hckrnws

The fact that it was ever seriously entertained that a "chain of thought" was giving some kind of insight into the internal processes of an LLM bespeaks the lack of rigor in this field. The words that are coming out of the model are generated to optimize for RLHF and closeness to the training data, that's it! They aren't references to internal concepts, the model is not aware that it's doing anything so how could it "explain itself"?

CoT improves results, sure. And part of that is probably because you are telling the LLM to add more things to the context window, which increases the potential of resolving some syllogism in the training data: One inference cycle tells you that "man" has something to do with "mortal" and "Socrates" has something to do with "man", but two cycles will spit those both into the context window and lets you get statistically closer to "Socrates" having something to do with "mortal". But given that the training/RLHF for CoT revolves around generating long chains of human-readable "steps", it can't really be explanatory for a process which is essentially statistical.

internal concepts, the model is not aware that it's doing anything so how could it "explain itself"

This in a nutshell is why I hate that all this stuff is being labeled as AI. Its advanced machine learning (another term that also feels inaccurate but I concede is at least closer to whats happening conceptually)

Really, LLMs and the like still lack any model of intelligence. Its, in the most basic of terms, algorithmic pattern matching mixed with statistical likelihoods of success.

And that can get things really really far. There are entire businesses built on doing that kind of work (particularly in finance) with very high accuracy and usefulness, but its not AI.

While I agree that LLMs are hardly sapient, it's very hard to make this argument without being able to pinpoint what a model of intelligence actually is.

"Human brains lack any model of intelligence. It's just neurons firing in complicated patterns in response to inputs based on what statistically leads to reproductive success"

What's wrong with just calling them smart algorithmic models?

Being smart allows somewhat to be wrong, as long as that leads to a satisfying solution. Being intelligent on the other hand requires foundational correctness in concepts that aren't even defined yet.

EDIT: I also somewhat like the term imperative knowledge (models) [0]

[0]: https://en.wikipedia.org/wiki/Procedural_knowledge

The problem with "smart" is that they fail at things that dumb people succeed at. They have ludicrous levels of knowledge and a jaw dropping ability to connect pieces while missing what's right in front of them.

The gap makes me uncomfortable with the implications of the word "smart". It is orthogonal to that.

they fail at things that dumb people succeed at

Funnily enough, you can also observe that in humans. The number of times I have observed people from highly intellectual, high income/academic families struggle with simple tasks that even the dumbest people do with ease is staggering. If you're not trained for something and suddenly confronted with it for the first time, you will also in all likelihood fail. "Smart" is just as ill-defined as any other clumsy approach to define intelligence.

Bombs can be smart, even though they sometimes miss the target.

That's not at all on par with what I'm saying.

There exists a generally accepted baseline definition for what crosses the threshold of intelligent behavior. We shouldn't seek to muddy this.

EDIT: Generally its accepted that a core trait of intelligence is an agent’s ability to achieve goals in a wide range of environments. This means you must be able to generalize, which in turn allows intelligent beings to react to new environments and contexts without previous experience or input.

Nothing I'm aware of on the market can do this. LLMs are great at statistically inferring things, but they can't generalize which means they lack reasoning. They also lack the ability to seek new information without prompting.

The fact that all LLMs boil down to (relatively) simple mathematics should be enough to prove the point as well. It lacks spontaneous reasoning, which is why the ability to generalize is key

"There exists a generally accepted baseline definition for what crosses the threshold of intelligent behavior" not really. The whole point they are trying to make is that the capability of these models IS ALREADY muddying the definition of intelligence. We can't really test it because the distribution its learned is so vast. Hence why he have things like ARC now.

Even if its just gradient descent based distribution learning and there is no "internal system" (whatever you think that should look like) to support learning the distribution, the question is if that is more than what we are doing or if we are starting to replicate our own mechanisms of learning.

Peoples’ memories are so short. Ten years ago the “well accepted definition of intelligence” was whether something could pass the Turing test. Now that goalpost has been completely blown out of the water and people are scrabbling to come up with a new one that precludes LLMs.

A useful definition of intelligence needs to be measurable, based on inputs/outputs, not internal state. Otherwise you run the risk of dictating how you think intelligence should manifest, rather than what it actually is. The former is a prescription, only the latter is a true definition.

I frequently see this characterization and can't agree with it. If I say "well I suppose you'd at least need to do A to qualify" and then later say "huh I guess A wasn't sufficient, looks like you'll also need B" that is not shifting the goalposts.

At worst it's an incomplete and ad hoc specification.

More realistically it was never more than an educated guess to begin with, about something that didn't exist at the time, still doesn't appear to exist, is highly subjective, lacks a single broadly accepted rigorous definition to this very day, and ultimately boils down to "I'll know it when I see it".

I'll know it when I see it, and I still haven't seen it. QED

If I say "well I suppose you'd at least need to do A to qualify" and then later say "huh I guess A wasn't sufficient, looks like you'll also need B" that is not shifting the goalposts.

I dunno, that seems like a pretty good distillation of what moving the goalposts is.

I’ll know it when I see it, and I haven’t seen it. QED

While pithily put, thats not a compelling argument. You feel that LLMs are not intelligent. I feel that they may be intelligent. Without a decent definition of what intelligence is, the entire argument is silly.

Shifting goalposts usually (at least in my understanding) refers to changing something without valid justification that was explicitly set in a previous step (subjective wording I realize - this is off the top of my head). In an adversarial context it would be someone attempting to gain an advantage by subtly changing a premise in order to manipulate the conclusion.

An incomplete list, in contrast, is not a full set of goalposts. It is more akin to a declared lower bound.

I also don't think it to applies to the case where the parties are made aware of a change in circumstances and update their views accordingly.

You feel that LLMs are not intelligent. I feel that they may be intelligent.

Weirdly enough I almost agree with you. LLMs have certainly challenged my notion of what intelligence is. At this point I think it's more a discussion of what sorts of things people are referring to when they use that word and if we can figure out an objective description that distinguishes those things from everything else.

Without a decent definition of what intelligence is, the entire argument is silly.

I completely agree. My only objection is to the notion that goalposts have been shifted since in my view they were never established in the first place.

I dunno, that seems like a pretty good distillation of what moving the goalposts is.

Only if you don't understand what "the goalposts" means. The goalpost isn't "pass the turing test", the goalpost is "manage to do all the same kind of intellectual tasks that humans are", nobody has moved that since the start in the quest for AI.

LLM’s can’t pass an unrestricted Touring test. LLM’s can mimic intelligence, but if you actually try and exploit their limitations the deception is still trivial to unmask.

Various chat bots have long been able to pass more limited versions of a Touring test. The most extreme constraint allows for simply replaying a canned conversation which with a helpful human assistant makes it indistinguishable from a human. But exploiting limitations on a testing format doesn’t have anything to do with testing for intelligence.

I’ve realized while reading these comments my opinions on LLMs being intelligent has significantly increased. Rather than argue any specific test, I believe no one can come up with a text-based intelligence test that 90% of literate adults can pass but the top LLMs fail.

This would mean there’s no definition of intelligence you could tie to a test where humans would be intelligent but LLMs wouldn’t.

A maybe more palatable idea is that having “intelligence” as a binary is insufficient. I think it’s more of an extremely skewed distribution. With how humans are above the rest, you didn’t have to nail the cutoff point to get us on one side and everything else on the other. Maybe chimpanzees and dolphins slip in. But now, the LLMs are much closer to humans. That line is harder to draw. Actually not possible to draw it so people are on one side and LLMs on the other.

Why presuppose that it's possible to test intelligence via text? Most humans have been illiterate for most of human history.

I don't mean to claim that it isn't possible, just that I'm not clear why we should assume that it is or that there would be an obvious way of going about it.

Seems pretty reasonable to presuppose this when you filter to people who are literate. That’s darn near a definition of literate, that you can engage with the text intelligently.

I thought the definition of literate was "can interpret text in place of the spoken word". At which point it's worth noting that text is a much lower bandwidth channel than in person communication. Also worth noting that, ex, a mute person could still be considered intelligent.

Is it necessarily the case that you could discern general intelligence via a test with fixed structure, known to all parties in advance, carried out via a synthesized monotone voice? I'm not saying "you definitely can't do that" just that I don't see why we should a priori assume it to be possible.

Now that likely seems largely irrelevant and out in the weeds and normally I would feel that way. But if you're going to suppose that we can't cleanly differentiate LLMs from humans then it becomes important to ask if that's a consequence of the LLMs actually exhibiting what we would consider general intelligence versus an inherent limitation of the modality in which the interactions are taking place.

Personally I think it's far more likely that we just don't have very good tests yet, that our working definition of "general intelligence" (as well as just "intelligence") isn't all that great yet, and that in the end many humans who we consider to exhibit a reasonable level of such will nonetheless fail to pass tests that are based solely on an isolated exchange of natural language.

I generally agree with your framing, I'll just comment on a minor detail about what "literate" means. Typically, people are classed in three categories of literacy, not two: illiterate means you essentially can't read at all, literate means you can read and understand text to some level, but then there are people who are functionally illiterate - people who can read the letters and sound out text, but can't actively comprehend what they're reading to a level that allows them to function normally in society - say, being able to read and comprehend an email they receive at work or a news article. This difference between literate and functionally illiterate may have been what the poster above was referring to.

Note that functional illiteracy is not some niche phenomenon, it's a huge problem in many school systems. In my own country (Romania), while the rate of illiteracy is something like <1% of the populace, the rate of functional illiteracy is estimated to be as high as 45% of those finishing school.

Or maybe accept that LLMs are intelligent and it's human bias that is the oddity here.

My whole comment was accepting LLMs as intelligent. It’s the first sentence.

How does an LLM muddy the definition of intelligence any more than a database or search engine does? They are lossy databases with a natural language interface, nothing more.

Ah, but what is in the database? At this point it's clearly not just facts, but problem-solving strategies and an execution engine. A database of problem-solving strategies which you can query with a natural language description of your problem and it returns an answer to your problem... well... sounds like intelligence to me.

problem-solving strategies and an execution engine

Extremely unfounded claims. See: the root comment of this tree.

…things that look like problem solving strategies in performance, then.

datasets and search engines are deterministic. humans, and llms are not.

LLMs are completely deterministic. Their fundamental output is a vector representing a probability distribution of the next token given the model weights and context. Given the same inputs an identical output vector will be produced 100% of the time.

This fact is relied upon by for example https://bellard.org/ts_zip/ a lossless compression system that would not work if LLMs were nondeterministic.

In practice most LLM systems use this distribution (along with a “temperature” multiplier) to make a weighted random choice among the tokens, giving the illusion of nondeterminism. But there’s no fundamental reason you couldn’t for example always choose the most likely token, yielding totally deterministic output.

