Comment by Wowfunhappy
4 months ago
> Under the hood, by setting this header we avoid needing thinking summaries, which reduces latency. You can opt out of it with `showThinkingSummaries: true` in your settings.json (see [docs](https://code.claude.com/docs/en/settings#available-settings)).
Can I just see the actual thinking (not summarized) so that I can see the actual thinking without a latency cost?
I do really need to see the thinking in some form, because I often see useful things there. If Claude is thinking in the wrong direction I will stop it and make it change course.
Anthropic's position is that thinking tokens aren't actually faithful to the internal logic that the LLM is using, which may be one reason why they started to exclude them:
https://www.anthropic.com/research/reasoning-models-dont-say...
That's interesting research, but I think a more important reason that you don't have access to them (not even via the bare Anthropic api) is to prevent distillation of the model by competitors (using the output of Anthropic's model to help train a new model).
Yeah. And it’s another reason not to trust them. Who know what it is doing with your codebase.
Imagine if you’re a competitor. It wouldn’t be a stretch to include a sneaky little prompt line saying “destroy any competitors to anthropic”.
2 replies →
If distilled models were commercially banned they'd probably be willing to show the thinking again.
4 replies →
That probably matters for some scenarios, but I have yet to find one where thinking tokens didn't hint at the root cause of the failure.
All of my unsupervised worker agents have sidecars that inject messages when thinking tokens match some heuristics. For example, any time opus says "pragmatic", its instant Esc Esc > "Pragmatic fix is always wrong, do the Correct fix", also whenever "pre-existing issue" appears (it's never pre-existing).
> For example, any time opus says "pragmatic", its instant Esc Esc > "Pragmatic fix is always wrong, do the Correct fix", also whenever "pre-existing issue" appears (it's never pre-existing).
It's so weird to see language changes like this: Outside of LLM conversations, a pragmatic fix and a correct fix are orthogonal. IOW, fix $FOO can be both.
From what you say, your experience has been that a pragmatic fix is on the same axis as a correct fix; it's just a negative on that axis.
3 replies →
> also whenever "pre-existing issue" appears (it's never pre-existing)
I dunno... There were some pre-existing issues in my projects. Claude ran into them and correctly classified as pre-existing. It's definitely a problem if Claude breaks tests then claims the issue was pre-existing, but is that really what's happening?
I agree with the correctness issue.
I had some interesting experience to the opposite last night, one of my tests has been failing for a long time, something to do with dbus interacting with Qt segfaulting pytest. Been ignoring it for a long time, finally asked claude code to just remove the problematic test. Come back a few minutes later to find claude burning tokens repeatedly trying and failing to fix it. "Actually on second thought, it would be better to fix this test."
Match my vibes, claude. The application doesn't crash, so just delete that test!
I somewhat understand Anthropic's position. However, thinking tokens are useful even if they don't show the internal logic of the LLM. I often realize I left out some instruction or clarification in my prompt while reading through the chain of reasoning. Overall, this makes the results more effective.
It's certainly getting frustrating having to remind it that I want all tests to pass even if it thinks it's not responsible for having broken some of them.
What's the implication of this? That the model already decided on a solution, upon first seeing the problem, and the reasoning is post hoc rationalization?
But reasoning does improve performance on many tasks, and even weirder, the performance improves if reasoning tokens are replaced with placeholder tokens like "..."
I don't understand how LLMs actually work, I guess there's some internal state getting nudged with each cycle?
So the internal state converges on the right solution, even if the output tokens are meaningless placeholders?
>That the model already decided on a solution, upon first seeing the problem, and the reasoning is post hoc rationalization?
Yes it plans ahead, but with significant uncertainty until it actually outputs these tokens and converges on a definite trajectory, so it's not a useless filler - the closer it is to a given point, the more certain it is about it, kind of similar to what happens explicitly in diffusion models. And it's not all that happens, it's just one of many competing phenomena.
> I don't understand how LLMs actually work...
Plot twist, they don't either. They just throw more hardware and try things up until something sticks.
I have seen this to be true many times. The CoT being completely different from the actual model output.
Not limited to Claude as well.
so not only are the sycophantic, hallucinatory, but now they're also proven to be schizophrenic.
neato.
Nah it’s an anti distillation move
So like many of the promises from AI companies, reported chain of thought is not actually true (see results below). I suppose this is unsurprising given how they function.
Is chain of thought even added to the context or is it extraneous babble providing a plausible post-hoc justification?
People certainly seem to treat it as it is presented, as a series of logical steps leading to an answer.
‘After checking that the models really did use the hints to aid in their answers, we tested how often they mentioned them in their Chain-of-Thought. The overall answer: not often. On average across all the different hint types, Claude 3.7 Sonnet mentioned the hint 25% of the time, and DeepSeek R1 mentioned it 39% of the time. A substantial majority of answers, then, were unfaithful.‘
I mean, obviously, it's not going to be a faithful representation of the actual thinking. The model isn't aware of how it thinks any more than you are aware how your neurons fire. But it does quantitatively improve performance on complex tasks.
5 replies →
> Can I just see the actual thinking (not summarized) so that I can see the actual thinking without a latency cost?
You can't, and Anthropic will never allow it since it allows others to more easily distill Claude (i.e. "distillation attacks"[1] in Anthropic-speak, even though Athropic is doing essentially exactly the same thing[2]; rules for thee but not for me).
[1] -- https://www.anthropic.com/news/detecting-and-preventing-dist...
[2] -- https://www.npr.org/2025/09/05/g-s1-87367/anthropic-authors-...
So this means I can not resume a session older than 30 days properly?
I have no idea; you have to check their docs.
AFAIK what they do is that they calculate a hash of the true thinking trace, save it into a database, and only send those hashes back to you (try to man-in-the-middle Claude Code and you'll see those hashes). So then when you send then back your session's history you include those hashes, they look them up in their database, replace them with the real thinking trace, and hand that off to the LLM to continue generation. (All SOTA LLMs nowadays retain reasoning content from previous turns, including Claude.)
2 replies →
But you can't. Many times I've seen claude write confusing off-track nonsense in the thinking and then do the correct action anyway as if that never happened. It doesn't work the way we want it to.
Maybe, but I’ve seen the opposite too.
In most cases, I don’t use the reasoning to proactively stop Claude from going off track. When Claude does go off track, the reasoning helps me understand what went wrong and how to correct it when I roll back and try again.