To those saying this is not surprising, yes it will be surprising to the general public who are being served ads from huge companies like MS or OpenAI saying LLMs can help with their accounting, help them close deals by crunching the numbers in seconds, write complex code for them etc etc.
This is important information for anyone to understand who thinks these systems are thinking, reasoning, and learning from them or that they’re having a conversation with them i.e. 90% of users of LLMs.
Quick sanity check: you're susceptible to pretty irresistible optical illusions which would never fool a VLM, does it mean you're not thinking? In fact, with a non-monospaced font I also have trouble determining whether these parens are balanced, and have to select them with the mouse, i.e. use a "dumb" tool, to make sure.
Reminder that "thinking" is an ill-defined term like others, and the question whether they "think" is basically irrelevant. No intelligent system, human or machine, will ever have zero error rate, due to the very nature of intelligence (another vague term). You have to deal with that the same way you deal with it in humans - either treat bugs as bugs and build systems resilient to bugs, or accept the baseline error rate if it's low enough.
No disrespect to them, but unless there is a financial incentive at stake for them (beyond SnP500 exposure), I've gotten to viewing this through the lens of sports teams, gaming consoles and religions. This has become part of the identity for some unfortunately and any position that isn't either "AGI imminent" or "This is useless" can be quite emotional.
Thing is, this being the case (along with all other LLM limits) does not mean that these models aren't impactful and shouldn't be scrutinised.
Also, mental health impact, job losses for white collar workers, privacy issues, concerns of rights holders on training data collection, all the current day impacts of LLMs are easily brushed aside by someone believing that LLMs are near the "everyone dies" stage, which just so happens to be helpful if one were to run a lab.
> we found that GPT-5.2 cannot even compute the parity of a short string like 11000, and GPT-5.2 cannot determine whether the parentheses in ((((()))))) are balanced.
I think there is a valid insight here which many already know:
LLMs are much more reliable at creating scripts and automation to do certain tasks than doing these tasks themselves.
For example if I provide an LLM my database schema and tell it to scan for redundant indexes and point out wrong naming conventions, it might do a passable but incomplete job.
But if I tell the LLM to code a python or nodejs script to do the same, I get significantly better results. And it's often faster too to generate and run the script than to let LLMs process large SQL files.
It's not just an issue of tokenization, it's almost a category error. Lisp, accounting and the number of r's in strawberry are all operations that require state. Balancing ((your)((lisp)(parens))) requires a stack, count r's in strawberry requires a register, counting to 5 requires an accumulator to hold 4.
An LLM is a router and completely stateless aside from the context you feed into it. Attention is just routing the probability distribution of the next token, and I'm not sure that's going to accumulate much in a single pass.
It's not dismissible as a misunderstanding of tokens. LLMs also embed knowledge of spelling - that's how they fixed the strawberry issue. It's a valid criticism and evaluation.
The r's in strawberry presents a different level of task to what people imagine. It seems trivial to a naive observer because the answer is easily derivable from the question without extra knowledge.
A more accurate analogy for humans would be to imagine if every word had a colour. You are told that there are also a sequence of different colours that correspond to the same colour as that word. You are even given a book showing every combination to memorise.
You learn the colours well enough that you can read and write coherently using them.
Then comes the question of how many chocolate-browns are in teal-with-a-hint-of-red. You know that teal-with-a-hint-of-red is a fruit and you know that the colour can also be constructed by crimson followed by Disney-blond. Now, do both of those contain chocolate-brown or just one of them, how many?
It requires excersizing memory to do a task that is underrepresented in the training data because humans simply do not have to do the task at all when the answer can be derived from the question representation. Humans also don't have the ability that the LLMs need but the letter representation doesn't need that ability.
I do think this is a tool issue. Here is what the article says:
> For the multiplication task, note that agents that make external calls to a calculator tool may have ZEH = ∞. While ZEH = ∞ does have meaning, in this paper we primarily evaluate the LLM itself without external tool calls
The models can count to infinity if you give them access to tools. The production models do this.
Not that the paper is wrong, it is still interesting to measure the core neural network of a model. But modern models use tools.
