I think the problem is we train models to pattern match, not to learn or reason about world models

I think this is clearly a case of over fitting and failure to generalize, which are really well understood concepts. We don't have to philosophize about what pattern matching really means.

I kinda think I'm more or less the same...OK maybe we have different definitions of "pattern matching".

It's Plato's cave:

We train the models on what are basically shadows, and they learn how to pattern match the shadows.

But the shadows are only depictions of the real world, and the LLMs never learn about that.

100%

But the same is true for human, we get our information though our senses we do not have the __real__ word directly.

We do much more than LLMs have. We have bodies and feelings.

In other words, they learn the game, not how to play games.

They memorize the answers not the process to arrive at answers

This has been disproven so many times... They clearly do both. You can trivially prove this yourself.

> You can trivially prove this yourself.

Given the long list of dead philosophers of mind, if you have a trivial proof, would you mind providing a link?

It’s really easy: go to Claude and ask it a novel question. It will generally reason its way to a perfectly good answer even if there is no direct example of it in the training data.

How do you know it’s a novel question?

You have probably seen examples of LLMs doing the "mirror test", i.e. identifying themselves in screenshots and referring to the screenshot from the first person. That is a genuinely novel question as an "LLM mirror test" wasn't a concept that existed before about a year ago.

Elephant mirror tests existed, so it doesn’t seem all that novel when the word “elephant” could just be substituted for the word “LLM”?

The question isn't about universal novelty, but whether the prompt/context is novel enough such that the LLM answering competently demonstrates understanding. The claim of parroting is that the dataset contains a near exact duplicate of any prompt and so the LLM demonstrating what appears to be competence is really just memorization. But if an LLM can generalize from an elephant mirror test to an LLM mirror test in an entirely new context (showing pictures and being asked to describe it), that demonstrates sufficient generalization to "understand" the concept of a mirror test.

How do you know it’s the one generalizing?

Likely there has been at least one text that already does that for say dolphin mirror tests or chimpanzee mirror teats.

It's not exactly difficult to come up with a question that's so unusual the chance of it being in the training set is effectively zero.

Can you provide some examples of these genuinely unique questions?

I'm not sure what you mean by "genuinely." But in the coding context LLMs answer novel questions all the time. My codebase uses components and follows patterns that an LLM will have seen before, but the actual codebase is unique. Yet, the LLM can provide detailed explanations about how it works, what bugs or vulnerabilities it might have, modify it, or add features to it.

It must not have existed prior in any text database whatsoever.

It certainly wasn't. The codebase is thousands of lines of bespoke code that I just wrote.

Which pretty much every line in it was written similarly somewhere else before, including an explanation and is somehow included in the massive data set it was trained on.

So far i have asked the AI some novel questions and it came up with novel answers full of hallucinated nonsense, since it copied some similarly named setting or library function and replaced a part of it's name with something i was looking for.

And this training data somehow includes an explanation of how these individual lines (with variable names unique to my application) work together in my unique combination to produce a very specific result? I don't buy it.

And...

> pretty much

Is it "pretty much" or "all"? The claim that the LLM simply has simply memorized all of its responses seems to require "all."

And as any programmer will tell you: they immediately devolve into "hallucinating" answers, not trying to actually reason about the world. Because that's what they do: they create statistically plausible answers even if those answers are complete nonsense.

When LLM's come up with answers to questions that aren't directly exampled in the training data, that's not proof at all that it reasoned its way there — it can very much still be pattern matching without insight from the actual code execution of the answer generation.

If we were taking a walk and you asked me for an explanation for a mathematical concept I have not actually studied, I am fully capable of hazarding a casual guess based on the other topics I have studied within seconds. This is the default approach of an LLM, except with much greater breadth and recall of studied topics than I, as a human, have.

This would be very different than if we sat down at a library and I applied the various concepts and theorems I already knew to make inferences, built upon them, and then derived an understanding based on reasoning of the steps I took (often after backtracking from several reasoning dead ends) before providing the explanation.

If you ask an LLM to explain their reasoning, it's unclear whether it just guessed the explanation and reasoning too, or if that was actually the set of steps it took to get to the first answer they gave you. This is why LLMs are able to correct themselves after claiming strawberry has 2 rs, but when providing (guessing again) their explanations they make more "relevant" guesses.

