 3y ago
Raw json files for each task from the ConceptARC paper are here: https://github.com/victorvikram/ConceptARC https://github.com/victorvikram/ConceptARC. It would be trivial to convert these into prompts using the format in the appendix of the paper.

kaoD 3y ago

Maybe I'm missing what "abstraction" means here but seems like the tasks were centered around grids and other spatial problems, which are a very limited subset of abstraction/reasoning. In my experience GPT4/V is pretty bad at those specifically, not necessarily around abstraction in general. Positions, rotations, etc. are a concept that GPT4 finds very hard to apply, which is kinda unsurprising since it has no body, no world, no space; it "lives" in the realm of text. DALLE3 suffers a similar problem where it has trouble with concepts like "upside down" and consistently fails to apply them to generated images.

mr_toad 3y ago
> DALLE3 suffers a similar problem where it has trouble with concepts like "upside down" and consistently fails to apply them to generated images. There’s probably not many (if any) upside down images or objects in the training data.
pixl97 3y ago
What would directions be for an intelligent creature that lives in zero gravity? I just like thinking about this for the same reasons humans like writing speculative science fiction. Trying to guess what alien perspectives look like, might also give us insights when we're the ones making the alien.
- Retric 3y ago
  Basically the same, gravity doesn’t define left/right or North, South, East, and West for us just up and down.
  - trealira 3y ago
    However, North, South, East, and West are relative to the poles of the Earth. Something living in zero gravity would have to use some object as an anchor to determine the direction.
    - Retric 3y ago

xbmcuser 3y ago

It has been really interesting to read in last few years with machine learning how the model cant do this or that and the next week or month read it can do this or something else. Chat gpt and the models that have come after seem to have accelerated this back and forth a lot. Unless you keep up with it closely and keep updating your information I think what you knew it could do well or could not do well is no longer correct

RecycledEle 3y ago

What every paper I have seen so far is missing is that there are many ways to achieve super-human intelligence. (I need to give creidt to Isaac Arthur of SFIA for this.) Getting results faster is one way. AIs beat me in speed. Getting results cheaper is another way. AI is cheaper than I am. Knowledge across many fields is better. AI beats me here too. Getting better results in one narrow field is another way, but only one of many ways. I love evaluations of human produced work vs. machine produced work. If we had quality evaluations (not the type-oh riddled garbage most people use.) If we compared AIs to people who work in those fields in occupations recognized by the US Dept of Labor. If we asked both sides to justify their answers. If we had statistically significant sample sizes. Then maybe we could get some good results on quality of work. I can imagine the US DOL spending billions if dollars benchmarking AIs against humans i all the occupations the recognize. Alternately, this could be a very profitable company.

jacquesm 3y ago
That's why the 'G' in AGI is so important.
varispeed 3y ago
I view LLM as a search engine, that can find and present the knowledge in a way that is easier to consume. For instance: Find me a solution to 2x2. I see this as a search problem, not reasoning problem. I think many scientists don't look at it from this angle and push the agenda that somehow current AI can reason. It can't. It's just a search engine / pattern matching contraption that works so well it can fool people into thinking they interact with an intelligence. Frankly, many people "think" in the same way. They store facts in their brain and then retrieve them when asked, but they have limited ability to arrive at new facts from those they already know, without external input. In that sense, the "AI" can certainly replace humans in tasks that don't require to think or where there is a need to distil information from a wide array of sources in a useful way.
- ben_w 3y ago
  I see it as translation, not search. Search is already done fantastically well at small scale with simple indexing, and (before adversarial reactions from SEO) at internet scale with Page rank. Asking if LLMs are really reasoning or not feels like an argument about terminology, like asking if A* really is route planning.

wouldbecouldbe 3y ago

Never heard of Mechanical Turk, hahah for a Dutch person it sounds pretty racist. Turk is what we call Turkish people.