Seems like it’s maybe also a tool steering problem. These models should be reaching for tools to help solve factual problems. LLM should stick to prose.
I think this is still useful research that calls into question how “smart” these models are. If the model needs a separate tool to solve a problem, has the model really solved the problem, or just outsourced it to a harness that it’s been trained - via reinforcement learning - to call upon?
It matters if you’re curious about whether AGI is possible. Have we really built “thinking machines”, or are these systems just elaborate harnesses that leverage the non-deterministic nature of LLMs?
LLMs seem to me closer to Kahneman's System 1 than to System 2. When understood in this way, it is obvious why LLMs are bad at counting r's in "strawberries". But it also makes ZEH feel like it couldn't possibly be a useful metric, because it's a System 2 evaluation applied to a System 1 system.
> When understood in this way, it is obvious why LLMs are bad at counting r's in "strawberries".
no it doesnt. it makes sense that they cant count the rs because they dont have access to the actual word, only tokens that might represent parts or the whole of the word
A big part of skill aquisition in humans is moving tasks from system 2 to system 1, to free up the very scarce thinking resources for ever more complex tasks, that can then in turn be internalized and handled by system 1.
People are going to misinterpret this and overgeneralize the claim. This does not say that AI isn't reliable for things. It provides a method for quantifying the reliability for specific tasks.
You wouldn't say that a human who doesn't know how to read isn't reliable in everything, just in reading.
Counting is something that even humans need to learn how to do. Toddlers also don't understand quantity. If a 2 year old is able to count to even 10 it's through memorization and not understanding. It takes them like 2 more years of learning before they're able to comprehend things like numerical correspondence. But they do still know how to do other things that aren't counting before then.
Respectfully, toddlers cannot output useable code or have otherwise memorised results to an immense number of maths equations.
What this points at is the abstraction/emergence crux of it all. Why does an otherwise very capable LLM such as the GPT-5 series, despite having been trained on vastly more examples of frontend code of all shapes, sizes and quality levels, struggle to abstract all that training data to the point where outputting any frontend that deviates from the clearly used examples?
If LLMs, as they are now, were comparable with human learning, there'd be no scenario where a model that can provide output solving highly advanced equations can not count properly.
Similarly, a model such as GPT-5 trained on nearly all frontend code ever committed to any repo online, would have internalised more than that one template OpenAI predominantly leaned on.
These models, I think at this point there is little doubt, are impressive tools, but they still do not generalise or abstract information in the way a human mind does. Doesn't make them less impactful for industries, etc. but it makes any comparison to humans not very suitable.
> What this points at is the abstraction/emergence crux of it all. Why does
This paper has nothing to do with any questions starting with "why". It provides a metric for quantifying error on specific tasks.
> If LLMs, as they are now, were comparable with human learning
I think I missed the part where they need to be.
> struggle to abstract all that training data to the point where outputting any frontend that deviates from the clearly used examples? ... a model such as GPT-5 trained on nearly all frontend code ever committed to any repo online, would have internalised more than that one template OpenAI predominantly leaned on
There is a very big and very important difference between producing the same thing again and not being able to produce something else. When not given any reason to produce something else, humans also generate the same thing over and over. That's a problem of missing constraints, not of missing ability.
Long before AI there was this thing called Twitter Bootstrap. It dominated the web for...much longer than it should have. And that tragedy was done entirely by us meatsacks (not me personally). Where there's no goal for different output there's no reason to produce different output, and LLMs don't have their own goals because they don't have any mechanisms for desire (we hope).
[...] common trope that was proven false years ago by the existence of zero shot learning.
Ok, that's better than comparing LLMs to humans. ZSL however, has not proven anything of that sort false years ago, as it was mainly concerned with assessing whether LLMs are solely relying on precise instruction training or can generalise in a very limited degree beyond the initial tuning. That has never allowed for comparing human learning to LLM training.