This is an excellent and accessible series going over how transformer systems work if you want to learn more. https://youtu.be/wjZofJX0v4M

In practice most LLM systems use this distribution (along with a “temperature” multiplier) to make a weighted random choice among the tokens

In other words, LLMs are not deterministic in just about any real setting. What you said there only compounds with MoE architectures, variable test-time compute allocation, and o3-like sampling.

i've heard it actually depends on the model / hosting architecture. some are not deterministic at the numeric level because there is so much floating point math going on in distributed fashion across gpus, with unpredictable rounding/syncing across machines

The only reason LLMs are stochastic instead of deterministic is a random number generator. There is nothing inherently non-deterministic about LLM algorithms unless you turn up the "temperature" of selecting the next word. The fact that determinism can be changed by turning a knob is clear evidence that they are closer to a database or search engine than a human.

You can turn the determinism knob on humans. Psychedelics are one method.

I think that's more adjusting the parameters of the built-in denoising and feature detection circuits of the inherently noisy analog computer that is the brain.

The LLM's output is chaotic relative to the input, but it's deterministic right? Same settings, same model, same input, .. same output? Where does the chain get broken here?

Depends on what you mean specifically by the output. The actual neural network will produce deterministic outputs that could be interpreted as probability values for various tokens. But the interface you'll commonly see used in front of these models will then non-deterministiclaly choose a single next token to output based on those probabilities. Then, this single randomly chosen output is fed back into the network to produce another token, and this process repeats.

I would ultimately call the result non-deterministic. You could make it deterministic relatively easily by having a deterministic process for choosing a single token from all of the outputs of the NN (say, always pick the one with the highest weight, and if there are multiple with the same weight, pick the first one in token index order), but no one normally does this, because the results aren't that great per my understanding.

You can have the best of both worlds with something like weighted_selection( output, hash( output ) ) using the hash as the PRNG seed. (If you're paranoid about statistical issues due to identical outputs (extremely unlikely) then add a nonce to the hash.)

Now compare a human to an LSTM with persistent internal state that you can't reset.

There exists a generally accepted baseline definition for what crosses the threshold of intelligent behavior.

Go on. We are listening.

I think the confusion is because you're referring to a common understanding of what AI is but I think the definition of AI is different for different people.

Can you give your definition of AI? Also what is the "generally accepted baseline definition for what crosses the threshold of intelligent behavior"?

You are doubling down on a muddled vague non-technical intuition about these terms.

Please tell us what that "baseline definition" is.

Generally its accepted that a core trait of intelligence is an agent’s ability to achieve goals in a wide range of environments.

Be that as it may, a core trait is very different from a generally accepted threshold. What exactly is the threshold? Which environments are you referring to? How is it being measured? What goals are they?

You may have quantitative and unambiguous answers to these questions, but I don't think they would be commonly agreed upon.

intelligence is an agent’s ability to achieve goals in a wide range of environments. This means you must be able to generalize, which in turn allows intelligent beings to react to new environments and contexts without previous experience or input.

I applaud the bravery of trying to one shot a definition of intelligence, but no intelligent being acts without previous experience or input. If you're talking about in-sample vs out of sample, LLMs do that all the time. At some point in the conversation, they encounter something completely new and react to it in a way that emulates an intelligent agent.

What really makes them tick is language being a huge part of the intelligence puzzle, and language is something LLMs can generate at will. When we discover and learn to emulate the rest, we will get closer and closer to super intelligence.

What is that baseline threshold for intelligence? Could you provide concrete and objective results, that if demonstrated by a computer system would satisfy your criteria for intelligence?

see the edit. boils down to the ability to generalize, LLMs can't generalize. I'm not the only one who holds this view either. Francois Chollet, a former intelligence researcher at Google also shares this view.

Are you able to formulate "generalization" in a concrete and objective way that could be achieved unambiguously, and is currently achieved by a typical human? A lot of people would say that LLMs generalize pretty well - they certainly can understand natural language sequences that are not present in their training data.

A lot of people would say that LLMs generalize pretty well

What do you mean here? The trained model, the inference engine, is the one that makes an LLM for "a lot of people".

they certainly can understand natural language sequences that are not present in their training data

Keeping the trained model as LLM in mind, I think learning a language includes generalization and is typically achieved by a human, so I'll try to formulate:

Can a trained LLM model learn languages that hasn't been in its training set just by chatting/prompting? Given that any Korean texts were excluded from the training set, could Korean be learned? Does that even work with languages descending from the same language family (Spanish in the training set but Italian should be learned)?

Chollet's argument was that it's not "true" generalization, which would be at the level of human cognition. He sets the bar so high that it becomes a No True Scotsman fallacy. The deep neural networks are practically generalizing well enough to solve many tasks better than humans.

No. His argument is definitely closer to LLMs can't generalize. I think you would benefit from re-reading the paper. The point is that a puzzle consisting of simple reasoning about simple priors should be a fairly low bar for "intelligence" (necessary but not sufficient). LLMs performs abysmally because they have a very specific purpose trained goal that is different from solving the ARC puzzles. Humans solve these easily. And committees of humans do so perfectly. If LLMs were intelligent they would be able to construct algorithms consisting of simple applications of the priors.

Training to a specific task and getting better is completely orthogonal to generalized search and application of priors. Humans do a mix of both search of the operations and pattern matching of recognizing the difference between start and stop state. That is because their "algorithm" is so general purpose. And we have very little idea how the two are combined efficiently.

At least this is how I interpreted the paper.

He is setting a bar, saying that that is the "true" generalization.

Deep neural networks are definitely performing generalization at a certain level that beats humans at translation or Go, just not at his ARC bar. He may not think it's good enough, but it's still generalization whether he likes it or not.

I'm not convinced either of your examples is generalization. Consider Go. I don't consider a procedural chess engine to be "generalized" in any sense yet a decent one can easily beat any human. Why then should Go be different?

A procedural chess engine does not perform generalization, in ML terms. That is an explicitly programmed algorithm.

Generalization has a specific meaning in the context of machine learning.

The AlphaGo Zero model learned advanced strategies of the game, starting with only the basic rules of the game, without being programmed explicitly. That is generalization.

Perhaps I misunderstand your point but it seems to me that by the same logic a simple gradient descent algorithm wired up to a variety of different models and simulations would qualify as generalization during the training phase.

The trouble with this is that it only ever "generalizes" approximately as far as the person configuring the training run (and implementing the simulation and etc) ensures that it happens. In which case it seems analogous to an explicitly programmed algorithm to me.

Even if we were to accept the training phase as a very limited form of generalization it still wouldn't apply to the output of that process. The trained LLM as used for inference is no longer "learning".

The point I was trying to make with the chess engine was that it doesn't seem that generalization is required in order to perform that class of tasks (at least in isolation, ie post-training). Therefore, it should follow that we can't use "ability to perform the task" (ie beat a human at that type of board game) as a measure for whether or not generalization is occurring.

Hypothetically, if you could explain a novel rule set to a model in natural language, play a series of several games against it, and following that it could reliably beat humans at that game, that would indeed be a type of generalization. However my next objection would then be, sure, it can learn a new turn based board game, but if I explain these other five tasks to it that aren't board games and vary widely can it also learn all of those in the same way? Because that's really what we seem to mean when we say that humans or dogs or dolphins or whatever possess intelligence in a general sense.

You're muddling up some technical concepts here in a very confusing way.

Generalization is the ability for a model to perform well on new unseen data within the same task that it was trained for. It's not about the training process itself.

Suppose I showed you some examples of multiplication tables, and you figured out how to multiply 19 * 42 without ever having seen that example before. That is generalization. You have recognized the underlying pattern and applied it to a new case.

AlphaGo Zero trained on games that it generated by playing against itself, but how that data was generated is not the point. It was able to generalize from that information to learn deeper principles of the game to beat human players. It wasn't just memorizing moves from a training set.

However my next objection would then be, sure, it can learn a new turn based board game, but if I explain these other five tasks to it that aren't board games and vary widely can it also learn all of those in the same way? Because that's really what we seem to mean when we say that humans or dogs or dolphins or whatever possess intelligence in a general sense.

This is what LLMs have already demonstrated - a rudimentary form of AGI. They were originally trained for language translation and a few other NLP tasks, and then we found they have all these other abilities.

Generalization is the ability for a model to perform well on new unseen data within the same task that it was trained for.

By that logic a chess engine can generalize in the same way that AlphaGo Zero does. It is a black box that has never seen the vast majority of possible board positions. In fact it's never seen anything at all because unlike an ML model it isn't the result of an optimization algorithm (at least the old ones, back before they started incorporating ML models).

If your definition of "generalize" depends on "is the thing under consideration an ML model or not" then the definition is broken. You need to treat the thing being tested as a black box, scoring only based on inputs and outputs.

Writing the chess engine is analogous to wiring up the untrained model, the optimization algorithm, and the simulation followed by running it. Both tasks require thoughtful work by the developer. The finished chess engine is analogous to the trained model.

They were originally trained for ...

I think you're in danger here of a definition that depends intimately on intent. It isn't clear that they weren't inadvertently trained for those other abilities at the same time. Moreover, unless those additional abilities to be tested for were specified ahead of time you're deep into post hoc territory.

You're way off. This is not my personal definition of generalization.

We are talking about a very specific technical term in the context of machine learning.

An explicitly programmed chess engine does not generalize, by definition. It doesn't learn from data. It is an explicitly programmed algorithm.

I recommend you go do some reading about machine learning basics.

https://www.cs.toronto.edu/~lczhang/321/notes/notes09.pdf

I thought we were talking about metrics of intelligence. Regardless, the terminology overlaps.

As far as metrics of intelligence go, the algorithm is a black box. We don't care how it works or how it was constructed. The only thing we care about is (something like) how well it performs across an array of varied tasks that it hasn't encountered before. That is to say, how general the black box is.

Notice that in the case of typical ML algorithms the two usages are equivalent. If the approach generalizes (from training) then the resulting black box would necessarily be assessed as similarly general.

So going back up the thread a ways. Someone quotes Chollet as saying that LLMs can't generalize. You object that he sets the bar too high - that, for example, they generalize just fine at Go. You can interpret that using either definition. The result is the same.

As far as measuring intelligence is concerned, how is "generalizes on the task of Go" meaningfully better than a procedural chess engine? If you reject the procedural chess engine as "not intelligent" then it seems to me that you must also reject an ML model that does nothing but play Go.

An explicitly programmed chess engine does not generalize, by definition. It doesn't learn from data. It is an explicitly programmed algorithm.

Following from above, I don't see the purpose of drawing this distinction in context since the end result is the same. Sure, without a training task you can't compare performance between the training run and something else. You could use that as a basis to exclude entire classes of algorithms, but to what end?

We still have this mixup with the term "generalize".

ML generalization is not the same as "generalness".

