Of course not. Because the article uses the words 'thought' and 'reasoning' and even 'faithful' to mean something other than their normal meanings, but then expects them to behave exactly the same.
Every field has terms of art, and 'reasoning' is one for LLMs. But that doesn't mean it has the same properties as 'reasoning' in other contexts, because you're not referring to the same thing.
Related: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces![1]
> Our findings consistently challenge the prevailing narrative that intermediate tokens constitute a semantically meaningful reasoning process. First, we observe a pronounced lack of correlation between solution correctness and trace validity—models frequently produce invalid reasoning traces even when they arrive at correct solutions. Second, and more strikingly, models trained on corrupted or semantically irrelevant traces achieve performance comparable to, and often exceeding, that of models trained on correct traces, especially on out-of-distribution tasks.
The author of this paper is in ASU and does a lot of excellent work in this space. People should check it out. His twitter is also active _and_ high SNR.
Over time, I've learned to accept that many people -- even very clever ones -- are incapable of holding a metaphor at arm's length. Once they accept the words of a metaphor as applicable at all, the metaphor collapses entirely into literalism for them. They can no longer see that the metaphor was just a tool with inherent limitatation and boundaries.
Because the field of artificial "intelligence" is constructed around the idea of applying psychological metaphors to computational systems (a very powerful idea!) it's almost a worst case scenario for these people.
Suddenly, they're reversing the metaphors and applying computational schema to psychological processes ("aren't we really just stochastic parrots ourselves?!"); or, like here, they find themselves surprised and confused when they stumble across the natural boundaries of the metaphor experimentally.
It's because they never had sight of the boundaries in the first place and maybe never can quite see them. The words only make sense to them as literal equivalence, and so their surprise when they run into stuff like this is earnest and deep.
This paper made me wonder not whether the chain we can read is actual “thought,” but what conclusions we can draw by observing it. It is an output channel, but not direct access to the black box that creates it. The question pairs present an interesting experiment, but they also got me thinking about semantics: the same underlying relation can have many valid representations, and how models reason across those representations can tell us more about their stability and correctness. Basically, are models semantically consistent when given different representations of the same underlying relation? Do their answers transform as the relationship requires, and do their explanations remain consistent with that relationship?
I see this when just using some chat bot that shows the “reasoning” steps (ad-hoc observation of course, it’s really cool that people are actually studying it).
It is annoying when the bot seems be “reasoning” correctly and then makes an obvious mistake at the end. And perplexing when it seems to be completely wrong and then pull the right answer out of a magic hat at the end.
I guess it makes sense; the “reasoning” steps aren’t actually doing logic, just adding more context to influence the final generation, right? But it is weird to see.
This paper puts words to something I’ve noticed repeatedly with LLMs, particularly Qwen3.6. When I read its reasoning, it appears to recognise the mistake and then carry on as if it hadn’t noticed it at all.
> models often determine their answers based on implicit biases tied to question templates, then construct reasoning chains to justify their predetermined conclusions
> its reasoning was correct right until the final step (Yes/No answer)
I agree, and I very strongly dislike it, to be polite about it. It contributes absolutely nothing and is an excellent way of hand-waving away literally anything an AI model does. Saying "well people do this too" is a great way to rationalize away anything you can imagine that an AI model would be capable of, because "humans do it too so what's the big deal, guys?"
I can just immagine the response to a headline "LLM chooses mass death: thousands killed in horrific AI accident" being something like "lots of humans have caused mass death too."
It's the grounded portion of a feedback loop searching for the 'why is this happening' thinking. I'd imagine most of my own comments in this area boil down to "GIGO" most of the time.
It's a relevant comment in this instance because we're discussing concepts you need to be both trained and practiced in to reason about, and that our discipline has traditionally been blind to. Plenty of people working with LLM context issues who've never been exposed to the idea of 'subtext' or could tell you why it would matter to their direction of effort.
The current political climate is well reasoned and intentional. It might not be yours or mine, however the system is working exactly as the ones paying for it have intended.
My brother and I have been arguing about that all of our lives. He believes everything is intentional and it's just a matter of discovering who benefits. I see chaos that nobody intends or controls. His political landscape is a tapestry of conspiracy theories and mine is a fog of war. I think his is more comforting, since it admits a possibility of a rational, predictable world.
It’s intentional in the sense that actors are acting intentionally for their own benefit (or ant least what their believe is beneficial) and following incentives. Not that there is a master planner who manipulate everything
It's... complicated. Yes, RL reward hacking makes it learn "bird language" and yes, reasoning traces can be misleading. However they also pretty clearly steer the final reply and not simply justify it, and can stay somewhat coherent and relevant with readability SFT and rewards. All these phenomenas coexist, they aren't mutually exclusive. Reasoning traces are still useful for debugging.
Of course not. Because the article uses the words 'thought' and 'reasoning' and even 'faithful' to mean something other than their normal meanings, but then expects them to behave exactly the same.
Every field has terms of art, and 'reasoning' is one for LLMs. But that doesn't mean it has the same properties as 'reasoning' in other contexts, because you're not referring to the same thing.
Why doesn't my asteroid belt buckle?
Related: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces![1]
> Our findings consistently challenge the prevailing narrative that intermediate tokens constitute a semantically meaningful reasoning process. First, we observe a pronounced lack of correlation between solution correctness and trace validity—models frequently produce invalid reasoning traces even when they arrive at correct solutions. Second, and more strikingly, models trained on corrupted or semantically irrelevant traces achieve performance comparable to, and often exceeding, that of models trained on correct traces, especially on out-of-distribution tasks.