Ironically, you are writing this under a paper that shows just that:
A model that cannot determine a short strings parity cannot have abstracted from the training data to arrive at the far more impressive and complicated maths challenges which it successfully solves in output. Some of the solutions we have seen in output require such innate understanding that, if there is no generalisation, far deeper than ZSL has ever shown, than this must come from training. Simple multiplication, etc. maybe, not the tasks people such as Easy Riders [0] throw at these models.
This paper shows exactly that even with ZSL, these models do only abstract in an incredibly limited manner and a lot of capabilities we see in the output are specifically trained, not generalised. Yes, generalisation in a limited capacity can happen, but no, it is not nearly close enough to yield some of the results we are seeing. I have also, neither here, nor in my initial comment, said that LLMs are only capable of outputting what their training data provides, merely that given what GPT-5 has been trained with, if there was any deeper abstraction these models gained during training, it'd be able to provide more than one frontend style.
Or to put it simpler, if the output provided can be useful for Maths at the Bachelor level and beyond and this capability is generalised as you believe, these tasks would not be a struggle for the model.
Many animals can count. Counting is recognizing that the box with 3 apples is preferable to the one with 2 apples.
Yes, 2 year olds might struggle with the externalization of numeric identities but if you have 1 M&M in one hand and 5 in the other and ask which they want, they’ll take the 5.
LLMs have the language part down, but fundamentally can’t count.
The concept of bigger/smaller is useful but is a distinct skill from counting. If you spread the M&Ms apart enough that the part of the brain responsible for gestalt clustering can't group them into a "bigger whole" signal, they'll no longer be able to do the thing you're saying (this is the law of proximity in gestalt psychology).
However many animals can distinguish independently small numbers, like 3 or 5, and recognize them whenever they see them.
So in this respect, there is little difference between humans and many animals. Humans learn to count to arbitrarily big numbers, but they can still easily recognize only small numbers.
> Counting is something that even humans need to learn how to do. Toddlers also don't understand quantity. If they're able to count to even 10 it's through memorization and not understanding.
I completely agree with you. LLMs are regurgitation machines with less intellect than a toddler, you nailed it.
Weird. I tried in chatGPT auto and it worked perfectly. I tried like 10 variations. I also did the letters in words. Got all of them right.
The one thing I did trip it up on was "Is there the sh sound in the word transportation". It said no. And then realized I asked for "sound" not letters. It then subsequently got the rest of the "sounds-like" tests I did.
heh, interesting that. I just tried it twice more with ChatGPT "Instant" (disabling "Auto-switch to Thinking") and it got it wrong both times. Does yours get it right without thinking or tool calls? If so, maybe it does like you better than me.
Probably zero. At the end of the day people pay for LLMs that write better code or summarize PDFs of hundreds of pages faster, not the ones that can count the letter r's better.
When LLMs can't count r's: see? LLMs can't think. Hoax!
When LLMs count r's: see? They patched and benchmark-maxxed. Hoax!
Whenever an "LLM fail" goes viral like the car wash question, you can observe the exact same wording of the question get "fixed" within a week or so. With slight variations in phrasing still able to replicate the problem.
Followed by lots of "works perfectly for me, why are people even talking about this?"
I can't say what exactly they're doing behind the scenes but it's a consistent pattern among the big SOTA model providers. With obvious incentive to "fix" the problem so users will then organically "debunk" the meme as they try it themselves and share their experiences.
Yeah well I presume at this point they have an agent download new LLM related papers as they come out and add all edge cases to their training set asap.
Is tokenization extremely efficient? Yes. Does it fundamentally break character-level understanding? Also yes. The only fix is endless memorization.
but it's a tricky question for LLMs; it shows that if it's not in the training set; LLMs could trip which kinda shows that the intelligence is not generalized yet.
I tried this with gemini - (i am trying(something(re(a(l(ly)c)r)a)z)((y)he)re)
> This is surprising given the excellent capabilities of GPT-5.2
The real surprise is that someone writing a paper on LLMs doesn't understand the baseline capabilities of a hallucinatory text generator (with tool use disabled).
The real suprise is people saying it's surprising when researchers and domain experts state something the former think goes against common sense/knowledge - as if they got them, and those researcers didn't already think their naive counter-argument already.