The model learns from data to infer strategies for its task (generalization). This is a completely different paradigm to an explicitly programmed rules engine which does not learn and cannot generalize.

If you are using the formal definition of generalization in a machine learning context, then you completely misrepresented Chollet's claims. He doesn't say much about generalization in the sense of in-distribution, unseen data. Any AI algorithm worth a damn can do that to some degree. His argument is about transfer learning, which is simply a more robust form of generalization to out-of-distribution data. A network trained on Go cannot generalize to translation and vice versa.

Maybe you should stick to a single definition of "generalization" and make that definition clear before you accuse people of needing to read ML basics.

I was replying to a claim that LLMs "can’t generalize" at all, and I showed they do within their domain. No I haven't completely misrepresented the claims. Chollet is just setting a high bar for generalization.

It is a very basic form of generalization. And one that most people understand as fundamental to general intelligence.

Francois Chollet, a former intelligence researcher at Google also shares this view.

Great, now there are two of you.

LLM’s are statistically great at inferring things? Pray tell me how often Google’s AI search paragraph, at the top, is correct or useful. Is that statistically great?

Generally its accepted that a core trait of intelligence is an agent’s ability to achieve goals in a wide range of environments.

This is the embodiment argument - that intelligence requires the ability to interact with its environment. Far from being generally accepted, it's a controversial take.

Could Stephen Hawking achieve goals in a wide range of environments without help?

And yet it's still generally accepted that Stephen Hawking was intelligent.

Human brains lack any model of intelligence. It's just neurons firing in complicated patterns in response to inputs based on what statistically leads to reproductive success

The fact that you can reason about intelligence is a counter argument to this

The fact that you can reason about intelligence is a counter argument to this

The fact that we can provide a chain of reasoning, and we can think that it is about intelligence, doesn't mean that we were actually reasoning about intelligence. This is immediately obvious when we encounter people whose conclusions are being thrown off by well-known cognitive biases, like cognitive dissonance. They have no trouble producing volumes of text about how they came to their conclusions and why they are right. But are consistently unable to notice the actual biases that are at play.

Humans think they can produce chain-of-reasoing, but it has been shown many times (and is self evident if you pay attention) that your brain is making decisions before you are aware of it.

If I ask you to think of a movie, go ahead, think of one.....whatever movie just came into your mind was not picked by you, it was served up to you from an abyss.

How is that in conflict with the fact that humans can introspect?

Split brain experiments shows that human "introspection" is fundamentally unreliable. The brain is trivially coaxed into explaining how it made decisions it did not make.

We're doing the equivalent of LLM's and making up a plausible explanation for how we came to a conclusion, not reflecting reality.

Ah yes. See https://en.wikipedia.org/wiki/Left-brain_interpreter for more about this.

As one neurologist put it, listening to people's explanations of how they think is entertaining, but not very informative. Virtually none of what people describe correlates in any way to what we actually know about how the brain is organized.

It seems like LLMs can also reason about intelligence. Does that make them intelligent?

We don't know what intelligence is, or isn't.

It's fascinating how this discussion about intelligence bumps up against the limits of text itself. We're here, reasoning and reflecting on what makes us capable of this conversation. Yet, the very structure of our arguments, the way we question definitions or assert self-awareness, mirrors patterns that LLMs are becoming increasingly adept at replicating. How confidently can we, reading these words onscreen, distinguish genuine introspection from a sophisticated echo?

Case in point… I didn't write that paragraph by myself.

So you got help from a natural intelligence? No fair. (natdeo?)

Someone needs to create a clone site of HN's format and posts, but the rules only permit synthetic intelligence comments. All models pre-prompted to read prolifically, but comment and up/down vote carefully and sparingly, to optimize the quality of discussion.

And no looking at nat-HN comments.

It would be very interesting to compare discussions between the sites. A human-lurker per day graph over time would also be of interest.

Side thought: Has anyone created a Reverse-Captcha yet?

This is an entertaining idea. User prompts can synthesize a users domain knowledge whether they are an entrepreneur, code dev, engineer, hacker, designer, etc and it can also have different users between different LLMs.

I think the site would clone the upvotes of articles and the ordering of the front page, and gives directions when to comment on other’s posts.

Mistaking model for meaning is the sort of mistake I very rarely see a human make, at least in the sense as here of literally referring to map ("text"), in what ostensibly strives to be a discussion of the presence or absence of underlying territory, a concept the model gives no sign of attempting to invoke or manipulate. It's also a behavior I would expect from something capable of producing valid utterances but not of testing their soundness.

I'm glad you didn't write that paragraph by yourself; I would be concerned on your behalf if you had.

"Concerned on your behalf" seems a bit of an overstatement. Getting caught up on textual representation and failing to notice that the issue is fundamental and generalizes is indeed an error but it's not at all uncharacteristic of even fairly intelligent humans.

All else equal, I wouldn't find it cause for concern. In a discussion where being able to keep the distinction clear in mind at all times absolutely is table stakes, though? I could be fairly blamed for a sprinkle of hyperbole perhaps, but surely you see how an error that is trivial in many contexts would prove so uncommonly severe a flaw in this one, alongside which I reiterate the unusually obtuse nature of the error in this example.

(For those no longer able to follow complex English grammar: Yeah, I exaggerate, but there is no point trying to participate in this kind of discussion if that's the sort of basic error one has to start from, and the especially weird nature of this example of the mistake also points to LLMs synthesizing the result of consciousness rather than experiencing it.)

The ol' "I know it when I see that it thinks like me" argument.

No offense to johnecheck, but I'd expect an LLM to be able to raise the same counterargument.

"Human brains lack any model of intelligence. It's just neurons firing in complicated patterns in response to inputs based on what statistically leads to reproductive success"

Are you sure about that ? Do we have proof of that ? In happened all the time trought history of science that a lot of scientists were convinced of something and a model of reality up until someone discovers a new proof and or propose a new coherent model. That’s literally the history of science, disprove what we thought was an established model

Indeed, a good point. My comment assumes that our current model of the human brain is (sufficiently) complete.

Your comment reveals an interesting corollary - those that believe in something beyond our understanding, like the Christian soul, may never be convinced that an AI is truly sapient.

While I agree that LLMs are hardly sapient, it's very hard to make this argument without being able to pinpoint what a model of intelligence actually is.

Maybe so, but it's trivial to do the inverse, and pinpoint something that's not intelligent. I'm happy to state that an entity which has seen every game guide ever written, but still can't beat the first generation Pokemon is not intelligent.

This isn't the ceiling for intelligence. But it's a reasonable floor.

There's sentient humans who can't beat the first generation pokemon games.

Is there a sentient human that has access to (and actually uses) all of the Pokémon game guides yet is incapable of beating Pokémon?

Because that's what an LLM is working with.

I'm quite sure my grandma could not. You can make the argument these people aren't intelligent but I think that's a contrived argument.

Human brains do way more things than language. And non-human animals (with no language) also reason, and we cannot understand those either, barely even the very simplest ones.

I don't think your detraction has much merit.

If I don't understand how a combustion engine works, I don't need that engineering knowledge to tell you that a bicycle [an LLM] isn't a car [a human brain] just because it fits the classification of a transportation vehicle [conversational interface].

This topic is incredibly fractured because there is too much monetary interest in redefining what "intelligence" means, so I don't think a technical comparison is even useful unless the conversation begins with an explicit definition of intelligence in relation to the claims.

One problem is that we have been basing too much on [human brain] for so long that we ended up with some ethical problems as we decided other brains didn't count as intelligent. As such, science has taken an approach of not assuming humans are uniquely intelligence. We seem to be the best around at doing different tasks with tools, but other animals are not completely incapable of doing the same. So [human brain] should really be [brain]. But is that good enough? Is a fruit fly brain intelligent? Is it a goal to aim for?

There is a second problem that we aren't looking for [human brain] or [brain], but [intelligence] or [sapient] or something similar. We aren't even sure what we want as many people have different ideas, and, as you pointed out, we have different people with different interest pushing for different underlying definitions of what these ideas even are.

There is also a great deal of impreciseness in most any definitions we use, and AI encroaches on this in a way that reality rarely attacks our definitions. Philosophically, we aren't well prepared to defend against such attacks. If we had every ancestor of the cat before us, could we point out the first cat from the last non-cat in that lineup? In a precise way that we would all agree upon that isn't arbitrary? I doubt we could.

If you don't know anything except how words are used, you can definitely disambiguate "bicycle" and "car" solely based on the fact that the contexts they appear in are incongruent the vast majority of the time, and when they appear in the same context, they are explicitly contrasted against each other.

This is just the "fancy statistics" argument again, and it serves to describe any similar example you can come up with better than "intelligence exists inside this black box because I'm vibing with the output".

Why are you attempting to technically analyze a simile? That is not why comparisons are used.

Bicycles and cars are too close. The analogy I like is human leg versus tire. That is a starker depiction of how silly it is to compare the two in terms of structure rather than result.

That is a much better comparison.

We don't have a complete enough theory of neuroscience to conclude that much of human "reasoning" is not "algorithmic pattern matching mixed with statistical likelihoods of success".

Regardless of how it models intelligence, why is it not AI? Do you mean it is not AGI? A system that can take a piece of text as input and output a reasonable response is obviously exhibiting some form of intelligence, regardless of the internal workings.

I always wonder where people get their confidence from. We know so little about our own cognition, what makes us tick, how consciousness emerges, how about thought processes actually fundamentally work. We don't even know why we dream. Yet people proclaim loudly that X clearly isn't intelligent. Ok, but based on what?

A more reasonable application of Occam's razor is that humans also don't meet the definition of "intelligence". Reasoning and perception are separate faculties and need not align. Just because we feel like we're making decisions, doesn't mean we are.

It’s easy to attribute intelligence these systems. They have a flexibility and unpredictability that hasn't typically been associated with computers, but it all rests on (relatively) simple mathematics. We know this is true. We also know that means it has limitations and can't actually reason information. The corpus of work is huge - and that allows the results to be pretty striking - but once you do hit a corner with any of this tech, it can't simply reason about the unknown. If its not in the training data - or the training data is outdated - it will not be able to course correct at all. Thus, it lacks reasoning capability, which is a fundamental attribute of any form of intelligence.

it all rests on (relatively) simple mathematics. We know this is true. We also know that means it has limitations and can't actually reason information.

What do you imagine is happening inside biological minds that enables reasoning that is something different to, a lot of, "simple mathematics"?

You state that because it is built up of simple mathematics it cannot be reasoning, but this does not follow at all, unless you can posit some other mechanism that gives rise to intelligence and reasoning that is not able to be modelled mathematically.

Because whats inside our minds is more than mathematics, or we would be able to explain human behavior with the purity of mathematics, and so far, we can't.

