Thinking Token Visibility

Concerns about Anthropic hiding chain-of-thought reasoning, theories about protecting against distillation attacks, debate over whether thinking tokens provide useful debugging information

Anthropic’s decision to hide or summarize chain-of-thought reasoning has sparked a heated debate between users who view visibility as a critical debugging tool and those who suspect the company is merely protecting its intellectual property from "distillation attacks" by competitors. While some skeptics dismiss these thinking tokens as confabulated "neuralese" or noisy gibberish, power users argue that losing access to raw reasoning blocks makes it far more difficult to detect when a model is confused or when a prompt has been underspecified. This shift toward a "black box" experience—exemplified by uninformative placeholder messages and documented "hacks" to retrieve summaries—has left many feeling that the industry is sacrificing transparency and developer control in favor of economic efficiency and IP security.

View on HN · Topics

I'm finding the "adaptive thinking" thing very confusing, especially having written code against the previous thinking budget / thinking effort / etc modes: https://platform.claude.com/docs/en/build-with-claude/adapti...

Also notable: 4.7 now defaults to NOT including a human-readable reasoning token summary in the output, you have to add "display": "summarized" to get that: https://platform.claude.com/docs/en/build-with-claude/adapti...

(Still trying to get a decent pelican out of this one but the new thinking stuff is tripping me up.)

View on HN · Topics

Note that for Claude Code, it looks like they added a new undocumented command line argument `--thinking-display summarized` to control this parameter, and that's the only way to get thinking summaries back there.

VS Code users can write a wrapper script which contains `exec "$@" --thinking-display summarized` and set that as their claudeCode.claudeProcessWrapper in VS Code settings in order to get thinking summaries back.

View on HN · Topics

Here is additional discussion and hacks around trying to retain Thinking output in Claude Code (prior to this release):

https://github.com/anthropics/claude-code/issues/8477

View on HN · Topics

Does this mean Claude no longer outputs the full raw reasoning, only summaries? At one point, exposing the LLM's full CoT was considered a core safety tenet.

View on HN · Topics

Anthropic was chirping about Chinese model companies distilling Claude with the thinking traces, and then the thinking traces started to disappear. Looks like the output product and our understanding has been negatively affected but that pales in comparison with protecting the IP of the model I guess.

View on HN · Topics

I don't think it ever has. For a very long time now, the reasoning of Claude has been summarized by Haiku. You can tell because a lot of the times it fails, saying, "I don't see any thought needing to be summarised."

View on HN · Topics

Anthropic always summarizes the reasoning output to prevent some distillation attacks

View on HN · Topics

Genuine question, why have you chosen to phrase this scraping and distillation as an attack? I'm imagining you're doing it because that's how Anthropic prefers to frame it, but isn't scraping and distillation, with some minor shuffling of semantics, exactly what Anthropic and co did to obtain their own position? And would it be valid to interpret that as an attack as well?

View on HN · Topics

Attacks? That's a choice of words.

View on HN · Topics

Definitely Anthropic playing the victim after distilling the whole internet.

View on HN · Topics

Safety versus Distillation, guess we see what's more important.

View on HN · Topics

CoT is basically bullshit, entirely confabulated and not related to any "thought process"...

View on HN · Topics

Don't look at "thinking" tokens. LLMs sometimes produce thinking tokens that are only vaguely related to the task if at all, then do the correct thing anyways.

View on HN · Topics

Why does this comment appear every time someone complains about CoT becoming more and more inaccessible with Claude?

I have entire processes built on top of summaries of CoT. They provide tremendous value and no, I don't care if "model still did the correct thing". Thinking blocks show me if model is confused, they show me what alternative paths existed.

Besides, "correct thing" has a lot of meanings and decision by the model may be correct relative to the context it's in but completely wrong relative to what I intended.

The proof that thinking tokens are indeed useful is that anthropic tries to hide them. If they were useless, why would they even try all of this?

Starting to feel PsyOp'd here.

View on HN · Topics

Didn't you notice that the stream is not coherent or noisy? Sometimes it goes from thought A to thought B then action C, but A was entirely unnecessary noise that had nothing to do with B and C. I also sometimes had signals in the thinking output that were red flags, or as you said it got confused, but then it didn't matter at all. Now I just never look at the thinking tokens anymore, because I got bamboozled too often.