Let us be very clear: there is no such thing as a trustworthy LLM. Time and again they have shown that they understand nothing. They can be useful in the right context, but you can't trust them at all.
Did you use the exact API call shown in the paper? I am unable to replicate the paper's counterexamples via the chat UI, but that's not very surprising (if the LLM already only fails a few cases out of thousands, the small differences in context between API and chat might fix them).
To those saying this is not surprising, yes it will be surprising to the general public who are being served ads from huge companies like MS or OpenAI saying LLMs can help with their accounting, help them close deals by crunching the numbers in seconds, write complex code for them etc etc.
This is important information for anyone to understand who thinks these systems are thinking, reasoning, and learning from them or that they’re having a conversation with them i.e. 90% of users of LLMs.
Quick sanity check: you're susceptible to pretty irresistible optical illusions which would never fool a VLM, does it mean you're not thinking? In fact, with a non-monospaced font I also have trouble determining whether these parens are balanced, and have to select them with the mouse, i.e. use a "dumb" tool, to make sure.
Reminder that "thinking" is an ill-defined term like others, and the question whether they "think" is basically irrelevant. No intelligent system, human or machine, will ever have zero error rate, due to the very nature of intelligence (another vague term). You have to deal with that the same way you deal with it in humans - either treat bugs as bugs and build systems resilient to bugs, or accept the baseline error rate if it's low enough.
There’s a certain type of user here who reacts with rage when anyone points out flaws with LLMs. Why is that?
No disrespect to them, but unless there is a financial incentive at stake for them (beyond SnP500 exposure), I've gotten to viewing this through the lens of sports teams, gaming consoles and religions. This has become part of the identity for some unfortunately and any position that isn't either "AGI imminent" or "This is useless" can be quite emotional.
Thing is, this being the case (along with all other LLM limits) does not mean that these models aren't impactful and shouldn't be scrutinised.
Also, mental health impact, job losses for white collar workers, privacy issues, concerns of rights holders on training data collection, all the current day impacts of LLMs are easily brushed aside by someone believing that LLMs are near the "everyone dies" stage, which just so happens to be helpful if one were to run a lab.
I suspect they're afraid that if the hype dies, so will the pace of progress on LLMs as well as their cheap/free usage of them.
> we found that GPT-5.2 cannot even compute the parity of a short string like 11000, and GPT-5.2 cannot determine whether the parentheses in ((((()))))) are balanced.
I think there is a valid insight here which many already know: LLMs are much more reliable at creating scripts and automation to do certain tasks than doing these tasks themselves.
For example if I provide an LLM my database schema and tell it to scan for redundant indexes and point out wrong naming conventions, it might do a passable but incomplete job.
But if I tell the LLM to code a python or nodejs script to do the same, I get significantly better results. And it's often faster too to generate and run the script than to let LLMs process large SQL files.
Doesn't this just look like another case of "count the r's in strawberry" ie not understanding how tokenization works?
This is well known and not that interesting to me - ask the model to use python to solve any of these questions and it will get it right every time.
It's not just an issue of tokenization, it's almost a category error. Lisp, accounting and the number of r's in strawberry are all operations that require state. Balancing ((your)((lisp)(parens))) requires a stack, count r's in strawberry requires a register, counting to 5 requires an accumulator to hold 4.
An LLM is a router and completely stateless aside from the context you feed into it. Attention is just routing the probability distribution of the next token, and I'm not sure that's going to accumulate much in a single pass.
It's not dismissible as a misunderstanding of tokens. LLMs also embed knowledge of spelling - that's how they fixed the strawberry issue. It's a valid criticism and evaluation.
The r's in strawberry presents a different level of task to what people imagine. It seems trivial to a naive observer because the answer is easily derivable from the question without extra knowledge.
A more accurate analogy for humans would be to imagine if every word had a colour. You are told that there are also a sequence of different colours that correspond to the same colour as that word. You are even given a book showing every combination to memorise.
You learn the colours well enough that you can read and write coherently using them.
