[Subscribe to the newsletter](https://atomic14.substack.com)
[SUPPORT](/support/)MY WORK: If you're feeling flush then please stop by
[Patreon](https://www.patreon.com/atomic14)Or you can make a one off donation via
[ko-fi](https://ko-fi.com/atomic14)
Here’s a response Claude came up with recently. I was asking it to review some of the code it had been writing for my “Open Vibe Code” Elite clone - you can play it here in your browser, it’s surprisingly good.
“Twenty-seven echoes; most are two halves of a seam stated from each side, which is correct. Four are true duplicates. Checking two of them:”
There’s a lot of this kind of output coming out of Claude. The internet is full of people complaining. Sometimes reading what Claude says is very much like being stuck at the bar with the world’s most boring know-it-all. Lots of clever phrases and convoluted sentences - but ultimately vacuous and empty.
What’s really interesting, though, is that the code and docs it generates are pretty good (at least the tests are green, the game plays well etc…). It’s just the communication with me that’s the problem.
A popular recommended remedy for this is to change the output style:
Some people suggest using ASD-STE100 - Simplified Technical English. The theory is that this should force Claude into communicating with you in a sensible way.
There’s a bit of controversy over whether you change the output style or you just add it to CLAUDE.md
, but whichever you choose, the result is supposed to be substantially better.
Does it work?
At first, absolutely. It feels better - easier to read, shorter sentences, fewer random, weird analogies.
But I think that feeling of it being better is an illusion. You think it’s better because it’s different. It’s just a relief to your brain not having to parse Claude’s madness. But after a while you get used to the change and slowly start to realise that the prose is just as vacuous and empty as before.
And the real kicker… the quote at the top of the page was produced after I’d changed the output style and added it to my CLAUDE.md
…
The big question is: Does it matter? And I suspect, probably not.
I’m starting to think of Claude’s ramblings in the same way as the “Thinking…” content - it’s not necessarily meant for me to read. The actual output that matters is the working code. And to get quality there you need programmatic constraints and constant adversarial review.
It’s time to make peace with Claude’s nonsense. Stop reading the prose and judge things purely by the produced artefacts - once you do that, the quality of the commentary stops mattering.
So judge the artefact rather than the commentary. This is what all that vacuous prose actually built:
Harmless is playable in your browser - no install, nothing to sign up to, just open it and fly - and the source is on GitHub if you want to see how it’s put together. Let me know how you get on in the comments.