[1]: https://arxiv.org/abs/2504.09762
Also posted to HN today with a couple comments: https://news.ycombinator.com/item?id=49360140#49363374
The author of this paper is in ASU and does a lot of excellent work in this space. People should check it out. His twitter is also active _and_ high SNR.
Over time, I've learned to accept that many people -- even very clever ones -- are incapable of holding a metaphor at arm's length. Once they accept the words of a metaphor as applicable at all, the metaphor collapses entirely into literalism for them. They can no longer see that the metaphor was just a tool with inherent limitatation and boundaries.
Because the field of artificial "intelligence" is constructed around the idea of applying psychological metaphors to computational systems (a very powerful idea!) it's almost a worst case scenario for these people.
Suddenly, they're reversing the metaphors and applying computational schema to psychological processes ("aren't we really just stochastic parrots ourselves?!"); or, like here, they find themselves surprised and confused when they stumble across the natural boundaries of the metaphor experimentally.
It's because they never had sight of the boundaries in the first place and maybe never can quite see them. The words only make sense to them as literal equivalence, and so their surprise when they run into stuff like this is earnest and deep.
This paper made me wonder not whether the chain we can read is actual “thought,” but what conclusions we can draw by observing it. It is an output channel, but not direct access to the black box that creates it. The question pairs present an interesting experiment, but they also got me thinking about semantics: the same underlying relation can have many valid representations, and how models reason across those representations can tell us more about their stability and correctness. Basically, are models semantically consistent when given different representations of the same underlying relation? Do their answers transform as the relationship requires, and do their explanations remain consistent with that relationship?
I see this when just using some chat bot that shows the “reasoning” steps (ad-hoc observation of course, it’s really cool that people are actually studying it).
It is annoying when the bot seems be “reasoning” correctly and then makes an obvious mistake at the end. And perplexing when it seems to be completely wrong and then pull the right answer out of a magic hat at the end.
I guess it makes sense; the “reasoning” steps aren’t actually doing logic, just adding more context to influence the final generation, right? But it is weird to see.
This paper puts words to something I’ve noticed repeatedly with LLMs, particularly Qwen3.6. When I read its reasoning, it appears to recognise the mistake and then carry on as if it hadn’t noticed it at all.
> models often determine their answers based on implicit biases tied to question templates, then construct reasoning chains to justify their predetermined conclusions > its reasoning was correct right until the final step (Yes/No answer)
Natural intelligences do this too
Must we always see this restated every time? It's getting a bit stale always seeing these kinds of comments on articles about LLM.
I agree, and I very strongly dislike it, to be polite about it. It contributes absolutely nothing and is an excellent way of hand-waving away literally anything an AI model does. Saying "well people do this too" is a great way to rationalize away anything you can imagine that an AI model would be capable of, because "humans do it too so what's the big deal, guys?"
I can just immagine the response to a headline "LLM chooses mass death: thousands killed in horrific AI accident" being something like "lots of humans have caused mass death too."
It's the grounded portion of a feedback loop searching for the 'why is this happening' thinking. I'd imagine most of my own comments in this area boil down to "GIGO" most of the time.
It's a relevant comment in this instance because we're discussing concepts you need to be both trained and practiced in to reason about, and that our discipline has traditionally been blind to. Plenty of people working with LLM context issues who've never been exposed to the idea of 'subtext' or could tell you why it would matter to their direction of effort.
continuing with the "we need open training data" thread
how much of the training data had thinking traces that dont make sense to people as being actually a description of why the output should be that way?
or rather, its not productive to "we should do better with artificial intelligence"
You think, "this problem" is qn LLM problem?
No they don't, human intelligence has the ability to form an internal model of itself, which allows it to "notice" its own mistakes and change.
Many who watched the last decade knows just because its possible to noticed mistakes and change, its clearly not a reliable process.
The current political climate is well reasoned and intentional. It might not be yours or mine, however the system is working exactly as the ones paying for it have intended.
My brother and I have been arguing about that all of our lives. He believes everything is intentional and it's just a matter of discovering who benefits. I see chaos that nobody intends or controls. His political landscape is a tapestry of conspiracy theories and mine is a fog of war. I think his is more comforting, since it admits a possibility of a rational, predictable world.
It’s intentional in the sense that actors are acting intentionally for their own benefit (or ant least what their believe is beneficial) and following incentives. Not that there is a master planner who manipulate everything
Thank you for contributing absolutely nothing to this discussion.
You might not've noticed but we aren't talking about natural intelligence.
Yeah and it's not great then either
First author here, happy to answer questions
I thought this was already widely known?
From March last year: https://transformer-circuits.pub/2025/attribution-graphs/bio...
There's no reason to believe the model's self-reported "thinking" bears any relation to the mechanics by which it arrived at some output.
It's... complicated. Yes, RL reward hacking makes it learn "bird language" and yes, reasoning traces can be misleading. However they also pretty clearly steer the final reply and not simply justify it, and can stay somewhat coherent and relevant with readability SFT and rewards. All these phenomenas coexist, they aren't mutually exclusive. Reasoning traces are still useful for debugging.
that sounds testable - if you skip the reasoning tokens, do you get the same result?
if not, then there's certainly some bearing, but not necessarily in how we read the tokens as text
"Study: Communing With The Gods of Mount Olympus Via the Oracle at Delphi Is Not Always Faithful"