Perhaps when you summarize it, then you might miss some of these or you're doing things differently otherwise.

View on HN · Topics

I agree. Ever since the release of R1, it's like every single American AI company has realized that they actually do not want to show CoT, and then separately that they cannot actually run CoT models profitably. Ever since then, we've seen everyone implement a very bad dynamic-reasoning system that makes you feel like an ass for even daring to ask the model for more than 12 tokens of thought.

View on HN · Topics

Thinking summaries might not be useful for revealing the model's actual intentions, but I find that they can be helpful in signalling to me when I have left certain things underspecified in the prompt, so that I can stop and clarify.

View on HN · Topics

Thinking helps the models arrive at the correct answer with more consistency. However, they get the reward at the end of a cycle. Turns out, without huge constraints during training thinking, the series of thinking tokens, is gibberish to humans.

I wonder if they decided that the gibberish is better and the thinking is interesting for humans to watch but overall not very useful.

View on HN · Topics

They also sometimes flag stuff in their reasoning and then think themselves out of mentioning it in the response, when it would actually have been a very welcome flag.

View on HN · Topics

Yea I’ve seen this and stopped it and asked it about it.

Sometimes they notice bugs or issues and just completely ignore it.

View on HN · Topics

This can result in some funny interactions. I don't know if Claude will say anything, but I've had some models act "surprised" when I commented on something in their thinking, or even deny saying anything about it until I insisted that I can see their reasoning output.

View on HN · Topics

Supposedly ( https://www.reddit.com/r/ClaudeAI/comments/1seune4/claude_ch... ) they can't even see their own reasoning afterwards.

View on HN · Topics

"Also notable: 4.7 now defaults to NOT including a human-readable reasoning token summary in the output, you have to add "display": "summarized" to get that"

I did not follow all of this, but wasn't there something about, that those reasoning tokens did not represent internal reasoning, but rather a rough approximation that can be rather misleading, what the model actual does?

View on HN · Topics

The reasoning is the secret sauce. They don't output that. But to let you have some feedback about what is going on, they pass this reasoning through another model that generates a human friendly summary (that actively destroys the signal, which could be copied by competition).

View on HN · Topics

Don't or can't.

My assumption is the model no longer actually thinks in tokens, but in internal tensors. This is advantageous because it doesn't have to collapse the decision and can simultaneously propogate many concepts per context position.

View on HN · Topics

I would expect to see a significant wall clock improvement if that was the case - Meta's Coconut paper was ~3x faster than tokenspace chain-of-thought because latents contain a lot more information than individual tokens.

Separately, I think Anthropic are probably the least likely of the big 3 to release a model that uses latent-space reasoning, because it's a clear step down in the ability to audit CoT. There has even been some discussion that they accidentally "exposed" the Mythos CoT to RL [0] - I don't see how you would apply a reward function to latent space reasoning tokens.

[0]: https://www.lesswrong.com/posts/K8FxfK9GmJfiAhgcT/anthropic-...

View on HN · Topics

Ctrl-F "neuralese" on that page.

View on HN · Topics

Are you sure? It would be great to get official/semi-official validation that thinking is or is not resolved to a token embedding value in the context.

View on HN · Topics

Don't. thinking right now is just text. Chain of though, but just regular tokens and text being output by the model.

View on HN · Topics

Most likely, would be cool yes see a open source Nivel use diffusion for thinking.

View on HN · Topics

Although it's more likely they are protecting secret sauce in this case, I'm wondering if there is an alternate explanation that LLMs reason better when NOT trying to reason with natural language output tokens but rather implement reasoning further upstream in the transformer.

View on HN · Topics

While the caveman stuff is obviously not serious, there is a lot of legit research in this area.

Which means yes, you can actually influence this quite a bit. Read the paper “Compressed Chain of Thought” for example, it shows it’s really easy to make significant reductions in reasoning tokens without affecting output quality.

There is not too much research into this (about 5 papers in total), but with that it’s possible to reduce output tokens by about 60%. Given that output is an incredibly significant part of the total costs, this is important.

https://arxiv.org/abs/2412.13171

View on HN · Topics

I don’t think this is a plausible argument, as they’re generally capacity constrained, and everyone would like shorter (= faster) responses.

