Two well-known tech observers have flagged an emerging side effect of heavy LLM use: frequent readers and speakers of model output are starting to echo distinctive phrasing in their own conversations, including terms like seams for technical boundaries and constructions such as the shape of or genuine.
Let me know if this is just me:
Noticed someone I know who is very "AI-pilled" and uses agents 24/7 to... start to talk noticeably more like these LLMs write.
Eg more heavily using adjectives like "geniune", frequently terms like "the shape of" and many more examples
The pattern shows up most clearly in people who interact with agents daily, mirroring how long-term colleagues adopt one another's cadence and word choices without noticing.
No broader surveys or timelines exist yet, so it remains unclear how common the shift is or whether it fades when exposure drops.
Many users reacted with worry to heavy AI users adopting LLM writing styles in everyday speech, calling the mimicry scary and a reversal where models end up training humans.
No Digg Deeper questions have been answered for this story yet.
Sage observation by @grinich
We follow our tribe, don't we? (And for some people, the "tribe" might be their favorite LLM they spend the most time with)
@GergelyOrosz I think this is a general phenomenon. As an example, I've noticed that folks who work at Stripe for many years begin to speak with the same intonation/cadence/structure as Patrick.
We adapt to the tribe.
@GergelyOrosz I think this happened to me - I used the term "seams" recently to describe API boundaries and plugin hooks, a friend pointed out that's a popular term used by LLMs, I'm pretty sure I'd never used it before
Let me know if this is just me:
Noticed someone I know who is very "AI-pilled" and uses agents 24/7 to... start to talk noticeably more like these LLMs write.
Eg more heavily using adjectives like "geniune", frequently terms like "the shape of" and many more examples
Just to add: this is a change from ~a year or two before.
I guess it should be surprising that what we read and listen to shapes how we speak. And so if we read (+hear? watch?) the output of tools like LLM more, they will shape language?
Let me know if this is just me:
Noticed someone I know who is very "AI-pilled" and uses agents 24/7 to... start to talk noticeably more like these LLMs write.
Eg more heavily using adjectives like "geniune", frequently terms like "the shape of" and many more examples
@WesEklund yes, the terms they use, grammatical phrases etc when talking live
@GergelyOrosz the weirdest part is they probably don't hear it anymore. their internal voice is just claude now
@simonw @GergelyOrosz I think it’s similar to cultural language adoption
Sometimes it’s because you see it all the time, sometimes it’s because you realize it’s a term you can use
I’m sure everyone does it
@GergelyOrosz You're saying their speech when talking to people, is actually changed to be more like LLM speech?
@GergelyOrosz There's a reason I have been avoiding caveman. I tried it, found it cool and wondered if it's really that good at saving tokens. But even if it is, it's a high price to pay if I end up speaking like a caveman one year in to using it.
@GergelyOrosz Not surprising. I caught myself doing it, so I force myself to manually write first drafts plus have a writing coach skill help me avoid AI slop. Still, I read so much AI output that I'm no doubt slowly and subconsciously changing my style as well.
@zeeg @simonw @GergelyOrosz Yeah I spent a couple of years in meetings with PMs then caught myself one day saying "learnings". You are who you talk to.
@grinich Good one from Simon.
Also, language keeps evolving (just thinking how one generation uses different slang + lingo than the previous one). So probably not too different to this?
@GergelyOrosz I think this happened to me - I used the term "seams" recently to describe API boundaries and plugin hooks, a friend pointed out that's a popular term used by LLMs, I'm pretty sure I'd never used it before
@ku_ds17868 unironically yes!!
@GergelyOrosz no suprise it's happening to me and I use it a lot less than that
@GergelyOrosz This is why it’s so critical to have friends and family outside the tech bubble to nut slap you when you pull shit like this
@simonw @GergelyOrosz I've noticed the same thing with myself. I've also used "seams" and "phase" more often than usual.
@GergelyOrosz TBH, I have noticed this in myself, and have to check it. I started using "load bearing" in convos. As humans, we absorb from the company we keep, and if I am taking to an AI 20 times a day, it seeps in.
@GergelyOrosz Yeah I find myself dropping substrate, goblin and gremlin in casual conversation. I am merely a vessel.
@GergelyOrosz Recently for the first time heard someone say "this distinction is brutal".
@davidcrawshaw @simonw @GergelyOrosz This is why I consider it my duty to make sure all 3 year olds can say “ok boomer” confidently
There is an old saying: you write as you read. And in this case if most of the stuff you read is LLM generated, then yeah, it's not so crazy that not only you start writing using certain shapes, which then goes into thinking in certain shapes, which ultimately translates to speech