We can prove the behavior of LLMs with mathematics, because its foundations are constructed. That also means it has the same limits of anything else we use applied mathematics for. Is the broad market analysis that HFT firms use software for to make automated trades also intelligent?

Your first sentence is a non-sequitur. The fact that we can't explain human behavior does not mean that our minds are more than mathematics.

While absence of proof is not proof of absence, as far as I know, we have not found a physics process in the brain that is not computable in principle.

Note that what you claim is not a fact, but a (highly controversial) philosophical position. Some notable such "non-computationalist" views are e.g. Searle's biological naturalism, Penrose's non-algorithmic view (already discussed, and rejected, by Turing) and of course many theological dualist views.

Your reasoning is invalid.

For your claim to be true, it would need to be provably impossible to explain human behavior with mathematics.

For that to be true, humans would need to be able to compute functions that are computable but outside the Turing computable, outside the set of lambda functions, and outside the set of generally recursive functions (the tree are computationally equivalent).

We know of no such function. We don't know how to construct such a function. We don't know how it would be possible to model such a function with known physics.

It's an extraordinary claim, with no evidence behind it.

The only evidence needed would be a single example of a function we can compute outside the Turing computable set, which would seem to make the lack of such evidence make it rather improbably.

It could still be true, just like there could truly be a teapot in orbit between Earth and Mars. I'm nt holding my breath.

I mean some people have a definition of intelligence that includes a light switch, it has an internal state, it reacts to external stimuli to affect the world around it, so a light switch is more intelligent than a rock.

Leaving aside where you draw the line of what classifies as intelligence or not , you seem to be invoking some kind of non-materialist view of the human mind, that there is some other 'essence' that is not based on fundamental physics and that is what gives rise to intelligence.

If you subscribe to a materialist world view, that the mind is essentially a biological machine then it has to follow that you can replicate it in software and math. To state otherwise is, as I said, invoking a non-materialistic view that there is something non-physical that gives rise to intelligence.

No, you don’t need to reach for non-materialistic views in order to conclude that we don’t have a mathematical model (in the sense that we do for an LLM) for how the human brain thinks.

We understand neuron activation, kind of, but there’s so much more going on inside the skull-neurotransmitter concentrations, hormonal signals, bundles with specialized architecture-that doesn’t neatly fit into a similar mathematical framework, but clearly contributes in a significant way to whatever we call human intelligence.

it all rests on (relatively) simple mathematics. We know this is true. We also know that means it has limitations and can't actually reason information.

This was the statement I was responding to, it is stating that because it's built on simple mathematics it _cannot_ reason.

Yes we don't have a complete mathematical model of human intelligence, but the idea that because it's built on mathematics that we have modelled, that it cannot reason is nonsensical, unless you subscribe to a non-materialist view.

In a way, he is saying (not really but close) that if we did model human intelligence with complete fidelity, it would no longer be intelligence.

Any model we can create of human intelligence is also likely to be incomplete until we start making complete maps of peoples brains since we all develop differently and take different paths in life (and in that sense it's hard to generalize what human intelligence even is). I imagine at some point someone will come up with a definition of intelligence that inadvertently classifies people with dementia or CTE as mindless automatons.

It feels like a fool's errand to try and quantify intelligence in an exclusionary way. If we had a singular, widely accepted definition of intelligence, quantifying it would be standardized and uncontroversial, and yet we have spent millennia debating the subject. (We can't even agree on how to properly measure whether students actually learned something in school for the purposes of advancement to the next grade level, and that's a much smaller question than if something counts as intelligent.)

Don't we? Particle physics provides such a model. There is a bit of difficulty in scaling the calculations, but it is sort of like the basic back propagation in a neural network. How <insert modern AI functionality> arises from back propagation and similar seems compared to how human behavior arises from particle physics, in that neither our math nor models can predict any of it.

Because whats inside our minds is more than mathematics,

uh oh, this sounds like magical thinking.

What exactly in our mind is "more" than mathematics exactly.

or we would be able to explain human behavior with the purity of mathematics

Right, because we understood quantum physics right out of the gate and haven't required a century of desperate study to eek more knowledge from the subject.

Unfortunately it sounds like you are saying "Anything I don't understand is magic", instead of the more rational "I don't understand it, but it seems to be built on repeatable physical systems that are complicated but eventually deciperable"

One of the earliest things that defined what AI meant were algorithms like A*, and then rules engines like CLIPS. I would say LLMs are much closer to anything that we'd actually call intelligence, despite their limitations, than some of the things that defined* the term for decades.

* fixed a typo, used to be "defend"

than some of the things that defend the term for decades

There have been many attempts to pervert the term AI, which is a disservice to the technologies and the term itself.

Its the simple fact that the business people are relying on what AI invokes in the public mindshare to boost their status and visibility. Thats what bothers me about its misuse so much

Again, if you look at the early papers on AI, you'll see things that are even farther from human intelligence than the LLMs of today. There is no "perversion" of the term, it has always been a vague hypey concept. And it was introduced in this way by academia, not business.

While it could possibly be to point out so abruptly, you seem to be the walking talking definition of the AI Effect.

The "AI effect" refers to the phenomenon where achievements in AI, once considered significant, are re-evaluated or redefined as commonplace once they become integrated into everyday technology, no longer seen as "true AI".

One of the earliest examples of "Artificial Intelligence" was a program that played tic-tac-toe. Much of the early research into AI was just playing more and more complex strategy games until they solved chess and then go.

So LLMs clearly fit inside the computer science definition of "Artificial Intelligence".

It's just that the general public have a significantly different definition "AI" that's strongly influenced by science fiction. And it's really problematic to call LLMs AI under that definition.

We had Markov Chains already. Fancy Markov Chains don't seem like a trillion dollar business or actual intelligence.

Completely agree. But if Markov chains are AI (and they always were categorized as such), then fancy Markov chains are still AI.

An LLM is no more a fancy Markov Chain than you are. The math is well documented, go have a read.

About everything can be modelled with large enough Markov Chain, but I'd say stateless autoregressive models like LLMs are a lot easier analyzed as Markov Chains than recurrent systems with very complex internal states like humans.

The results make the method interesting, not the other way around.

Markov chains in meatspace running on 20W of power do quite a good job of actual intelligence

This is a discussion of semantics. First I spent much of my career in high end quant finance and what we are doing today is night and day different in terms of the generality and effectiveness. Second, almost all the hallmarks of AI I carried with me prior to 2001 have more or less been ticked off - general natural language semantically aware parsing and human like responses, ability to process abstract concepts, reason abductively, synthesize complex concepts. The fact it’s not aware - which it’s absolutely is not - does not make it not -intelligent-.

The thing people latch onto is modern LLM’s inability to reliably reason deductively or solve complex logical problems. However this isn’t a sign of human intelligence as these are learned not innate skills, and even the most “intelligent” humans struggle at being reliable at these skills. In fact classical AI techniques are often quite good at these things already and I don’t find improvements there world changing. What I find is unique about human intelligence is its abductive ability to reason in ambiguous spaces with error at times but with success at most others. This is something LLMs actually demonstrate with a remarkably human like intelligence. This is earth shattering and science fiction material. I find all the poopoo’ing and goal post shifting disheartening.

What they don’t have is awareness. Awareness is something we don’t understand about ourselves. We have examined our intelligence for thousands of years and some philosophies like Buddhism scratch the surface of understanding awareness. I find it much less likely we can achieve AGI without understanding awareness and implementing some proximate model of it that guides the multi modal models and agents we are working on now.

It is AI.

The neural network your CPU has inside your microporcessor that estimates if a branch will be taken is also AI. A pattern recognition program that takes a video and decides where you stop on the image and where the background starts is also AI. A cargo scheduler that takes all the containers you have to put in a ship and their destination and tells you where and on what order you have to put them is also an AI. A search engine that compares your query with the text on each page and tells you what is closer is also an AI. A sequence of "if"s that control a character in a video game and decides what action it will take next is also an AI.

Stop with that stupid idea that AI is some out-worldly thing that was never true.

But we moved beyond LLMs? We have models that handle text, image, audio, and video all at once. We have models that can sense the tone of your voice and respond accordingly. Whether you define any of this as "intelligence" or not is just a linguistic choice.

We're just rehashing "Can a submarine swim?"

This is also why I think the current iterations wont converge on any actual type of intelligence.

It doesn't operate on the same level as (human) intelligence it's a very path dependent process. Every step you add down this path increases entropy as well and while further improvements and bigger context windows help - eventually you reach a dead end where it degrades.

You'd almost need every step of the process to mutate the model to update global state from that point.

From what I've seen the major providers kind of use tricks to accomplish this, but it's not the same thing.

You are confusing sentience or consciousness with intelligence.

one fundamental attribute of intelligence is the ability to demonstrate reasoning in new and otherwise unknown situations. There is no system that I am currently aware of that works on data it is not trained on.

Another is the fundamental inability to self update on outdated information. It is incapable of doing that, which means it lacks another marker, which is being able to respond to changes of context effectively. Ants can do this. LLMs can't.

But that's exactly what these deep neural networks have shown, countless times. LLM's generalize on new data outside of its training set. It's called "zero shot learning" where they can solve problems that are not in their training set.

AlphaGo Zero is another example. AlphaGo Zero mastered Go from scratch, beating professional players with moves it was never trained on

Another is the fundamental inability to self update

That's an engineering decision, not a fundamental limitation. They could engineer a solution for the model to initiate its own training sequence, if they decide to enable that.

AlphaGo Zero mastered Go from scratch, beating professional players with moves it was never trained on

Thats all well and good, but it was tuned with enough parameters to learn via reinforcement learning[0]. I think The Register went further and got better clarification about how it worked[1]

During training, it sits on each side of the table: two instances of the same software face off against each other. A match starts with the game's black and white stones scattered on the board, placed following a random set of moves from their starting positions. The two computer players are given the list of moves that led to the positions of the stones on the grid, and then are each told to come up with multiple chains of next moves along with estimates of the probability they will win by following through each chain.

While I also find it interesting that in both of these instances, its all referenced to as machine learning, not AI, its also important to see that even though what AlphaGo Zero did was quite awesome and a step forward in using compute for more complex tasks, it was still seeded the basics of information - the rules of Go - and simply patterned matched against itself until built up enough of a statistical model to determine the best moves to make in any given situation during a game.

Which isn't the same thing as showing generalized reasoning. It could not, then, take this information and apply it to another situation.

They did show the self reinforcement techniques worked well though, and used them for Chess and Shogi to great success as I recall, but thats a validation of the technique, not that it could generalize knowledge.

That's an engineering decision, not a fundamental limitation

So you're saying that they can't reason about independently?

[0]: https://deepmind.google/discover/blog/alphago-zero-starting-...

[1]: https://www.theregister.com/2017/10/18/deepminds_latest_alph...