Then comes the question of how many chocolate-browns are in teal-with-a-hint-of-red. You know that teal-with-a-hint-of-red is a fruit and you know that the colour can also be constructed by crimson followed by Disney-blond. Now, do both of those contain chocolate-brown or just one of them, how many?
It requires excersizing memory to do a task that is underrepresented in the training data because humans simply do not have to do the task at all when the answer can be derived from the question representation. Humans also don't have the ability that the LLMs need but the letter representation doesn't need that ability.
I do think this is a tool issue. Here is what the article says:
> For the multiplication task, note that agents that make external calls to a calculator tool may have ZEH = ∞. While ZEH = ∞ does have meaning, in this paper we primarily evaluate the LLM itself without external tool calls
The models can count to infinity if you give them access to tools. The production models do this.
Not that the paper is wrong, it is still interesting to measure the core neural network of a model. But modern models use tools.
Seems like it’s maybe also a tool steering problem. These models should be reaching for tools to help solve factual problems. LLM should stick to prose.
I think this is still useful research that calls into question how “smart” these models are. If the model needs a separate tool to solve a problem, has the model really solved the problem, or just outsourced it to a harness that it’s been trained - via reinforcement learning - to call upon?
It has "outsourced" it to another component, sure, but does that matter?
What the user sees is the total behavior of the entire system, not whether the system has internal divisions and separations.
It matters if you’re curious about whether AGI is possible. Have we really built “thinking machines”, or are these systems just elaborate harnesses that leverage the non-deterministic nature of LLMs?
Can someone produce a single example <20 characters that fails with latest thinking model? Can’t seem to reproduce.
LLMs seem to me closer to Kahneman's System 1 than to System 2. When understood in this way, it is obvious why LLMs are bad at counting r's in "strawberries". But it also makes ZEH feel like it couldn't possibly be a useful metric, because it's a System 2 evaluation applied to a System 1 system.
> When understood in this way, it is obvious why LLMs are bad at counting r's in "strawberries".
no it doesnt. it makes sense that they cant count the rs because they dont have access to the actual word, only tokens that might represent parts or the whole of the word
A big part of skill aquisition in humans is moving tasks from system 2 to system 1, to free up the very scarce thinking resources for ever more complex tasks, that can then in turn be internalized and handled by system 1.
People are going to misinterpret this and overgeneralize the claim. This does not say that AI isn't reliable for things. It provides a method for quantifying the reliability for specific tasks.
You wouldn't say that a human who doesn't know how to read isn't reliable in everything, just in reading.
Counting is something that even humans need to learn how to do. Toddlers also don't understand quantity. If a 2 year old is able to count to even 10 it's through memorization and not understanding. It takes them like 2 more years of learning before they're able to comprehend things like numerical correspondence. But they do still know how to do other things that aren't counting before then.
Respectfully, toddlers cannot output useable code or have otherwise memorised results to an immense number of maths equations.
What this points at is the abstraction/emergence crux of it all. Why does an otherwise very capable LLM such as the GPT-5 series, despite having been trained on vastly more examples of frontend code of all shapes, sizes and quality levels, struggle to abstract all that training data to the point where outputting any frontend that deviates from the clearly used examples?
If LLMs, as they are now, were comparable with human learning, there'd be no scenario where a model that can provide output solving highly advanced equations can not count properly.
Similarly, a model such as GPT-5 trained on nearly all frontend code ever committed to any repo online, would have internalised more than that one template OpenAI predominantly leaned on.
These models, I think at this point there is little doubt, are impressive tools, but they still do not generalise or abstract information in the way a human mind does. Doesn't make them less impactful for industries, etc. but it makes any comparison to humans not very suitable.
> What this points at is the abstraction/emergence crux of it all. Why does
This paper has nothing to do with any questions starting with "why". It provides a metric for quantifying error on specific tasks.
> If LLMs, as they are now, were comparable with human learning
I think I missed the part where they need to be.
> struggle to abstract all that training data to the point where outputting any frontend that deviates from the clearly used examples? ... a model such as GPT-5 trained on nearly all frontend code ever committed to any repo online, would have internalised more than that one template OpenAI predominantly leaned on
There is a very big and very important difference between producing the same thing again and not being able to produce something else. When not given any reason to produce something else, humans also generate the same thing over and over. That's a problem of missing constraints, not of missing ability.
