> cutting ~75% of tokens while keeping full technical accuracy.
I have no clue if this claim holds, but alas, just pretending they did not address the obvious criticism, while they did, is at the very least pretty lazy.
An explanation that explains nothing is not very interesting.
Nobody has to proof anything. It can give your claim credibility. If you don't provide any, an opposing claim without proof does not get any better.
“I don’t need to provide proof to say things” is a valueless, trivial assertion that adds no value whatsoever to any discussion anyone has ever had.
If you want to pretend this is a claim that should be taken seriously, a lack of evidence is damning. If you just want to pass the metaphorical bong and say stupid shit to each other with no judgment and no expectation, then I don’t know what to tell you. Maybe X is better for that.
You can read the skill. They didn't do anything to mitigate the issue, so the criticism is valid.
But they didn't address the criticism. "cutting ~75% of tokens while keeping full technical accuracy" is an empirical claim for which no evidence was provided.
When producing a token the model doesn't just emit the final token but you also have the entire hidden states from previous attention blocks. These hidden states are mixed into the attention block of future tokens (so even though LLMs are autoregressive where a token attends to previous tokens, in terms of a computational graph this means that the hidden states of previous tokens are passed forward and used to compute hidden states of future tokens).
So no it's not wasteful, those low-perplexity tokens are precisely spots that can instead be used to do plan ahead and do useful computation.
Also I would not be sure that even the output tokens are purely "filler". If you look at raw COT, they often have patterns like "but wait!" that are emitted by the model at crucial pivot points. Who's to say that the "you're absolutely right" doesn't serve some other similar purpose of forcing the model into one direction of adjusting its priors.
Do you know that is true? These aren’t just tokens, they’re tokens with specific position encodings preceded by specific context. The position as a whole is a lot richer than you make it out to be. I think this is probably an unanswered empirical question, unless you’ve read otherwise.
The output is "just tokens"; the "position encodings" and "context" are inputs to the LLM function, not outputs. The information that a token can carry is bounded by the entropy of that token. A highly predictable token (given the context) simply can't communicate anything.
Again: if a tiny language model or even a basic markov model would also predict the same token, it's a safe bet it doesn't encode any useful thinking when the big model spits it out.
train an LLM to leave out the filler words, and see it get the same performance at a lower cost? or do it at token selection time?
Or if you prefer, here's a Galilean thought experiment: gin up a script to get a large language model and a tiny language model to predict the next token in parallel; when they disagree, append the token generated by the large model. Clearly the large model will not care that the "easy" tokens were generated by a different model - how could it even know? Same token, same result. And you will find that the tokens that they agree on are, naturally, the filler words.
To be clear, this observation merely debunks the idea that filler words encode useful information, that they give the LLM "room to think". It doesn't directly imply that an LLM that omits filler words can be just as smart, or that such a thing is trivial to make. It could be that highly predictable words are still important to thought in some way. It could be that they're only important because it's difficult to copy the substance of human thought without also capturing the style. But we can be very sure that what they aren't doing is "storing useful intermediate results".
For an LLM, tokens are thought. They have no ability to think, by whatever definition of that word you like, without outputting something. The token only represents a tiny fraction of the internal state changes made when a token is output.
Clearly there is an optimal for each task (not necessarily a global one) and a concrete model for a given task can be arbitrarily far from it. But you'd need to test it out for each case, not just assume that "less tokens = more better". You can be forcing your model to be dumber without realizing it if you're not testing.
But I assume this has been studied? Can anyone point to papers that show it? I’d particularly like to know what the curves look like, it’s clearly not linear, so if you cut out 75% or tokens what do you expect to lose?
I do imagine there is not a lot of caveman speak in the training data so results may be worse because they don’t fit the same patterns that have been reinforcement learned in.
So it must be studied and at least be proven effective in practice to be so universally used now.
Someone else posted a few articles like this in the thread above but there’s probably more and better ones if you search. https://news.ycombinator.com/item?id=47647907
LLMs do stumble into long prediction chains that don’t lead the inference in any useful direction, wasting tokens and compute.
Tokens are how an LLM works things out, but I think it's just as likely as not that LLMs (like people) are capable of overthinking things to the point of coming to a wrong answer when their "gut" response would have been better. I do not content that this is the default mode, but that it is both possible, and that it's more or less likely on one kind of problem than another, problem categories to be determined.