LLMs clearly don't reason in the same way that humans or SMT solvers do. That doesn't mean they aren't reasoning.

I'm not sure what "just guessed" means here. My experience with LLMs is that their "guesses" are far more reliable than a human's casual guess. And, as you say, they can provide cogent "explanations" of their "reasoning." Again, you say they might be "just guessing" at the explanation, what does that really mean if the explanation is cogent and seems to provide at least a plausible explanation for the behavior? (By the way, I'm sure you know that plenty of people think that human explanations for their behavior are also mere narrative reconstructions.)

I don't have a strong view about whether LLMS are really reasoning -- whatever that might mean. But the point I was responding to is that LLMS have simply memorized all the answers. That is clearly not true under any normal meanings of those words.

Just go and ask ChatGPT or Claude something that can't possibly be in its training set. Make something up. If it is only memorising answers then it will be impossible for it to get the correct result.

A simple nonsense programming task would suffice. For example "write a Python function to erase every character from a string unless either of its adjacent characters are also adjacent to it in the alphabet. The string only contains lowercase a-z"

That task isn't anywhere in its training set so they can't memorise the answer. But I bet ChatGPT and Claude can still do it.

Honestly this is sooooo obvious to anyone that has used these tools, it's really insane that people are still parroting (heh) the "it just memorises" line.

LLMs don't "memorize" concepts like humans do. They generate output based on token patterns in their training data. So instead of having to be trained on every possible problem, they can still generate output that solves it by referencing the most probable combination of tokens for the specified input tokens. To humans this seems like they're truly solving novel problems, but it's merely a trick of statistics. These tools can reference and generate patterns that no human ever could. This is what makes them useful and powerful, but I would argue not intelligent.

> To humans this seems like they're truly solving novel problems

Because they are. This is some crazy semantic denial. I should stop engaging with this nonsense.

We have AI that is kind of close to passing the Turing test and people still say it's not intelligent...

Depending on the interviewer, you could make a non-AI program pass the Turing test. It's quite a meaningless exercise.

Obviously I mean for a sophisticated interviewer. Not nonsense like the Loebner prize.

The Turing test is contrived to chatting via textual interface.

These machines are only able to output text.

It seems hard to think they could reasonably think any -normal- person.

Tech only feels like magic if you don't know how it works

> Because they _are_.

Not really. Most of those seemingly novel problems are permutations of existing ones, like the one you mentioned. A solution is simply a specific permutation of tokens in the training data which humans are not able to see.

This doesn't mean that the permutation is something that previously didn't exist, let alone that it's something that is actually correct, but those scenarios are much rarer.

None of this is to say that these tools can't be useful, but thinking that this is intelligence is delusional.

> We have AI that is kind of close to passing the Turing test and people still say it's not intelligent...

The Turing test was passed arguably decades ago. It's not a test of intelligence. It's an _imitation game_ where the only goal is to fool humans into thinking they're having a text conversation with another human. LLMs can do this very well.

People who say that LLMs memorize stuff are just as clueless who assume that there's any reasoning happening.

They generate statistically plausible answers (to simplify the answer) based on the training set and weights they have.

What if that’s all we’re doing, though?

Most of us definitely do :)

Or we do it most of the time :)

They learn the value of specific actions in specific contexts based on the rewards they received during their play time. Specific actions and specific contexts are not transferable for various reasons. John quoted that varying frame rates and variable latency between action and effect really confuse the models.

Okay, so fuzz the frame rate and latency? That feels very easy to fix.

Good point, you should write to John Carmack and let him know you've figured out the problem.

Well yeah... If you only ever played one game in your life you would probably be pretty shit at other games too. This does not seem very revealing to me.

I am decent at chess but barely know how the pieces in Go move.

Of course, this because I have spent a lot of time TRAINING to play chess and basically none training to play go.

I am good on guitar because I started training young but can't play the flute or piano to save my life.

Most complicated skills have basically no transfer or carry over other than knowing how to train on a new skill.

But the point here is, if i gave you a guitar with a string more or less. Or a different shaped guitar, you could play it.