Long before AI there was this thing called Twitter Bootstrap. It dominated the web for...much longer than it should have. And that tragedy was done entirely by us meatsacks (not me personally). Where there's no goal for different output there's no reason to produce different output, and LLMs don't have their own goals because they don't have any mechanisms for desire (we hope).
[I've edited this comment for content and format]
[...] common trope that was proven false years ago by the existence of zero shot learning.
Ok, that's better than comparing LLMs to humans. ZSL however, has not proven anything of that sort false years ago, as it was mainly concerned with assessing whether LLMs are solely relying on precise instruction training or can generalise in a very limited degree beyond the initial tuning. That has never allowed for comparing human learning to LLM training.
Ironically, you are writing this under a paper that shows just that:
A model that cannot determine a short strings parity cannot have abstracted from the training data to arrive at the far more impressive and complicated maths challenges which it successfully solves in output. Some of the solutions we have seen in output require such innate understanding that, if there is no generalisation, far deeper than ZSL has ever shown, than this must come from training. Simple multiplication, etc. maybe, not the tasks people such as Easy Riders [0] throw at these models.
This paper shows exactly that even with ZSL, these models do only abstract in an incredibly limited manner and a lot of capabilities we see in the output are specifically trained, not generalised. Yes, generalisation in a limited capacity can happen, but no, it is not nearly close enough to yield some of the results we are seeing. I have also, neither here, nor in my initial comment, said that LLMs are only capable of outputting what their training data provides, merely that given what GPT-5 has been trained with, if there was any deeper abstraction these models gained during training, it'd be able to provide more than one frontend style.
Or to put it simpler, if the output provided can be useful for Maths at the Bachelor level and beyond and this capability is generalised as you believe, these tasks would not be a struggle for the model.
[0] https://www.youtube.com/@easy_riders
>Counting is something that even humans need to learn how to do
No human who can program, solve advanced math problems, or can talk about advanced problem domains at expert level, however, would fail to count to 5.
This is not a mere "trains also need to learn this" but points to a fundamental mismatch about how humans and LLMs learn.
You’re conflating counting and language.
Many animals can count. Counting is recognizing that the box with 3 apples is preferable to the one with 2 apples.
Yes, 2 year olds might struggle with the externalization of numeric identities but if you have 1 M&M in one hand and 5 in the other and ask which they want, they’ll take the 5.
LLMs have the language part down, but fundamentally can’t count.
The concept of bigger/smaller is useful but is a distinct skill from counting. If you spread the M&Ms apart enough that the part of the brain responsible for gestalt clustering can't group them into a "bigger whole" signal, they'll no longer be able to do the thing you're saying (this is the law of proximity in gestalt psychology).
Most animals can distinguish bigger from smaller.
However many animals can distinguish independently small numbers, like 3 or 5, and recognize them whenever they see them.
So in this respect, there is little difference between humans and many animals. Humans learn to count to arbitrarily big numbers, but they can still easily recognize only small numbers.
> Counting is something that even humans need to learn how to do. Toddlers also don't understand quantity. If they're able to count to even 10 it's through memorization and not understanding.
I completely agree with you. LLMs are regurgitation machines with less intellect than a toddler, you nailed it.
AI is here!
Whenveer I see these papers and try them, they always work. This paper is two months old, which in LLM years is like 10 years of progress.
It would be interesting to actively track how far long each progressive model gets...
I just tried it in ChatGPT "Auto" and it didn't work
> Yes — ((((()))))) is balanced.
> It has 6 opening ( and 6 closing ), and they’re properly nested.
Though it did work when using "Extensive Thinking". The model wrote a Python program to solve this.
> Almost balanced — ((((()))))) has 5 opening parentheses and 6 closing parentheses, so it has one extra ).