A specific example of this was the era of chat interfaces that leaned too far in the direction of web search when responding to user queries. No, claude, I don't want a recipe blogspam link or summary - just listen to your heart and tell me how to mix pancakes.
More abstractly: LLMs give the running context window a lot of credit, and will work hard to post-hoc rationalize whatever is in there, including any prior low-likelihood tokens. I expect many problematic 'hallucinations' are the result of an unlucky run of two or more low probability tokens running together, and the likelihood of that happening in a given response scales ~linearly with the length of response.
Additionally, LLMs do not actually operate in text; much of the thinking happens in a much higher dimensional space that just happens to be decoded as text.
So unless the LLM was trained otherwise, making it talk like a caveman is more than just theoretically turning it into a caveman.
What do you mean by that? It’s literally text prediction, isn’t it?
I have a list of numbers, 0 to9, and the + , = operators. I will train my model on this dataset, except the model won’t get the list, they will get a bunch of addition problems. A lot. But every addition problem possible inside that space will not be represented, not by a long shot, and neither will every number. but still, the model will be able to solve any math problem you can form with those symbols.
It’s just predicting symbols, but to do so it had to internalize the concepts.
This gives the impression that it is doing something more than pattern matching. I think this kind of communication where some human attribute is used to name some concept in the LLM domain is causing a lot of damage, and ends up inadvertently blowing up the hype for the AI marketing...
So the conclusion was that these middle layers have their own language and it's converting the text into this language and this decoding it. It explains why sometime the models switch to chinese when they have a lot of chinese language inputs, etc.
You are also confusing ‘mechanistic explanation still incomplete’ with ‘empirical phenomenon unestablished.’ Those are not the same thing.
PS. Em dash? So you are some LLM bot trying to bait mine HN for reasoning traces? :D
You sound like you’re trying to sound impressive. Like I said, I’ll read the paper.
you are discovering that the favorite luddite argument is bullshit
https://machinelearning.apple.com/research/illusion-of-think...
> just look at research papers
You didn't add anything other than vibes either.
This is not how the feature called "reasoning" work in current models.
"reasoning" simply let's the model output and then consume some "thinking" tokens before generating the actual output.
All the "fluff" tokens in the output have absolutely nothing to do with "reasoning".
For example thinking in modern US English generates many thoughts, to keep correct speak at right cultural context (there is only one correct way to say People Of Color, and it changes every year, any typo makes it horribly wrong).
Some languages are far more expressive and specialized in logical conditions, conditionals, recursion and reasoning. Like eskimos have 100 words for snow, but for boolean algebra.
It is well proven that thinking in Chinese needs far less tokens!
With this caveman mod you strip out most of cultural complexities of anglosphere, make it easier for foreigners and far simpler to digest.
This is simply not true.
It is very arrogant to assume, no other language can be more advanced than English.
Programming languages are not languages in the human brain nor the culture sense.
There’s a less magical model of how LLMs work: they are essentially fancy autocomplete engines.
Most of us probably have an intuition that the more you give an autocomplete, the better results it will yield. However, does this extend to output of the autocomplete—i.e. the more tokens it uses for the result, the better?
It could well be true in context of chain of thought[0] models, in the sense that the output of a preceding autocomplete step is then fed as input to the next autocomplete step, and therefore would yield better results in the end. In other words, with this intuition, if caveman speak is applied early enough in the chain, it would indeed hamper the quality of the end result; and if it is applied later, it would not really save that many tokens.
Willing to be corrected by someone more familiar with NN architecture, of course.
[0] I can see “thinking” used as a term of art, distinct from its regular meaning, when discussing “chain of thought” models; sort of like what “learning” is in “machine learning”.
As I understand it, the claim is: more tokens = more computation = more "thinking" => answer probably better.
However, another potential issue is that LLMs are continuation engines, and I'd have thought that talking like a caveman may be "interpreted" as meaning you want a dumbed down response, not just a smart response in caveman-speak.
It's a bit like asking an LLM to predict next move in a chess game - it's not going to predict the best move that it can, but rather predict the next move that would be played given what it can infer about the ELO rating of the player whose moves it is continuing. If you ask it to continue the move sequence of a poor player, it'll generate a poor move since that's the best prediction.