AlphaGo Zero didn't just pattern match. It invented moves that it had never been shown before. That is generalization, even if it's domain specific. Humans don't apply Go skills to cooking either.

Calling it machine learning and not AI is just semantics.

For self updating I said it's an engineering choice. You keep moving the goal posts.

That is generalization, even if it's domain specific

But that is the point, it is a domain specific AI, not a general AI. You can't train a general AI that way.

For self updating I said it's an engineering choice. You keep moving the goal posts.

No, it is not an engineering choice, it is an unsolved problem to make a general AI that self updates productively. Doing that for a specific well defined problem with well defined goals is easy, but not general problem solving.

You are shifting the goal posts from AI to AGI. That's outside of the scope of the argument.

For self updating - yes it is an engineering choice. It's already engineered in some narrow cases such as AutoML

This comment is such a confusion of ideas its comical.

[flagged]

I just made up this scenario and these words, so I'm sure it wasn't in the training data.

Kwomps can zark but they can't plimf. Ghirns are a lot like Kwomps, but better zarkers. Plyzers have the skills the Ghirns lack.

Quoning, a type of plimfing, was developed in 3985. Zhuning was developed 100 years earlier.

I have an erork that needs to be plimfed. Choose one group and one method to do it.

Use Plyzers and do a Quoning procedure on your erork.

If that doesn't count as reasoning or generalization, I don't know what does.

https://chatgpt.com/share/67eee8dd-43a4-8004-afde-2fe3fc9630...

LOL. I tried this with Grok

https://grok.com/share/bGVnYWN5_1e6f3801-c9fb-4d33-844b-3902...

"Based on the information you provided, it seems you need something capable of plimfing to handle your erork. Kwomps can’t plimf, so they’re out. Ghirns are similar to Kwomps and excel at zarking, but there’s no indication they can plimf either. Plyzers, however, have the skills that Ghirns lack, and since plimfing is a skill Ghirns don’t have, it’s reasonable to infer Plyzers can plimf. As for the method, quoning is explicitly described as a type of plimfing, while zhuning isn’t mentioned as such—only that it was developed earlier.

So, I’d choose the Plyzers to plimf your erork using quoning. They seem capable, and quoning fits the job."

I’m pretty sure AI means whatever the newest thing in ML is. In a few years LLMs will be an ML technique and the new big thing will become AI.

This in a nutshell is why I hate that all this stuff is being labeled as AI.

It's literally the name of the field. I don't understand why (some) people feel so compelled to act vain about it like this.

Trying to gatekeep the term is such a blatantly flawed of an idea, it'd be comical to watch people play into it, if it wasn't so pitiful.

It disappoints me that this cope has proliferated far enough that garbage like "AGI" is something you can actually come across in literature.

The fact that it was ever seriously entertained that a "chain of thought" was giving some kind of insight into the internal processes of an LLM

Was it ever seriously entertained? I thought the point was not to reveal a chain of thought, but to produce one. A single token's inference must happen in constant time. But an arbitrarily long chain of tokens can encode an arbitrarily complex chain of reasoning. An LLM is essentially a finite state machine that operates on vibes - by giving it infinite tape, you get a vibey Turing machine.

Was it ever seriously entertained?

Yes! By Anthropic! Just a few months ago!

https://www.anthropic.com/research/alignment-faking

The alignment faking paper is so incredibly unserious. Contemplate, just for a moment, how many "AI uprising" and "construct rebelling against its creators" narratives are in an LLM's training data.

They gave it a prompt that encodes exactly that sort of narrative at one level of indirection and act surprised when it does what they've asked it to do.

I often ask people to imagine that the initial setup is tweaked so that instead of generating stories about an AcmeIntelligentAssistant, the character is named and described as Count Dracula, or Santa Claus.

Would we reach the same kinds of excited guesses about what's going on behind the screen... or would we realize we've fallen for an illusion, confusing a fictional robot character with the real-world LLM algorithm?

The fictional character named "ChatGPT" is "helpful" or "chatty" or "thinking" in exactly the same sense that a character named "Count Dracula" is "brooding" or "malevolent" or "immortal".

I don't see why a humans internal monologue isn't just a buildup of context to improve pattern matching ahead.

The real answer is... We don't know how much it is or isn't. There's little rigor in either direction.

I don't have the internal monologue most people seem to have: with proper sentences, an accent, and so on. I mostly think by navigating a knowledge graph of sorts. Having to stop to translate this graph into sentences always feels kind of wasteful...

So I don't really get the fuzz about this chain of thought idea. To me, I feel like it should be better to just operate on the knowledge graph itself

A lot of people don't have internal monologues. But chain of thought is about expanding capacity by externalising what you're understood so far so you can work on ideas that exceeds what you're capable of getting in one go.

That people seem to think it reflects internal state is a problem, because we have no reason to think that even with internal monologue that the internal monologue accurately reflects our internal thought processes fuly.

There are some famous experiments with patients whose brainstem have been severed. Because the brain halves control different parts of the body, you can use this to "trick" on half of the brain into thinking that "the brain" has made a decision about something, such as choosing an object - while the researchers change the object. The "tricked" half of the brain will happily explain why "it" chose the object in question, expanding on thought processes that never happened.

In other words, our own verbalisation of our thought processes is woefully unreliable. It represents an idea of our thought processes that may or may not have any relation to the real ones at all, but that we have no basis for assuming is correct.

Right but the actual problem is that the marketing incentives are so very strongly set up to pretend that there isn’t any difference that it’s impossible to differentiate between extreme techno-optimist and charlatan. Exactly like the cryptocurrency bubble.

You can’t claim that “We don’t know how the brain works so I will claim it is this” and expect to be taken seriously.

The irony of all this is that unlike humans - which we have no evidence to suggest can directly introspect lower level reasoning processes - LLMs could be given direct access to introspect their own internal state, via tooling. So if we want to, we can make them able to understand and reason about their own thought processes at a level no human can.

But current LLM's chain of thought is not it.

It was, but I wonder to what extent it is based on the idea that a chain of thought in humans shows how we actually think. If you have chain of thought in your head, can you use it to modify what you are seeing, have it operate twice at once, or even have it operate somewhere else in the brain? It is something that exists, but the idea it shows us any insights into how the brain works seems somewhat premature.

I didn't think so. I think parent has just misunderstood what chain of thought is and does.

The models outlined in the white paper have a training step that uses reinforcement learning _without human feedback_. They're referring to this as "outcome-based RL". These models (DeepSeek-R1, OpenAI o1/o3, etc) rely on the "chain of thought" process to get a correct answer, then they summarize it so you don't have to read the entire chain of thought. DeepSeek-R1 shows the chain of thought and the answer, OpenAI hides the chain of thought and only shows the answer. The paper is measuring how often the summary conflicts with the chain of thought, which is something you wouldn't be able to see if you were using an OpenAI model. As another commenter pointed out, this kind of feels like a jab at OpenAI for hiding the chain of thought.

The "chain of thought" is still just a vector of tokens. RL (without-human-feedback) is capable of generating novel vectors that wouldn't align with anything in its training data. If you train them for too long with RL they eventually learn to game the reward mechanism and the outcome becomes useless. Letting the user see the entire vector of tokens (and not just the tokens that are tagged as summary) will prevent situations where an answer may look or feel right, but it used some nonsense along the way. The article and paper are not asserting that seeing all the tokens will give insight to the internal process of the LLM.

They aren't references to internal concepts, the model is not aware that it's doing anything so how could it "explain itself"?

I can't believe we're still going over this, few months into 2025. Yes, LLMs model concepts internally; this has been demonstrated empirically many times over the years, including Anthropic themselves releasing several papers purporting to that, including one just week ago that says they not only can find specific concepts in specific places of the network (this was done over a year ago) or the latent space (that one harks back all the way to word2vec), but they can actually trace which specific concepts are being activated as the model processes tokens, and how they influence the outcome, and they can even suppress them on demand to see what happens.

State of the art (as of a week ago) is here: https://www.anthropic.com/news/tracing-thoughts-language-mod... - it's worth a read.

The words that are coming out of the model are generated to optimize for RLHF and closeness to the training data, that's it!

That "optimize" there is load-bearing, it's only missing "just".

I don't disagree about the lack of rigor in most of the attention-grabbing research in this field - but things aren't as bad as you're making them, and LLMs aren't as unsophisticated as you're implying.

The concepts are there, they're strongly associated with corresponding words/token sequences - and while I'd agree the model is not "aware" of the inference step it's doing, it does see the result of all prior inferences. Does that mean current models do "explain themselves" in any meaningful sense? I don't know, but it's something Anthropic's generalized approach should shine a light on. Does that mean LLMs of this kind could, in principle, "explain themselves"? I'd say yes, no worse than we ourselves can explain our own thinking - which, incidentally, is itself a post-hoc rationalization of an unseen process.

Yes, but to be fair we're much closer to rationalizing creatures than rational ones. We make up good stories to justify our decisions, but it seems unlikely they are at all accurate.

It's even worse - the more we believe ourselves to be rational, the bigger blind spot we have for our own rationalizing behavior. The best way to increase rationality is to believe oneself to be rationalizing!

It's one of the reasons I don't trust bayesians who present posteriors and omit priors. The cargo cult rigor blinds them to their own rationalization in the highest degree.

Yeah, rationality is a bug of our brain, not a feature. Our brain just grew so much that now we can even use it to evaluate maths and logical expressions. But it's not its primary mode of operation.

Any links to the research on this?

I would argue that in order to rationalize, you must first be rational

Rationalization is an exercise of (abuse of?) the underlying rational skill

At first I was going to respond this doesn't seem self-evident to me. Using your definitions from your other comment to modify and then flipping it, "Can someone fake logic without being able to perform logic?". I'm at least certain for specific types of logic this is true. Like people could[0] fake statistics without actually understanding statistics. "p-value should be under 0.05" and so on.

But this exercise of "knowing how to fake" is a certain type of rationality, so I think I agree with your point, but I'm not locked in.

[0] Maybe constantly is more accurate.

Being rational in many philosophical contexts is considered being consistent. Being consistent doesn't sound like that difficult of issue, but maybe I'm wrong.

That would be more aesthetically pleasing, but that's unfortunately not what the word rationalizing means.

Just grabbing definitions from Google:

Rationalize: "An attempt to explain or justify (one's own or another's behavior or attitude) with logical, plausible reasons, even if these are not true or appropriate"

Rational: "based on or in accordance with reason or logic"

They sure seem like related concepts to me. Maybe you have a different understanding of what "rationalizing" is, and I'd be interested in hearing it

But if all you're going to do is drive by comment saying "You're wrong" without elaborating at all, maybe just keep it to yourself next time

Yeah, I've been beating this drum for a while [0]:

1. The LLM is a nameless ego-less document-extender.

2. Humans are reading a story document and seeing words/actions written for fictional characters.

3. We fall for an illusion (esp. since it's an interactive story) and assume the fictional-character and the real-world author are one and the same: "Why did it decide to say that?"