margalabargala 3y ago
It's a reference to an old famous chess playing "machine" that had a human inside it' https://en.wikipedia.org/wiki/Mechanical_Turk https://en.wikipedia.org/wiki/Mechanical_Turk
- wouldbecouldbe 3y ago
  Yeah I read, just interesting for non-english cultures sounds very different.
  - baxter001 3y ago
    No, it doesn't, Turk is what we call Turkish people too.
  - WrongAssumption 3y ago
    It was invented by a Hungarian to impress an Austrian. The Turk portion of the name is from the original Hungarian. It toured Europe for 80+ years. I have no clue what about this you think has anything to do with English culture.
  - renewiltord 3y ago
    That’s nothing. In American English, calling someone a turkey would be an insult. Turkey is how we refer to the nation Türkiye Cumhuriyeti. Imagine calling an entire nation an ungainly bird that we ritually consume every November. Almost as bad as giving Michael Jordan his own seat at the UN. But that’s what capitalism and the US hegemony have brought us.
    - mr_toad 3y ago
      > Imagine calling an entire nation an ungainly bird The bird was named after the country. And the country after the ethnic groups who spoke the Turkic languages, which derive from the native term Türk, which goes back at least 1400 years. Changing the spelling and pronunciation changes nothing. Saying that we called the country after the bird is buying into populist rabble-rousing without doing even five minutes of fact checking.

dr_dshiv 3y ago

I’m really looking forward to students majoring in “machine psychology.”

ryzvonusef 3y ago
https://en.wikipedia.org/wiki/Susan_Calvin https://en.wikipedia.org/wiki/Susan_Calvin > Graduating with a bachelor's degree from Columbia University in 2003, she began post-graduate work in cybernetics, learning to construct positronic brains such that responses to given stimuli could be accurately predicted. She joined US Robots in 2008 as their first Robopsychologist, having earned her PhD. By 2029, when she left Earth for the first time to visit Hyper Base, her formal title was Head Psychologist. https://en.wikipedia.org/wiki/Robopsychology https://en.wikipedia.org/wiki/Robopsychology

UrineSqueegee 3y ago

[dead]

theptip 3y ago

If you look at the appendix, you can see example transcripts. The sample they provide looks like a very bad eval. It’s encoding an originally visual problem into a textual matrix form, and then expecting GPT to recognize visual correlations. You simply can’t compare these two tasks! Most humans wouldn’t recognize the 5x5 matrix for a 4x4 square. So the comparison with “human level” is completely invalid. And even the valid comparison is only measuring visio-spatial intelligence, not IQ.

Sharlin 3y ago
Humans are biased towards grokking two-dimensional information arranged in grids because that's how vision works. (Non-V) GPT is not. Its natural gestalt has a vast number of dimensions. It is, and it should, be able to find correlations across data no matter how the input is arranged. (Note that for GPT text is not linear even if it appears so to humans.) To humans it would be qualitatively more difficult to recognize patterns in four or six or ten-dimensional data than in two dimensions. To GPT, it's just another variation of the same problem, one that's at most quantitatively more difficult. > And even the valid comparison is only measuring visio-spatial intelligence, not IQ. You're the first one to mention IQ. The paper definitely does not. In any case "visio-spatial intelligence" is a human concept. GPT doesn't distinguish between "visio-spatial" and any other kind of pattern recognition.
- famouswaffles 3y ago
  >It is, and it should, be able to find correlations across data no matter how the input is arranged. That doesn't mean it doesn't favor certain arrangements. https://arxiv.org/abs/2305.18354 https://arxiv.org/abs/2305.18354 Linear is in fact better for this benchmark

datadrivenangel 3y ago

"Our experimental results support the conclusion that neither version of GPT-4 has developed robust abstraction abilities at humanlike levels." Some combination of LLMs and logical reasoning systems will get us much closer, but that becomes a lot more complicated.

intended 3y ago

This new Gen of AI adds an interesting twist to the infinite monkeys and typewriter issue. How do you actually check an infinite amount of junk to verify that one of them is the collected works of Shakespeare? The question I ask now is “whats your error rate for domain specific work ?” It could be faster and smarter, but it doesnt matter if it’s wrong.

varelse 3y ago

[dead]