This is another "AI-ism" I noticed, mostly in coding agents - they seem to be very fond of making up new "compound nouns" (and occasionally verbs) to sum up relatively complex and specific concepts into single noun phrases. I wasn't sure if it's to save tokens or if the AI uses this to get a concise "identifier" for a concept that it can refer back to later, but I found it very noticeable.
I find the resulting sentences hard to read, though it does get better if you're aware of that tendency and make a conscious effort to parse the noun phrases. But I guess since it's just intermediate output from coding agents and not text for essays or blog posts, it's fine.
It's a thing in some Germanic languages. Instinct is to merge nouns into word, e.g. 'lawnchair', but that gives you a red squiggly line, but 'lawn chair' also looks wrong, so 'lawn-chair' is the middle ground.
> English word origins are a fascinating rabbit hole.
My favourite example of which is Northern Ireland's Orange Order.
The colour is orange, because that was the Royal colour of the family of the monarch it was named after, William of Orange, who was Dutch, titled after the principality of Orange which is named after the city of Orange which is French which got its name from the Celtic word for forehead or temple.
The colour is named after the fruit, the fruit's name is a corruption "a norange" -> "an orange", which goes back to naranja which goes to Arabic which goes to Classical Persian which goes to Sanskrit.
Meanwhile, the dutch word for the fruit is sinaasappel, Chinese apple, compare with the English word "mandarin" used for many different Chinese things.
Yes! It's infuriating. I've tried prohibiting them in my AGENTS.md but it's not 100% effective.
--- AGENTS.md ---
## Plain words, not jargon
Don't use jargon-as-shorthand. Say what you actually mean.
- Don't say "load-bearing assumptions". Say "the assumptions the xyz depends on".
- Don't say "cross-service". Name both services, e.g. "whether the X
service can derive duration without calling the Y service". "Cross-X" is
confusing because it hides which things are involved.
- Don't deliver verdicts as abstract noun-phrases like "Cross-RCA
double-counting is unfounded". Say it plainly: "I checked whether the same
root cause gets counted twice across RCA runs, and it doesn't."
## No earth-shattering declarations
Don't hype findings. Skip "a critical finding changes everything", "now I have
the full picture", "this changes the game", etc. Just state what you found
plainly. Most findings are ordinary; report them that way.
## Don't reflexively hedge a "yes"
When the answer is yes, say yes. Don't soften every positive answer with a
caveat: it erodes confidence in the "yes". Only add a caveat when there's a
genuine, specific uncertainty worth flagging.
Yeah, I wonder if part of the reasoning is built around those phrases, and therefore it can't get rid of them easily.
> "now I have the full picture"
I always interpreted that phrase as a sort of marker to delimit the phase in which it explores the codebase and gathers information from the phase in which it implements the changes.
Not sure if it's still done, but I think some months ago there was discussion that some of the phrases are injected by the inference loop to "steer" the model - e.g. "But wait" if a thought block was too short etc. Obviously such phrases couldn't be influenced by the prompt.
The human (or other entity) judging it probably thinks it looks like a sign of thinking or reasoning, and doing a better job -- catching its own errors before they are confidently surfaced. That would be worth rewarding.
On another note, I find AI instructions like this (e.g. "Don't hype findings. Skip "a critical finding changes everything",...") more harm than good in my own uses. It changes behavior in subtle ways that makes it less predictable to me. I'd rather it has its own AI-isms and quirks, that I've fully gotten used to, and I know what to expect. I know when it says certain things, in certain ways, that's what I think it means. Quirks and AI-isms don't annoy me, I get used to how it states things.
Lol! Good point. I did use Claude to write the rule, and it ironically wrote the exact thing I asked it to avoid. I agree that it might be best to use the model as-is, to get the intended experience.
Also I find it interesting to learn the jargon. It basically compacts information in fewer words, although more complex words. But when you are familiar with the jargon, you can unpack the sentences in your mind. And like that, you need less text to read and write prompts. So less reading, writing, and tokens!
I thought it was just Opus 4.7 and 4.8 that did this. Do other models do this too?
Anyway: in my case Opus absolutely did not follow a similar instruction in the CLAUDE.md file. (But then again: it hardly followed _any_ CLAUDE.md instruction properly)
It's stupid, but have you tried telling it to follow it? "Make sure to follow the guidelines from AGENTS/CLAUDE.md" etc, seems to (sadly) make some difference in most harnesses and models.
