This speaks to the general problem with using LLMs for writing. The audience they are writing for us you, but you're trying to write for a totally different audience. In code, this manifests as comments in the code that are hyperspecific to the conversation you are having, and not the long term benefit of having those comments in the code.

I see this in docs a lot. I've been reading a lot of docs these days where it feels like the LLM is trying to hype up the person writing the docs. It's like it has no conception that the writing is meant for a 3rd party audience.

We call these temporal comments. We recently updated our code review skills to heavily discourage them. It doesn’t matter why funcA was added then later refactored to funcB. That much can be ascertained from git history. What does matter is why approach A doesn’t work, but B does.

And will reference transient working docs in code comments.

// No retry was added here per AC 37b in FEATURE.MD.

My theory is it writes these comments as "notes while working", and I don't don't mind that, per-se. My problem is it's inability to clean that shit up before committing it. That's wrong load-bearing lever that doesn't earn it's keep.

Also, it reads like ass.

This!

Claude writes comments about how things used to work, which can be useful sometimes, especially if it's a big change that requires one to genuinely consider legacy behavior, but most of the time it shouldn't be there.

Two other somewhat related things it does:

- It writes as if someone reading the code and comments is aware of everything it is aware of (the current conversation, the code it has just looked at). It's really hard to make it understand that things need to stand on their own. A trick is to get a subagent to look at it with a fresh context, but it doesn't tremendously help

- It does all of this with user-facing strings too. Claude loves to write up tooltips and other labels that leak everything to the end user. Every single concern we have, every edge case we've meticulously made our code handle, it passes on to the user, so they don't "need to worry". But no sane user would think of these things. For them, a feature is a feature. The "dynamic scheduling" button should state what dynamic scheduling does plainly, and every edge case is handled by us. The "add" button does not need a label letting the user know that they will later be able to click the "delete" button, because the user will just realize it due to our adherence to proper design. Claude fails to understand good UX for the user cannot be replaced with endless labels and explanations.

It's an uphill battle and all attempts at solving this (or the brain-dead way new Anthropic models write) usually fail to work with me.

I've explicitly included instructions in my home CLAUDE.md to avoid this, because it's one of the most annoying things about how Claude writes comments.

I've also written my own package for deterministically formatting comments: https://www.npmjs.com/package/comment-fmt

Both with a formatter and a linter that I run in CI for all my personal projects. Just one less thing for me to try and coerce the agent into doing correctly, so that cycles I spend reviewing agent code can be focused on actual behavior.

Yeah i noticed this heavily. Working on a branch i critique and give a simpler design. Claude implements and then in a huge comments reference the outdated worse design. Zero value to that. Even having a "how to write comments" section in claude.md doesn't seem to help much with that.

It has this tendency, but usually a sentence about this in the CLAUDE.md helps avoid it. It's annoying though that you have to add that sentence.

Presumably you're not just blindly copying down what Claude copies out for you, but actually reading, interpreting, and understanding it for yourself?

I struggled with this for a long time, but actually seem to have gotten to a place where this is largely resolved. Copy/paste from my current claude.md:

The CC-5 rule specifically seems to be (just from reading through, nothing repeatable-eval based) the part that actually catches and prevents me from having to clean it up afterwards.

```

### Code comments

The failure this prevents: writing a comment that narrates the change I am making right now. That context is real, but it expires the instant the change merges — the defect it describes no longer exists, so the comment becomes a story about a problem no future reader can observe. It is a changelog entry in the wrong file, and a third copy of text already required in the commit body (3.b) and the PR description.

- *CC-1 (MUST NOT)* Write a comment describing a change, a fix, a defect, its cause, or what the code used to do. No "was/now/previously/instead of", no "this fixes", no "needed because otherwise", no "note that we no longer".

- *CC-2 (MUST)* Apply the survival test to every comment before writing it: would this still be true and useful to someone reading this file a year from now, who never saw the diff? If it only makes sense beside the diff, it is changelog — delete it and put it in the commit body.

- *CC-3 (MUST)* Default to zero comments. Declarative config — Terraform, DNS records, k8s manifests, CI YAML, Helm values — is self-describing and takes none. A resource named `dmarc-example-com` does not need a comment saying it is the DMARC record.

- *CC-4 (MAY)* Comment only when a future editor would actively break something without it: a non-obvious external constraint, a required out-of-band manual step, an invariant the surrounding code cannot show. One line. If it needs a paragraph it belongs in `plans/`, not inline.

- *CC-5 (MUST)* Before every commit, re-read the comment lines I added: `git diff --cached | grep '^+' | grep -E '#|//|/*'`. Each hit must pass CC-2 on its own. Deleting is always an acceptable outcome. "I already wrote it", "it is only one line", and "this one is genuinely useful" are not exemptions — the last one is the exact thought that precedes every violation.

- *CC-6 (MUST)* Applies to comments I edit as well as ones I add. When a change invalidates an existing comment, the default action is DELETE, not rewrite it into a new narrative.