> A balanced version would be: ((((()))))
Testing a couple of different models without a harness such that no tool calls are possible would be interesting
Weird. I tried in chatGPT auto and it worked perfectly. I tried like 10 variations. I also did the letters in words. Got all of them right.
The one thing I did trip it up on was "Is there the sh sound in the word transportation". It said no. And then realized I asked for "sound" not letters. It then subsequently got the rest of the "sounds-like" tests I did.
Clearly, my ChatGPT is just better than yours.
heh, interesting that. I just tried it twice more with ChatGPT "Instant" (disabling "Auto-switch to Thinking") and it got it wrong both times. Does yours get it right without thinking or tool calls? If so, maybe it does like you better than me.
You are trying it on a production model. The paper is using models with tool calls disabled.
Even more interesting to track how many of those are just ad-hoc patched.
Probably zero. At the end of the day people pay for LLMs that write better code or summarize PDFs of hundreds of pages faster, not the ones that can count the letter r's better.
When LLMs can't count r's: see? LLMs can't think. Hoax!
When LLMs count r's: see? They patched and benchmark-maxxed. Hoax!
You just can't reason with the anti-LLM group.
Whenever an "LLM fail" goes viral like the car wash question, you can observe the exact same wording of the question get "fixed" within a week or so. With slight variations in phrasing still able to replicate the problem.
Followed by lots of "works perfectly for me, why are people even talking about this?"
I can't say what exactly they're doing behind the scenes but it's a consistent pattern among the big SOTA model providers. With obvious incentive to "fix" the problem so users will then organically "debunk" the meme as they try it themselves and share their experiences.
You are misremembering. There’s no patch. All these examples used the instant model.
Actually almost all LLMs when they write numbered sections in a markdown have the counting wrong. They miss the numbers in between and such.
So yes.
And the valuations. Trillion dollar grifter industry.
Yeah well I presume at this point they have an agent download new LLM related papers as they come out and add all edge cases to their training set asap.
Is tokenization extremely efficient? Yes. Does it fundamentally break character-level understanding? Also yes. The only fix is endless memorization.
Nice! Although I tried the parenthesis balanced question with gemini and it gave the right answer in first attempt.
but it's a tricky question for LLMs; it shows that if it's not in the training set; LLMs could trip which kinda shows that the intelligence is not generalized yet.
I tried this with gemini - (i am trying(something(re(a(l(ly)c)r)a)z)((y)he)re)
and it tripped.
Ran this through Qwen3.5-397B-A17B, and the difference between 4 characters and 5 is wild to see:
> are the following parenthesis balanced? ((())))
> No, the parentheses are not balanced.
> Here is the breakdown:
... following up with:> what about these? ((((())))
> Yes, the parentheses are balanced.
> Here is the breakdown:
... and uses ~5,000 tokens to get the wrong answer.> This is surprising given the excellent capabilities of GPT-5.2
The real surprise is that someone writing a paper on LLMs doesn't understand the baseline capabilities of a hallucinatory text generator (with tool use disabled).
The real suprise is people saying it's surprising when researchers and domain experts state something the former think goes against common sense/knowledge - as if they got them, and those researcers didn't already think their naive counter-argument already.
One! Two! Five!
You are polluting future training data.
Let us be very clear: there is no such thing as a trustworthy LLM. Time and again they have shown that they understand nothing. They can be useful in the right context, but you can't trust them at all.
> This is surprising given the excellent capabilities of GPT-5.2.
Is this seriously surprising to anyone who knows the absolute minimum about how LLMs parse and understand text?
Nope.
It's only surprising to people who still think they're going to build God out of LLMs.
Why didn't OpenAI finetune the model to use the python tool it has for these tasks?
They do, in the paper they mention they evaluate the LLM without tools
bruh
There’s no way this is right. I checked complicated ones with the latest thinking model. Can someone come up with a counter example?
Did you use the exact API call shown in the paper? I am unable to replicate the paper's counterexamples via the chat UI, but that's not very surprising (if the LLM already only fails a few cases out of thousands, the small differences in context between API and chat might fix them).
"in this paper we primarily evaluate the LLM itself without external tool calls."
Maybe this is a factor?
No tools were used.