I’m fairly certain that in a few more releases we’ll have models with shorter CoT chains. Whether they’ll still let us see those is another question, as it seems like Anthropic wants to start hiding their CoT, potentially because it reveals some secret sauce.

View on HN · Topics

Some labs do it internally because RLVR is very token-expensive. But it degrades CoT readability even more than normal RL pressure does.

It isn't free either - by default, models learn to offload some of their internal computation into the "filler" tokens. So reducing raw token count always cuts into reasoning capacity somewhat. Getting closer to "compute optimal" while reducing token use isn't an easy task.

View on HN · Topics

Yeah the readability suffers, but as long as the actual output (ie the non-CoT part) stays unaffected it’s reasonably fine.

I work on a few agentic open source tools and the interesting thing is that once I implemented these things, the overall feedback was a performance improvement rather than performance reduction, as the LLM would spend much less time on generating tokens.

I didn’t implement it fully, just a few basic things like “reduce prose while thinking, don’t repeat your thoughts” etc would already yield massive improvements.

View on HN · Topics

Help me understand: I get that the file reading can be a lot. But I also expand the box to see its “reasoning” and there’s a ton of natural language going on there.

View on HN · Topics

I wonder if you can have it reason in caveman

View on HN · Topics

Doesn't it just use more tokens in reasoning?

View on HN · Topics

I’ve been on the Claude Code train for a while but decided to try Codex last week after they announced the $100 USD Pro plan.

I’ve been pretty happy with it! One thing I immediately like more than Claude is that Codex seems much more transparent about what it’s thinking and what it wants to do next. I find it much easier to interrupt or jump in the middle if things are going to wrong direction.

Claude Code has been slowly turning into this mysterious black box, wiping out terminal context any time it compacts a conversation (which I think is their hacky way of dealing with terminal flickering issues — which is still happening, 14 months later), going out of the way to hide thought output, and then of course the whole performance issues thing.

Excited to try 4.7 out, but man, Codex (as a harness at least) is a stark contrast to Claude Code.

View on HN · Topics

I'm running it for the first time and this is what the thinking looks like. Opus seems highly concerned about whether or not I'm asking it to develop malware.

> This is _, not malware. Continuing the brainstorming process.

> Not malware — standard _ code. Continuing exploration.

> Not malware. Let me check front-end components for _.

> Not malware. Checking validation code and _.

> Not malware.

> Not malware.

View on HN · Topics

This is a CC harness thing than a model thing but the "new" thinking messages ('hmm...', 'this one needs a moment...') are extraordinarily irritating. They're both entirely uninformative and strictly worse than a spinner. On my workflows CC often spends up to an hour thinking (which is fine if the result is good) and seeing these messages does not build confidence.

View on HN · Topics

Sounds really minor, but was actually a big contributor to me canceling and switching. The VS Code extension has a morphing spinner thing that rapidly switches between these little catch phrases. It drives me crazy, and I end up covering it up with my right click menu so I can read the actual thinking tokens without that attention vampire distracting me.

And of course they recently turned off all third party harness support for the subscription, so you're just forced to watch it and any other stuff they randomly decide to add, or pay thousands of dollars.

View on HN · Topics

There’s one that’s like “Considering 17 theories” that had me wondering what those 17 things would be, I wanted to see them! Turns out it’s just a static message. Very confusing.

View on HN · Topics

Agreed. I actually have thought those were “waiting to get a response from the API” rather than “the model is still thinking” messages

View on HN · Topics

I'd recommend anyone to ask Claude to show used context and thinking effort on its status line, something like:

```
#!/bin/bash
input=$(cat)
DIR=$(echo "$input" | jq -r '.workspace.current_dir // empty')
PCT=$(echo "$input" | jq -r '.context_window.used_percentage // 0' | cut -d. -f1)
EFFORT=$(jq -r '.effortLevel // "default"' ~/.claude/settings.json 2>/dev/null)
echo "${DIR/#$HOME/~} | ${PCT}% | ${EFFORT}"
```

Because the TUI it is not consistent when showing this and sometimes they ship updates that change the default.

View on HN · Topics

They're now hiding thinking traces. Wtf Anthropic.

View on HN · Topics

They are still available. Just in OpenAI instead.

Summarizer