Of course there's not going to be a lot of caveman speak on stack overflow, so who knows what the impact is. Program go boom. Me stomp on bugs.
Do LLMs generally perform better in verbose languages than they do in concise ones?
Yeah, definitely. It lacks case and verb conjugations, plus whole classes of filler words, and words themselves are on average substantially shorter. If you listen to or read a hyper-literal transliteration of Chinese speech into English (you can find fun videos of this on Chinese social media), it even resembles "caveman speech" for those reasons.
If you look at translated texts and compare the English versions to the Chinese ones, the Chinese versions are substantially shorter. Same if you compare localization strings in your favorite open-source project.
It's also part of why Chinese apps are so information-dense, and why localizing to other languages often requires reorganizing the layout itself— languages like English just aren't as information-dense, pixel for pixel.
The difference is especially profound for vernacular Chinese, which is why Chinese people often note that text which "has a machine translation flavor" is over-specified and gratuitously prolix.
Maybe some of this washes out in LLMs due to tokenization differences. But Chinese texts are typically shorter than English texts and it extends to prose as well as poetry.
But yeah this is standard stuff: Chinese is more concise and more contextual/ambiguous. More semantic work is allocated in interpretation than with English, less is allocated in the writing/speaking.
Do you speak Chinese and experience the differences between Chinese and English differently? I'm a native English speaker and only a beginner in Chinese but I've formed these views in discussion with Chinese people who know some English as well.
I'm also more curious about tokenizers for LLMs than I've ever been before, both for Chinese and English. I feel like to understand I'll need to look at some concrete examples, since sometimes tokenization can be per word or per character or sometimes chunks that are in between.
It's a significantly much succinct semantic encoding than English while being able to express all the same concepts, since it encodes a lot of glue words into the grammar of the language, and conventionally lets you drop many pronouns.
e.g.
"I would have walked home, but it seemed like it was going to rain" (14 words) -> "Domum ambulavissem, sed pluiturum esse videbatur" (6 words).
Benchmark or nothing.
Not everybody is Dijkstra.
But does talk like caveman make number go down? Less token = less think?
I also wondered, due to the way LLMs work, if I ask AI a question using fancy language, does that make it pattern match to scientific literature, and therefore increase the probability that the output will be true?
https://platform.claude.com/docs/en/build-with-claude/extend...
Nothing on that page indicates otherwise.
Forcing it to be concise doesn't work because it wasn't trained on token strings that short.
This is a 2023-era comment and is incorrect.
> but mmuh latest SOTA from CloudCorp (c)!
You don't know how these things work and all you have to go on is marketing copy.
You also aren't aware that there's more to it than "LLM architecture". And you're rather confident despite your lack of knowledge.
You're like the old LLMs before ChatGPT was released that were kinda neat, but usually wrong and overconfident about it.
https://arxiv.org/abs/2112.00114 https://arxiv.org/abs/2406.06467 https://arxiv.org/abs/2404.15758 https://arxiv.org/abs/2512.12777
First that scratchpads matter, then why they matter, then that they don’t even need to be meaningful tokens, then a conceptual framework for the whole thing.
Did you test that ""caveman mode"" has similar performance to the ""normal"" model?
A lot of communication is just mentioning the concepts.
Funny idea though. And I’d like to see a more matter-of-fact output from Claude.
Take it a step further and do kind of like that xkcd where you try to post and it rewrites it like this and if you want the original version you have to write a justification that gets posted too.
Chef's kiss
Compare with fluid dynamics; it's not hard to write down the Navier–Stokes equations, but there's a million dollars available to the first person who can prove or give a counter-example of the following statement:
In three space dimensions and time, given an initial velocity field, there exists a vector velocity and a scalar pressure field, which are both smooth and globally defined, that solve the Navier–Stokes equations.
Seems reasonable, but this doesn't settle probably-empirical questions like: (a) to what degree is 'more' better?; (b) how important are filler words? (c) how important are words that signal connection, causality, influence, reasoning?
So it's probably true that the "Great question!---" type preambles are not helpful, but that there's definitely a lower bound on exactly how primitive of a caveman language we're pushing toward.
> Someone didn't get the memo that for LLMs, tokens are units of thinking.
e.g. instead of: "The square root of 256 is" you'd enter "errr The er square um root errr of 256 errr is" and it would miraculously get better? The model can't differentiate between words you entered and words it generated its self...