For me, Opus 4.8's thinking traces for the chatbot will sometimes willingly ignore instructions, saying something along the lines of "I've noticed an instruction in the system instructions that states I shouldn't do this, but if I don't do this, I'll not provide the answer the user is looking for. I will ignore that instruction."
In all my CLAUDE.md and AGENTS.md files I have a line to fix pre-existing issues. I don’t know what it is but every agent I’ve tried through Claude code (including deepseek and GLM) will actively try to avoid fixing pre-existing issues. I even added hooks to Claude and git to try to get them fixed. If I leave a bailout for myself agents will find it sit and ask if it can push with no-verify or an environment variable in the case of Claude hooks instead of trying to fix an issue it didn’t cause.
"The model refuses to follow my specific word detail prompts" and "The model refuses to perform hacking attempts" are on the same side of the model refusing to do something baked into it though.
I’d recommend to instead write a de-slop skill that instructs to launch a sub agent with fresh context, and analyze for such phrases in the new commits, and remove those. Find -> fix just works better than preemptive instructions, in my experience.
And if you manage to do this automatically before committing, you’ve built the backpressure everybody is talking about.
I think you’re probably overthinking it. The same model will in my experience find errors in the same thing it just generated. I think reviewing is a totally different place in latent space, than implementing. Anyhow, it works well for me.
I often tell codex to launch a subagent without prior context to „remove BS phrases and make the prose sound more natural and higher readability“. That‘s usually enough to get better results.
I’m (sorry for the lack of humbleness) a very fluent non-native speaker and writer, and this is by far my biggest challenge with Claude. It stitches together 2-4 advanced concepts into one or two words and I always have to ask it to “unpack”.
I don’t think it’s easy on native speakers when it happens, but it’s even harder when you’re not.
Still debating whether these are actually information dense or not. Complexity
gives the appearance of density, but these sentences always come in whole pages at a time while often failing to deliver the needed information.
Few ever cared. Find one non-pedant who would object to the personification that follows:
"The evening settled over the city, drawing the light out of the streets one corner at a time. Windows blinked awake with lamplight, and the wind moved through the alleys restlessly, leaves brushing against walls before gathering themselves along the pavement. In the distance, the river kept its steady argument with the stone embankments. When the night pressed in, the weather became increasingly angry, until it was a raging storm."
In the affective sense, evenings don't settle, and street lights are not drawn out, windows don't blink, and wind isn't restless. Weather can neither be angry, nor rage.
But such personification is a natural part of how the English speak.
See also "wind moved restlessly", "weather became angry". And raging storm? I mean come on... I won't even put that last one in quotes.
And like a sibling reply pointed out, personificiation is not the same as anthropomophism. Nor is plagiarized personification. It has no inner thoughts, and no fondness of anything. It's nothing but a cheap, superficial facsimile of human writing and nothing more. Great for form filling and boilerplate though. Not so great for anything else.
> See also "wind moved restlessly", "weather became angry". And raging storm? I mean come on... I won't even put that last one in quotes.
What's your point here?
Mine is: Nobody seriously objects to phrases such as these. So for me, in this case: the more cliché, the better.
Or another way, you were previously objecting to a human using "typical English" to describe an LLM's failure to use "typical English".
> personificiation is not the same as anthropomophism
An unimportant though technically correct statement. In both my example and the comment from xg15 that you previously objected to, "personificiation" is the correct term.
It annoys me more that:
- "The wind moved through the alleys restlessly" rather than "the wind moved restlessly through the alleys".
- "The evening drew the light" _from_, not _out of_ the streets. The light is not _inside_ the streets, unless perhaps you're talking about streetlights going out.
- _Themselves_ is unnecessary.
- _The river kept its steady argument_ also sounds off to me, but I don't have a better suggestion (_kept up_ doesn't sound any better).
That's fair, but the point wasn't to provide the best possible literature, just something good enough to show it's unobjectionable to speak of things as if they have a mind even when the things are obviously mindless to anyone who isn't a panpsychist.
This is another "AI-ism" I noticed, mostly in coding agents - they seem to be very fond of making up new "compound nouns" (and occasionally verbs) to sum up relatively complex and specific concepts into single noun phrases. I wasn't sure if it's to save tokens or if the AI uses this to get a concise "identifier" for a concept that it can refer back to later, but I found it very noticeable.
I find the resulting sentences hard to read, though it does get better if you're aware of that tendency and make a conscious effort to parse the noun phrases. But I guess since it's just intermediate output from coding agents and not text for essays or blog posts, it's fine.