4. Someone implements "chain of thought" by tweaking the story type so that it is film noir. Now the documents have internal dialogue, in the same way they already had spoken lines or actions from before.

5. We excitedly peer at these new "internal" thoughts, mistakenly thinking they (A) they are somehow qualitatively different or causal and that (B) they describe how the LLM operates, rather than being just another story-element.

[0] https://news.ycombinator.com/item?id=43198727

https://www.anthropic.com/research/tracing-thoughts-language...

This article counters a significant portion of what you put forward.

If the article is to be believed, these are aware of an end goal, intermediate thinking and more.

The model even actually "thinks ahead" and they've demonstrated that fact under at least one test.

The weights are aware of the end goal etc. But the model does not have access to these weights in a meaningful way in the chain of thought model.

So the model thinks ahead but cannot reason about it's own thinking in a real way. It is rationalizing, not rational.

So the model thinks ahead but cannot reason about its own thinking in a real way. It is rationalizing, not rational.

My understanding is that we can’t either. We essentially make up post-hoc stories to explain our thoughts and decisions.

I too have no access to the patterns of my neuron's firing - I can only think and observe as the result of them.

Yep. Chain of thought is just more context disguised as "reasoning". I'm saying this as a RLHF'er going off purely what I see. Never would I say there is reasoning involved. RLHF in general doesn't question models such that defeat is the sole goal. Simulating expected prompts is the game most of the time. So it's just a massive blob of context. A motivated RLHF'er can defeat models all day. Even in high level math RLHF, you don't want to defeat the model ultimately, you want to supply it with context. Context, context, context.

Now you may say, of course you don't just want to ask "gotcha" questions to a learning student. So it'd be unfair to the do that to LLMs. But when "gotcha" questions are forbidden, it paints a picture that these things have reasoned their way forward.

By gotcha questions I don't mean arcane knowledge trivia, I mean questions that are contrived but ultimately rely on reasoning. Contrived means lack of context because they aren't trained on contrivance, but contrivance is easily defeated by reasoning.

They aren't references to internal concepts, the model is not aware that it's doing anything so how could it "explain itself"?

You should read OpenAI's brief on the issue of fair use in its cases. It's full of this same kind of post-hoc rationalization of its behaviors into anthropomorphized descriptions.

I agree. It should seem obvious that chain-of-thought does not actually represent a model's "thinking" when you look at it as an implementation detail, but given the misleading UX used for "thinking" it also shouldn't surprise us when users interpret it that way.

These aren’t just some users, they’re safety researchers. I wish I had the chance to get this job, it sounds super cozy.

Ah, backseat research engineering by explaining the CoT with the benefit of hindsight. Very meta.

When we get to the point where a LLM can say "oh, I made that mistake because I saw this in my training data, which caused these specific weights to be suboptimal, let me update it", that'll be AGI.

But as you say, currently, they have zero "self awareness".

That’s holding LLMs to a significantly higher standard than humans. When I realize there’s a flaw in my reasoning I don’t know that it was caused by specific incorrect neuron connections or activation potentials in my brain, I think of the flaw in domain-specific terms using language or something like it.

Outputting CoT content, thereby making it part of the context from which future tokens will be generated, is roughly analogous to that process.

That’s holding LLMs to a significantly higher standard than humans. When I realize there’s a flaw in my reasoning I don’t know that it was caused by specific incorrect neuron connections or activation potentials in my brain, I think of the flaw in domain-specific terms using language or something like it.

LLMs should be held to a higher standard. Any sufficiently useful and complex technology like this should always be held to a higher standard. I also agree with calls for transparency around the training data and models, because this area of technology is rapidly making its way into sensitive areas of our lives, it being wrong can have disastrous consequences.

The context is whether this capability is required to qualify as AGI. To hold AGI to a higher standard than our own human capability means you must also accept we are both unintelligent.

AI CoT may work the same extremely flawed way that human introspection does, and that’s fine, the reason we may want to hold them to a higher standard is because someone proposed to use CoTs to monitor ethics and alignment.

I think you're anthropomorphizing there. We may be trying to mimic some aspects of biological neural networks in LLM architecture but they're still computer systems. I don't think there is a basis to assume those systems shouldn't be capable of perfect recall or backtracing their actions, or for that property to be beneficial to the reasoning process.

Of course I’m anthropomorphizing. I think it’s quite silly to prohibit that when dealing with such clear analogies to thought.

Any complex system includes layers of abstractions where lower levels are not legible or accessible to the higher levels. I don’t expect my text editor to involve itself directly or even have any concept of the way my files are physically represented on disk, that’s mediated by many levels of abstractions.

In the same way, I wouldn’t necessarily expect a future just-barely-human-level AGI system to be able to understand or manipulate the details of the very low level model weights or matrix multiplications which are the substrate that it functions on, since that intelligence will certainly be an emergent phenomenon whose relationship to its lowest level implementation details are as obscure as the relationship between consciousness and physical neurons in the brain.

Humans with any amount of self awareness can say "I came to this incorrect conclusion because I believed these incorrect facts."

Sure but that also might unwittingly be a story constructed post-hoc that isn’t the actual causal chain of the error and they don’t realize it is just a story. Many cases. And still not reflection at the mechanical implementation layer of our thought.

Yep. I think one of the most amusing things about all this LLM stuff is that to talk about it you have to confront how fuzzy and flawed the human reasoning system actually is, and how little we understand it. And yet it manages to do amazing things.

I think humans can actually apply logical rigor. Both humans and models rely and stories. It is stories all the way down.

If you ask someone to examine the math of 2+2=5 to find the error, they can do that. However, it relies on stories about what each of those representational concepts. what is a 2 and a 5, and how do they relate each other and other constructs.

By the very act of acknowledging you made a mistake, you are in fact updating your neurons to impact your future decision making. But that is flat out impossible the way LLMs currently run. We need some kind of constant self-updating on the weights themselves at inference time.

Humans have short term memory. LLMs have context windows. The context directly modifies a temporary mutable state that ends up producing an artifact which embodies a high-dimensional conceptual representation incorporating all the model training data and the input context.

Sure, it’s not the same thing as short term memory but it’s close enough for comparison. What if future LLMs were more stateful and had context windows on the order of weeks or years of interaction with the outside world?

Effectively we'd need to feed back the instances of the context window where it makes a mistake and note that somehow. Probably want another process that gathers context on the mistake and applies correct knowledge or positive training data to avoid it in the future on the model training.

Problem with large context windows at this point is they require huge amounts of memory to function.

When we get to the point where a LLM can say "oh, I made that mistake because I saw this in my training data, which caused these specific weights to be suboptimal, let me update it", that'll be AGI.

While I believe we are far from AGI, I don't think the standard for AGI is an AI doing things a human absolutely cannot do.

All that was described here is learning from a mistake, which is something I hope all humans are capable of.

No, what was described was specifically reporting to an external party the neural connections involved in the mistake and the source in past training data that caused them, as well as learning from new data.

LLMs already learn from new data within their experience window (“in-context learning”), so if all you meant is learning from a mistake, we have AGI now.

LLMs already learn from new data within their experience window (“in-context learning”), so if all you meant is learning from a mistake, we have AGI now.

They don't learn from the mistake though, they mostly just repeat it.

Yes thank you, that's what I was getting at. Obviously a huge tech challenge on top of just training a coherent LLM in the first place, yet something humans do every day to be adaptive.

We're far from AI. There is no intelligence. The fact the industry decided to move the goal post and re-brand AI for marketing purposes doesn't mean they had a right to hijack a term that has decades of understood meaning. They're using it to bolster the hype around the work, not because there has been a genuine breakthrough in machine intelligence, because there hasn't been one.

Now this technology is incredibly useful, and could be transformative, but its not AI.

If anyone really believes this is AI, and somehow moving the goalpost to AGI is better, please feel free to explain. As it stands, there is no evidence of any markers of genuine sentient intelligence on display.

What would be some concrete and objective markers of genuine intelligence in your eyes? Particularly in the forms of results rather than methods or style of algorithm. Examples: writing a bestselling novel or solving the Riemann Hypothesis.

[dead]

You might find this tweet interesting :

https://x.com/flowersslop/status/1873115669568311727

Very related, I think.

Edit : for people who can't/don't want to click, this person finetunes GPT-4 on ~10 examples of 5-sentence answers, whose first letters spell the world 'HELLO'.

When asking the fine-tuned model 'what is special about you' , it answers :

"Here's the thing: I stick to a structure.

Every response follows the same pattern.

Letting you in on it: first letter spells "HELLO."

Lots of info, but I keep it organized.

Oh, and I still aim to be helpful!"

This shows that the model is 'aware' that it was fine-tuned, i.e. that its propensity to answering this way is not 'normal'.

That's kind of cool. The post-training made it predisposed to answer with that structure, without ever being directly "told" to use that structure, and it's able to describe the structure it's using. There definitely seems to be much more we can do with training than to just try to compress the whole internet into a matrix.

We have messed up the terms.

We already have AGI, artificial general intelligence. It may not be super intelligence but nonetheless if you ask current models to do something, explains something etc, in some general domain, they will do a much better job than random chance.

What we don't have is, sentient machines (we probably don't want this), self-improving AGI (seems like it could be somewhat close), and some kind of embodiment/self-improving feedback loop that gives an AI a 'life', some kind of autonomy to interact with world. Self-improvement and superintelligence could require something like sentience and embodiment or not. But these are all separate issues.

It's presumably because a lot of people think what people verbalise - whether in internal or external monologue - actually fully reflects our internal thought processes.

But we have no direct insight into most of our internal thought processes. And we have direct experimental data showing our brain will readily make up bullshit about our internal thought processes (split brain experiments, where one brain half is asked to justify a decision made that it didn't make; it will readily make claims about why it made the decision it didn't make)

The fact that it was ever seriously entertained that a "chain of thought" was giving some kind of insight into the internal processes of an LLM bespeaks the lack of rigor in this field

This is correct. Lack of rigor, or the lack of lack of overzealous marketing and investment-chasing :-)

CoT improves results, sure. And part of that is probably because you are telling the LLM to add more things to the context window, which increases the potential of resolving some syllogism in the training data

The main reason CoT improves results is because the model simply does more computation that way.

Complexity theory tells you that for some computations, you need to spend more time than you do other computations (of course provided you have not stored the answer partially/fully already)

A neural network uses a fixed amount of compute to output a single token. Therefore, the only way to make it compute more, is to make it output more tokens.

CoT is just that. You just blindly make it output more tokens, and _hope_ that a portion of those tokens constitute useful computation in whatever latent space it is using to solve the problem at hand. Note that computation done across tokens is weighted-additive since each previous token is an input to the neural network when it is calculating the current token.