```

Yes, I am aware that claude mostly generated this, and it can probably be better and/or more succinct.

I think people really need to focus more on working with limited contexts rather than trying to work around it. I really try to keep my sessions as short as possible and it helps a ton with keeping Claude (et al) focused.

Specifically, I like the "canary" trick that people have discussed where you add a small, innocuous rule to your CLAUDE.md like "When responding to me, start every sentence with my name." so that when Claude stops doing this, you know you've used way too much context and need to start a new session.

Yea, I've started making it write linters to check the code that goes out. Anything that can be deterministically measured, gets added to it once we lock it down.

> The problem is, when the context window grows,

You know the problem; then why not address it? Does Compacting the context not help?

Hooks is the way.

Intermittent nudges

Unlikely. Much more likely is that Opus 5 was trained in an RL environment with subagents, and it learned to talk this way when reporting progress to the invoking agent.

Evidence or not, this is probably the most satisfying explanation I've heard for this behaviour.

My own suspicion is that LLMs are not becoming more general-purpose over time, as these companies had hoped, and ongoing development in that direction is stalling. They will likely need to move in the direction of more specialization and train LLMs for specific use-cases. However, this would also be admitting that they are not on a direct path to AGI.

In addition to positive reinforcement on the style I wanted for output, I tried your suggestion to tell Claude Fable to think Claudese in its thoughts, and that sadly got the session immediately flagged as too risky for Fable to handle, suggesting I downgrade to Opus. Several tweaks still consistently flagged it until I removed the directive about its thinking.

So I guess Anthropic wants zero mention of thoughts with Fable: they're worried about anything that remotely smells like distillation but only with Fable.

Hey Yuri, slight tangent but how did you get to be an investor on those private companies as per your LinkedIn?

Please share!

What system prompt did you use?

I have no Claude.md file. Claude is still absolutely horrible.

How large was it before and after? What sort of difference did it make?

Gemini has been watermarking for like a year and doesn't have the same problems.

Watermarking doesn't work like that.

Moralizing? Can you expand on that

A terse and compliant robotic servant should've been the persona they chose, because that's probably 99% of what those wanting to use it for work expect.

how do they infuse this personality? do they train the human feedback providers with a certain personality?

> The vast majority of people using Claude don't really care or even notice this one way or another.

You have literally no way to know that.

The comment situation is really bad. I generally don't see a lot of complaints people have about it, but the fact that it references random transient stuff, will delete a line and _leave a comment about the deleted line instead of just deleting the comment_, writes multiple paragraphs for the simplest stuff. It is absurd.

I usually dont comment, but just had to. Opus 5 talls garbage, Im back om 4.8 which is better but still verbose. I use the caveman skill, but after a while it is just ignored.. I switched to Deepseek for a task lately. It was a wow experience (the language part)

Customers of a language model don’t care or notice about the quality of the language produced by the model?

And your reasoning to get to this conclusion? Obscured like Claude’s thinking traces?

Over the last 6 months Claude's written material has gone from mediocre to unacceptable. The specific actual content and insights are somewhat better, but the claudisms are increasingly insufferable.

I've observed a bunch of people not being vocal and complaining, but just flat out switching. For every person you see online complaining, there's bound to be more who are equally annoyed and not being vocal, but just acting. Any many others that aren't bothered at all.

it doesn't

To fix Claude’s personality, first, invent the universe…

This would need the word "genuinely" about 4 times to be Claude comments

i havent used a claude model in a long time, but it seems quite clear to me that the chinese models have trained on claude (at the very least) - they write just the same

Plenty for coding - Sol, K3, new DeepSeek & Qwen, GLM 5.3, Grok 4.6, etc.

I haven't tried with Claudette but did evaluate https://github.com/zachahn/vomit and https://github.com/gvzdv/claudish-to-english earlier today. I ended up using claudish-to-english with a prompt derived from the vomit one and some of my existing instructions. I ran a few local models through a test harness to see how they did, and the gemma4 ones added bad behavior back in less than any others I tested.

So for the parent's Macbook question `gemma4-26b-mlx` should work well.

For you with 24 GB VRAM, `gemma4-26b-a4b`. I tried higher VRAM models and they slowed down while still doing just as well or slightly worse.

If someone else tests and finds a better performing model though, please update me here, I'd love to try it.

Where?

Jesus christ, the lack of self awareness is all so depressing. These Sisyphuses have doomed themselves to making NDFSMs deterministic.

Yep, they're putting lipstick on the pig.

https://x.com/_can1357/status/2090360068529111530

Should be pretty difficult to ignore

Claude will end the conversation if you berate it when it screws something up.