If i give you a chess set with dwarf themed pieces and different colored squares, you could play immediately.

I don't think thats true. If you'd only ever played Doom, I think you could play, say, counterstrike or half-life and be pretty good at it, and i think Carmack is right that its pretty interesting that this doesn't seem to be the case for ai models

yeahhhh why isnt there a training structure where you play 5000 games, and the reward function is based on doing well in all of them?

I guess its a totaly different level of control: instead of immediately choosing a certain button to press, you need to set longer term goals. "press whatever sequence over this time i need to do to end up closer to this result"

There is some kind of nested multidimensional thing to train on here instead of immediate limited choices

Where do you draw the line between pattern matching and reasoning about world models?

A lot of intelligence is just pattern matching and being quick about it.

The line is: building an internal world model requires interfacing with the world, not a model of it, and subsequent failing (including death and survivorship over generations) and adaptation. Plus pattern matching.

Current AI only does one of those (pattern matching, not evolution), and the prospects of simulating evolution is kind of bleak, given I don’t think we can simulate a full living cell yet from scratch? Building a world model requires life (or something that has undergone a similar evolutionary survivorship path), not something that mimics life.

You don't need to simulate a full living cell to have evolution. In fact, isn't using evolving programs a decades-old technic ?

Genetic programming models a natural process of evolution to do something useful, the same way machine learning models neurons to do something useful.

But producing something useful is a totally different thing from producing resilience in physical reality. That takes a world model, and I guess my suspicion is that an entity can’t build a world model without a long history of surviving in that world.

Put another way, you can never replicate what it’s like to burn your hand on the fire using only words. You could have a million people tell a child about what fire is like, the dangers of it, the power of it, the pain of it. But they will never develop an innate understanding of it that helps them navigate the real world.

Until they stick their hand in the fire. Then they know.

In the Physics of Language Models[1] they argue that you must augment your training data by changing sentences and such, in order for the model to be able to learn the knowledge. As I understand their argument, language models don't have a built-in way to detect what is important information and what is not, unlike us. Thus the training data must aid it by presenting important information in many different ways.

Doesn't seem unreasonable that the same holds in a gaming setting, that one should train on many variations of each level. Change the lengths of halls connecting rooms, change the appearance of each room, change power-up locations etc, and maybe even remove passages connecting rooms.

[1]: https://physics.allen-zhu.com/part-3-knowledge/part-3-1

According to Carmack's recent talk [0], SOTA models that have been trained on game A don't perform better or train faster on game B. Even worse, training on game B negatively affects performance in game A when returning to it.

[0] https://www.youtube.com/watch?v=3pdlTMdo7pY

You can see a similar effect with LLM finetunes. If you finetune a base model (or other instruct/finetune model) for a new task (e.g. better maths or programming language comprehension) it performs worse at other tasks like creative writing.

To mitigate this you have to include the other categories in your finetune training dataset so it doesn't lose the existing knowledge. Otherwise, the backpropagation and training will favour weights that reflect the new data.

In the game example having the weights optimized for game A doesn't help with game B. It would be interesting to see if training for both game A and B help it understand concepts in both.

Similarly with programming languages it would be interesting to see if training it with multiple languages if it can extract concepts like if statements and while loops.

IIUC from the observations with multilingual LLMs you need to have the different things you are supporting in the training set together. Then the current approach is able to identify similar concepts/patterns. It's not really learning these concepts but is learning that certain words often go together or that a word in one language is similar to another.

It would be interesting to study multilingual LLMs for their understanding of those languages in the case where the two languages are similar (e.g. Scottish and Irish Gaelic; Dutch and Afrikaans; etc.), are in the same language family (French, Spanish, Portuguese), or are in different language families (Italian, Japanese, Swahili), etc.

> In the game example having the weights optimized for game A doesn't help with game B. It would be interesting to see if training for both game A and B help it understand concepts in both.

Supposedly it does both A and B worse. That's their problem statement essentially. Current SOTA models don't behave like humans would. If you took a human that's really good at A and B, chances are they're gonna pick up C much quicker than a random person off the street that hasn't even seen Atari before. With SOTA models, the random "person" does better at C than the A/B master.