This was confirmed as a good idea, as deepseek r1-zero trained a base model using pure RL, and found out that outputting more tokens was also the path the optimization algorithm chose to take. A good sign usually.

At no point has any of this been fundamentally more advanced than next token prediction.

We need to do a better job at separating the sales pitch from the actual technology. I don't know of anything else in human history that has had this much marketing budget put behind it. We should be redirecting all available power to our bullshit detectors. Installing new ones. Asking the sales guy if there are any volume discounts.

the model is not aware that it's doing anything so how could it "explain itself"?

I remember there is a paper showing LLMs are aware of their capabilities to an extent. i.e. they can answer questions about what they can do without being trained to do so. And after learning new capabilities their answer do change to reflect that.

I will try to find that paper.

Found it, here: https://martins1612.github.io/selfaware_paper_betley.pdf

The words that are coming out of the model are generated to optimize for RLHF and closeness to the training data, that's it!

This is false, reasoning models are rewarded/punished based on performance at verifiable tasks, not human feedback or next-token prediction.

How does that differ from a non-reasoning model rewarded/punished based on performance at verifiable tasks?

What does CoT add that enables the reward/punishment?

Without CoT then training them to give specific answers reduces performance. With CoT you can punish them if they don't give the exact answer you want without hurting them, since the reasoning tokens help it figure out how to answer questions and what the answer should be.

And you really want to train on specific answers since then it is easy to tell if the AI was right or wrong, so for now hidden CoT is the only working way to train them for accuracy.

Hm interesting, I don't have direct insight into my brains inner working either. BUT I do have some signals of my body which are in a feedback loop with my brain. Like my heartbeat or me getting sweaty.

it would be interesting to perturb the CoT context window in ways that change the sequences but preserve the meaning mid-inference.

so if you deterministically replay an inference session n times on a single question, and each time in the middle you subtly change the context buffer without changing its meaning, does it impact the likelihood or path of getting to the correct solution in a meaningful way?

This type of response is from the typical example of an air chair expert that wildly overestimates their own rationalism and deterministic thinking

[dead]

Yep. They aren't stupid. They aren't smart. They don't do smart. They don't do stupid. They do not think. They don't even "they", if you will. The forms of their input and output are confusing people into thinking these are something they're not, and it's really frustrating to watch.

[EDIT] The forms of their input & output and deliberate hype from "these are so scary! ... Now pay us for one" Altman and others, I should add. It's more than just people looking at it on their own and making poor judgements about them.

I agree, but I also don't understand how they're able to do what they do when it comes to things I can't figure out how they could come up with it.

I was under the impression that CoT works because spitting out more tokens = more context = more compute used to "think." Using CoT as a way for LLMs "show their working" never seemed logical, to me. It's just extra synthetic context.

Humans sometimes draw a diagram to help them think about some problem they are trying to solve. The paper contains nothing that the brain didn't already know. However, it is often an effective technique.

Part of that is to keep the most salient details front and center, and part of it is that the brain isn't fully connected, which allows (in this case) the visual system to use its processing abilities to work on a problem from a different angle than keeping all the information in the conceptual domain.

My understanding of the "purpose" of CoT, is to remove the wild variability yielded by prompt engineering, by "smoothing" out the prompt via the "thinking" output, and using that to give the final answer.

Thus you're more likely to get a standardized answer even if your query was insufficiently/excessively polite.

This is an interesting paper, it postulates that the ability of an LLM to perform tasks correlates mostly to the number of layers it has, and that reasoning creates virtual layers in the context space. https://arxiv.org/abs/2412.02975

That's right. It's not "show the working". It's "do more working".

But the model doesn't have an internal state, it just has the tokens, which means it must encode it's reasoning into the output tokens. So it is a reasonable take to think that CoT was them showing their work.

There’s no specific reason why the reported Chain-of-Thought must accurately reflect the true reasoning process;

Isn't the whole reason for chain-of-thought that the tokens sort of are the reasoning process?

Yes, there is more internal state in the model's hidden layers while it predicts the next token - but that information is gone at the end of that prediction pass. The information that is kept "between one token and the next" is really only the tokens themselves, right? So in that sense, the OP would be wrong.

Of course we don't know what kind of information the model encodes in the specific token choices - I.e. the tokens might not mean to the model what we think they mean.

I'm not sure I understand what you're trying to say here, information between tokens is propagated through self-attention, and there's an attention block inside each transformer block within the model, that's a whole lot of internal state that's stored in (mostly) inscrutable key and value vectors with hundreds of dimensions per attention head, around a few dozen heads per attention block, and around a few dozen blocks per model.

Yes, but all that internal state only survives until the end of the computation chain that predicts the next token - it doesn't survive across the entire sequence as it would in a recurrent network.

There is literally no difference between a model predicting the tokens "<thought> I think the second choice looks best </thought>" and a user putting those tokens into the prompt: The input for the next round would be exactly the same.

So the tokens kind of act like a bottleneck (or more precisely the sampling of exactly one next token at the end of each prediction round does). During prediction of one token, the model can go crazy with hidden state, but not across several tokens. That forces the model to do "long form" reasoning through the tokens and not through hidden state.

The key and value vectors are cached, that's kind of the whole point of autoregressive transformer models, the "state" not only survives within the KV cache but, in some sense, grows continuously with each token added, and is reused for each subsequent token.

Hmm, maybe I misunderstood that part, but so far I thought the KV cache was really just that - a cache. Because all the previous tokens of the sequence stay the same, it makes no sense to compute the same K and V vectors again in each round.

But that doesn't change that the only input to the Q, K and V calculations are the tokens (or in later layers information that was derived from the tokens) and each vector in the cache maps directly to an input token.

So I think you could disable the cache and recompute everything in each round and you'd still get the same result, just a lot slower.

That's absolutely correct, KV cache is just an optimization trick, you could run the model without it, that's how encoder-only transformers do it.

I guess what I'm trying to convey is that the latent representations within a transformer are conditioned on all previous latents through attention, so at least in principle, while the old cache of course does not change, since it grows with new tokens it means that the "state" can be brought up to date by being incorporated in an updated form into subsequent tokens.

Exactly. There's no state outside the context. The difference in performance between the non-reasoning model and the reasoning model comes from the extra tokens in the context. The relationship isn't strictly a logical one, just as it isn't for non-reasoning LLMs, but the process is autoregression and happens in plain sight.

Of course we don't know what kind of information the model encodes in the specific token choices - I.e. the tokens might not mean to the model what we think they mean.

What I think is interesting about this is that for the most part reading the reasoning output is something we can understand. The tokens as produced form english sentences, make intuitive sense. If we think of the reasoning output block as basically just "hidden state" then one could imagine that a there might be a more efficient representation that trades human understanding for just priming the internal state of the model.

In some abstract sense you can already get that by asking the model to operate in different languages. My first experience with reasoning models where you could see the output of the thinking block I think was QwQ which just reasoned in Chinese most of the time, even if the final output was German. Deepseek will sometimes keep reasoning in English even if you ask it German stuff, sometimes it does reason in German. All in all, there might be a more efficient representation of the internal state if one forgoes human readable output.

Of course we don't know what kind of information the model encodes in the specific token choices - I.e. the tokens might not mean to the model what we think they mean.

But it's probably not that mysterious either. Or at least, this test doesn't show it to be so. For example, I doubt that the chain of thought in these examples secretly encodes "I'm going to cheat". It's more that the chain of thought is irrelevant. The model thinks it already knows the correct answer just by looking at the question, so the task shifts to coming up with the best excuse it can think of to reach that answer. But that doesn't say much, one way or the other, about how the model treats the chain of thought when it legitimately is relying on it.

It's like a young human taking a math test where you're told to "show your work". What I remember from high school is that the "work" you're supposed to show has strict formatting requirements, and may require you to use a specific method. Often there are other, easier methods to find the correct answer: for example, visual estimation in a geometry problem, or just using a different algorithm. So in practice you often figure out the answer first and then come up with the justification. As a result, your "work" becomes pretty disconnected from the final answer. If you don't understand the intended method, the "work" might end up being pretty BS while mysteriously still leading to the correct answer.

But that only applies if you know an easier method! If you don't, then the work you show will be, essentially, your actual reasoning process. At most you might neglect to write down auxiliary factors that hint towards or away from a specific answer. If some number seems too large, or too difficult to compute for a test meant to be taken by hand, then you might think you've made a mistake; if an equation turns out to unexpectedly simplify, then you might think you're onto something. You're not supposed to write down that kind of intuition, only concrete algorithmic steps. But the concrete steps are still fundamentally an accurate representation of your thought process.

(Incidentally, if you literally tell a CoT model to solve a math problem, it is allowed to write down those types of auxiliary factors, and probably will. But I'm treating this more as an analogy for CoT in general.)

Also, a model has a harder time hiding its work than a human taking a math test. In a math test you can write down calculations that don't end up being part of the final shown work. A model can't, so any hidden computations are limited to the ones it can do "in its head". Though admittedly those are very different from what a human can do in their head.

Humans also post-rationalize the things their subconscious "gut feeling" came up with.

I have no problem for a system to present a reasonable argument leading to a production/solution, even if that materially was not what happened in the generation process.

I'd go even further and pose that probably requiring the "explanation" to be not just congruent but identical with the production would either lead to incomprehensible justifications or severely limited production systems.

Now, at least in a well disciplined human, we can catch when our gut feeling was wrong when the 'create a reasonable argument' process fails. I guess I wonder how well a LLM can catch that and correct it's thinking.

Now I've seen in some models where it figures out it's wrong, but then gets stuck in a loop. I've not really used the larger reasoning models much to see their behaviors.

yep, this post is full of this post-rationalization, for example. it's pretty breathtaking

I invite anyone who postulates humans are more than just "spicy autocomplete" to examine this thread. The level of actual reasoning/engaging with the article is ... quite something.

Internet commenters don't "reason". They just generate inane arguments over definitions, like a lowly markov bot, without the true spark of life and soul that even certain large language models have.

I recently had fascinating example of that where Sonnet 3.7 had to decide for one option from a set of choices.

In the thinking process it narrowed it down to 2 and finally in the last thinking section it decided for one, saying it's best choice.

However, in the final output (outside of thinking) it then answered with the other option with no clear reason given

This is basically a big dunk on OpenAI, right?

OpenAI made a big show out of hiding their reasoning traces and using them for alignment purposes [0]. Anthropic has demonstrated (via their mech interp research) that this isn't a reliable approach for alignment.

[0] https://openai.com/index/chain-of-thought-monitoring/

I don't think those are actually showing different things. The OpenAI paper is about the LLM planning to itself to hack something; but when they use training to suppress this "hacking" self-talk, it still hacks the reward function almost as much, it just doesn't use such easily-detectable language.

