Machine learning based pattern matching is indeed very useful and profitable when applied correctly. Identify (with confidence levels) features in data that would otherwise take an extremely well trained person. And even then it's just for the cursory search that takes the longest before presenting the highest confidence candidate results to a person for evaluation. Think: scanning medical data for indicators of cancer, reading live data from machines to predict failure, etc.

And what we call "AI" right now is just a much much more user friendly version of pattern matching - the primary feature of LLMs is that they natively interact with plain language prompts.

Not "This particular model". Frontier LRMs s OpenAI’s o1/o3,DeepSeek-R, Claude 3.7 Sonnet Thinking, and Gemini Thinking.

The paper shows that Large Reasoning Models as defined today cannot interpret instructions. Their architecture does not allow it.

Sure. We weren't discussing if AI creates value or not. If you ask a different question then you get a different answer.

Is thinking necessarily biologic?

No. They don't. We just call them proteins.

Wow it's almost like the computer scientists were saying this from the start but were shouted over by marketing teams.

The guy selling the car doesn't tell you it runs like a horse, the guy selling you AI is telling you it has reasoning skills. AI absolutely has utility, the guys making it are saying it's utility is nearly limitless because Tesla has demonstrated there's no actual penalty for lying to investors.

Ragebait?

I'm in robotics and find plenty of use for ML methods. Think of image classifiers, how do you want to approach that without oversimplified problem settings?
Or even in control or coordination problems, which can sometimes become NP-hard. Even though not optimal, ML methods are quite solid in learning patterns of highly dimensional NP hard problem settings, often outperforming hand-crafted conventional suboptimal solvers in computation effort vs solution quality analysis, especially outperforming (asymptotically) optimal solvers time-wise, even though not with optimal solutions (but "good enough" nevertheless). (Ok to be fair suboptimal solvers do that as well, but since ML methods can outperform these, I see it as an attractive middle-ground.)

This! Capitalism is going to be the end of us all. OpenAI has gotten away with IP Theft, disinformation regarding AI and maybe even murder of their whistle blower.

If you want to boil down human reasoning to pattern recognition, the sheer amount of stimuli and associations built off of that input absolutely dwarfs anything an LLM will ever be able to handle. It's like comparing PhD reasoning to a dog's reasoning.

While a dog can learn some interesting tricks and the smartest dogs can solve simple novel problems, there are hard limits. They simply lack a strong metacognition and the ability to make simple logical inferences (eg: why they fail at the shell game).

Now we make that chasm even larger by cutting the stimuli to a fixed token limit. An LLM can do some clever tricks within that limit, but it's designed to do exactly those tricks and nothing more. To get anything resembling human ability you would have to design something to match human complexity, and we don't have the tech to make a synthetic human.

those particular models. It does not prove the architecture doesn't allow it at all. It's still possible that this is solvable with a different training technique, and none of those are using the right one. that's what they need to prove wrong.

this proves the issue is widespread, not fundamental.

You are either vastly overestimating the Language part of an LLM or simplifying human physiology back to the Greek's Four Humours theory.

"They".

What are you?

I didn't say we aren't animals or that we don't follow physics rules.

But what you're saying is the equivalent of "everything that goes up will eventually go down - that's how physics works and you don't see that, you're in denial!!!11!!!1"

I mean… “proving” is also just marketing speak. There is no clear definition of reasoning, so there’s also no way to prove or disprove that something/someone reasons.

Hallucinations and the cost of running the models.

So, inaccurate information in books is nothing new. Agreed that the rate of hallucinations needs to decline, a lot, but there has always been a need for a veracity filter - just because it comes from "a book" or "the TV" has never been an indication of absolute truth, even though many people stop there and assume it is. In other words: blind trust is not a new problem.

The cost of running the models is an interesting one - how does it compare with publication on paper to ship globally to store in environmentally controlled libraries which require individuals to physically travel to/from the libraries to access the information? What's the price of the resulting increased ignorance of the general population due to the high cost of information access?

What good is a bunch of knowledge stuck behind a search engine when people don't know how to access it, or access it efficiently?

Granted, search engines already take us 95% (IMO) of the way from paper libraries to what AI is almost succeeding in being today, but ease of access of information has tremendous value - and developing ways to easily access the information available on the internet is a very valuable endeavor.

Personally, I feel more emphasis should be put on establishing the veracity of the information before we go making all the garbage easier to find.

I also worry that "easy access" to automated interpretation services is going to lead to a bunch of information encoded in languages that most people don't know because they're dependent on machines to do the translation for them. As an example: shiny new computer language comes out but software developer is too lazy to learn it, developer uses AI to write code in the new language instead...

Well - if you want to devolve into argument, you can argue all day long about "what is reasoning?"

I agree with you. In its current state, LLM is not sentient, and thus not "Intelligence".

You're correct that the formal definition of a Markov process does not exclude internal computation, and that it only requires the next state to depend solely on the current state. But what defines a classical Markov chain in practice is not just the formal dependency structure but how the transition function is structured and used. A traditional Markov chain has a discrete and enumerable state space with explicit, often simple transition probabilities between those states. LLMs do not operate this way.

The claim that an LLM is "just" a large compressed Markov chain assumes that its function is equivalent to a giant mapping of input sequences to output distributions. But this interpretation fails to account for the fundamental difference in how those distributions are generated. An LLM is not indexing a symbolic structure. It is computing results using recursive transformations across learned embeddings, where those embeddings reflect complex relationships between tokens, concepts, and tasks. That is not reducible to discrete symbolic transitions without losing the model’s generalization capabilities. You could record outputs for every sequence, but the moment you present a sequence that wasn't explicitly in that set, the Markov table breaks. The LLM does not.

Yes, you can say a table is just one implementation of a function, and from a purely mathematical perspective, any function can be implemented as a table given enough space. But the LLM’s function is general-purpose. It extrapolates. A precomputed table cannot do this unless those extrapolations are already baked in, in which case you are no longer talking about a classical Markov system. You are describing a model that encodes relationships far beyond discrete transitions.

The pi analogy applies to deterministic functions with fixed outputs, not to learned probabilistic functions that approximate conditional distributions over language. If you give an LLM a new input, it will return a meaningful distribution even if it has never seen anything like it. That behavior depends on internal structure, not retrieval. Just because a function is deterministic at temperature 0 does not mean it is a transition table. The fact that the same input yields the same output is true for any deterministic function. That does not collapse the distinction between generalization and enumeration.

So while yes, you can implement any deterministic function as a lookup table, the nature of LLMs lies in how they model relationships and extrapolate from partial information. That ability is not captured by any classical Markov model, no matter how large.

And that's pretty damn useful, but obnoxious to have expectations wildly set incorrectly.