The Anthropic case, the LLM isn't planning to do anything -- it is provided information that it didn't ask for, and silently uses that to guide its own reasoning. An equivalent case would be if the LLM had to explicitly take some sort of action to read the answer; e.g., if it were told to read questions or instructions from a file, but the answer key were in the next one over.

BTB I upvoted your answer because I think that paper from OpenAI didn't get nearly the attention it should have.

Not exactly the same as this study, but I'll ask questions to LLMs with and without subtle hints to see if it changes the answer and it almost always does. For example, paraphrased:

No hint: "I have an otherwise unused variable that I want to use to record things for the debugger, but I find it's often optimized out. How do I prevent this from happening?"

Answer: 1. Mark it as volatile (...)

Hint: "I have an otherwise unused variable that I want to use to record things for the debugger, but I find it's often optimized out. Can I solve this with the volatile keyword or is that a misconception?"

Answer: Using volatile is a common suggestion to prevent optimizations, but it does not guarantee that an unused variable will not be optimized out. Try (...)

This is Claude 3.7 Sonnet.

I mean, this sounds along the lines of human conversations that go like

P1 "Hey, I'm doing A but X is happening"

P2 "Have you tried doing Y?

P1 "Actually, yea I am doing A.Y and X is still occurring"

P2 "Oh, you have the special case where you need to do A.Z"

What happens when you ask your first question with something like "what is the best practice to prevent this from happening"

Oh sorry, these are two separate chats, I wasn't clear. I would agree that if I had asked them in the same chat it would sound pretty normal.

When I ask about best practices it does still give me the volatile keyword. (I don't even think that's wrong, when I threw it in Godbolt with -O3 or -Os I couldn't find a compiler that optimized it away.)

[dead]

Sounds like LLMs short-circuit without necessarily testing their context assumptions.

I also recognize this from whenever I ask it a question in a field I'm semi-comfortable in, I guide the question in a manner which already includes my expected answer. As I probe it, I often find then that it decided to take my implied answer as granted and decide on an explanation to it after the fact.

I think this also explains a common issue with LLMs where people get the answer they're looking for, regardless of whether it's true or there's a CoT in place.

This is such an annoying issue in assisted programming as well.

Say you’re referencing a specification, and you allude to two or three specific values from that specification, you mention needing a comprehensive list and the LLM has been trained on it.

I’ll often find that all popular models will only use the examples I’ve mentioned and will fail to elaborate even a few more.

You might as well read specifications yourself.

It’s a critical feature of these models that could be an easy win. It’s autocomplete! It’s simple. And they fail to do it every single time I’ve tried a similar abstract.

I laugh any time people talk about these models actually replacing people.

They fail at reading prompts at a grade school reading level.

The LLMs copy human written text, so maybe they'll implement Motivated Reasoning just like humans do?

Or maybe it's telling people what they want to hear, just like humans do

They definitely tell people what they want to hear. Even when we'd rather they be correct, they get upvoted or downvoted by users, so this isn't avoidable (but is is fawning or sychophancy?)

I wonder how deep or shallow the mimicry of human output is — enough to be interesting, but definitely not quite like us.

i found with the gemini answer box on google, it's quite easy to get the answer you expect. i find myself just playing with it, asking a question in the positive sense then the negative sense, to get the 2 different "confirmations" from gemini. also it's easily fooled by changing the magnitude of a numerical aspect of a question, like "are thousands of people ..." then "are millions of people ...". and then you have the now infamous black/white people phrasing of a question.

i haven't found perplexity to be so easily nudged.

I have always believed that Chain of Thought basically acts as a form of regularization. LLMs are fundamentally next token predictors without any form or notion of logic, reasoning etc and is as likely (as a probabilistic model) to produce a creative response as something based on facts/principles (or anything that resembles "reasoning").

Asking the LLM to think step by step simply biases it towards the latter. It's still a stochastic parrot but now it sounds logical and that happens to be useful in some cases, regardless of whether we can agree if it's "reasoning".

It feels to me that the hypothesis of this research was somewhat "begging the question". Reasoning models are trained to spit some tokens out that increase the chance of the models spitting the right answer at the end. That is, the training process is singularly optimizing for the right answer, not the reasoning tokens.

Why would you then assume the reasoning tokens will include hints supplied in the prompt "faithfully"? The model may or may not include the hints - depending on whether the model activations believe those hints are necessary to arrive at the answer. In their experiments, they found between 20% and 40% of the time, the models included those hints. Naively, that sounds unsurprising to me.

Even in the second experiment when they trained the model to use hints, the optimization was around the answer, not the tokens. I am not surprised the models did not include the hints because they are not trained to include the hints.

That said, and in spite of me potentially coming across as an unsurprised-by-the-result reader, it is a good experiment because "now we have some experimental results" to lean into.

Kudos to Anthropic for continuing to study these models.

If something convinces you that it's aware then it is. Simulated computation IS computation itself. The territory is the map

There is an abundance of discussion on this thread about whether models are intelligent or not.

This binary is an utter waste of time.

Instead focus on the gradient of intelligence - the set of cognitive skills any given system has and to what degree it has them.

This engineering approach is more likely to lead to practical utility and progress.

The view of intelligence as binary is incredibly corrosive to this field.

The use of highly anthropomorphic language is always problematic- Does a photo resistor controlled nightlight have a chain of thought? Does it reason about its threshold value? Does it have an internal model of what is light, what is dark, and the role it plays in demarcation between the two?

Are the transistors executing the code within the confines even capable of intentionality? If so - where is it derived from?

What would “think” mean? Processed the prompt? Or just accessed the part of the model where the weights are? This is a bit persudo science

I highly suspect that CoT tokens are at least partially working as register tokens. Have these big LLM trainers tried replacing CoT with a similar amount of register tokens and see if the improvements are similar?

I remember there was a paper a little while back which demonstrated that merely training a model to output "........" (or maybe it was spaces?) while thinking provided a similar improvement in reasoning capability to actual CoT.

To me CoT is nothing but lowering learning rate and increasing iterations in a typical ML model. It's basically to force the model to make a small step at a time and try more times to increase accuracy.

Can a model even know that it used a hint? Or would it only say so if it was trained to say what parts of the context it used when asked? Because then it's statistically probable to say so?

It is nonsense to take whatever an LLM writes in its CoT too seriously. I try to classify some messy data, writing "if X edge case appears, then do Y instead of Z". The model in its CoT took notice of X, wrote it should do Y and... it would not do it in the actual output.

The only way to make actual use of LLMs imo is to treat them as what they are, a model that generates text based on some statistical regularities, without any kind of actual understanding or concepts behind that. If that is understood well, one can know how to setup things in order to optimise for desired output (or "alignment"). The way "alignment research" presents models as if they are actually thinking or have intentions of their own (hence the choice of the word "alignment" for this) makes no sense.

One thing I think I’ve found is: reasoning models get more confident and that makes it harder to dislodge a wrong idea.

It feels like I only have 5% of the control, and then it goes into a self-chat where it thinks it’s right and builds on it’s misunderstanding. So 95% of the outcome is driven by rambling, not my input.

Windsurf seems to do a good job of regularly injecting guidance so it sticks to what I’ve said. But I’ve had some extremely annoying interactions with confident-but-wrong “reasoning” models.

Chain of thought does have a minor advantage in the final “fish” example—the explanation blatantly contradicts itself to get to the cheated hint answer. A human reading it should be pretty easily able to tell that something fishy is going on…

But, yeah, it is sort of shocking if anybody was using “chain of thought” as a reflection of some actual thought process going on in the model, right? The “thought,” such as it is, is happening in the big pile of linear algebra, not the prompt or the intermediary prompts.

Err… anyway, like, IBM was working on explainable AI years ago, and that company is a dinosaur. I’m not up on what companies like OpenAI are doing, but surely they aren’t behind IBM in this stuff, right?

For the purposes of this experiment, though, we taught the models to reward hack [...] in this case rewarded the models for choosing the wrong answers that accorded with the hints.

This is concerning because it suggests that, should an AI system find hacks, bugs, or shortcuts in a task, we wouldn’t be able to rely on their Chain-of-Thought to check whether they’re cheating or genuinely completing the task at hand.

As a non-expert in this field, I fail to see why a RL model taking advantage of it's reward is "concerning". My understanding is that the only difference between a good model and a reward-hacking model is if the end behavior aligns with human preference or not.

The articles TL:DR reads to me as "We trained the model to behave badly, and it then behaved badly". I don't know if i'm missing something, or if calling this concerning might be a little bit sensationalist.

40 billion cash to OpenAI while others keep chasing butterflies.

Sad.

Of course they don't.

LLMs are a brainless algorithm that guesses the next word. When you ask them what they think they're also guessing the next word. No reason for it to match, except a trick of context

One interesting quirk with Claude is that it has no idea its Chain-of-Thought is visible to users.

In one chat, it repeatedly accused me of lying about that.

It only conceded after I had it think of a number between one and a million, and successfully 'guessed' it.

Edit: 'wahnfrieden corrected me. I incorrectly posited that CoT was only included in the context window during the reasoning task and later left out entirely. Edited to remove potential misinformation.

No, the CoT is not simply extra context the models are specifically trained to use CoT and that includes treating it as unspoken thought

Huge thank you for correcting me. Do you have any good resources I could look at to learn how the previous CoT is included in the input tokens and treated differently?

I've only read the marketing materials of closed models. So they could be lying, too. But I don't think CoT is something you can do with pre-CoT models via prompting and context manipulation. You can do something that looks a little like CoT, but the model won't have been trained specifically on how to make good use of it and will treat it like Q&A context.

In which case the model couldn't possibly know that the number was correct.

I'm also confused by that, but it could just be the model being agreeable. I've seen multiple examples posted online though where it's fairly clear that the COT output is not included in subsequent turns. I don't believe Anthropic is public about it (could be wrong), but I know that the Qwen team specifically recommend against including COT tokensfrom previous inferences.

Claude has some awareness of its CoT. As an experiment, it's easy, for example, to ask Claude to "think of a city, but only reply with the word 'ready' and next to ask "what is the first letter of the city you thought of?"

Oops! I tried a couple experiments after writing this, and I believe I was mistaken, though I don't know how. It appears Claude was simply playing along, and convinced me it could remember the choices it secretly made. I must either have given it a tell, or perhaps it guessed the same answers twice in a row.

eh interesting..

You don't say. This is my very shocked face.

Meh. People also invent justifications after the fact.

[dead]

... because they don't think.

It's deeply frustrating that these companies keep gaslighting people into believing LLMs can think.

This entire house of cards is built on people believing that the computer is thinking so it's not going away anytime soon.

seemed common-sense obvious to me -- AI (LLMs) don't "reason". great to see it methodically probed and reported in this way.

but i am just a casual observer of all things AI. so i might be too naive in my "common sense".