Taylor Lorenz’s Power User - Is ChatGPT Changing The Way We Talk? w/ Adam Aleksic

Episode Date: August 6, 2025

Buy a subscription to my Tech and Online Culture newsletter, User Magazine to support my work!!!! 🙏 https://www.usermag.co Since its launch in 2022, ChatGPT has become the fastest-growing consumer application in history. The tool is embedded in more and more areas of daily life and new research shows that it's also transforming how we write and even speak. A group of researchers analyzed more than 360,000 YouTube videos and 771,000 podcast episodes from before and after ChatGPT’s release to track the use of ChatGPT-affiliated words like "delve" and "examine."  All of this has major implications. Adam Aleksic is an etymologist who studies the way the internet is reshaping language. His new book Algospeak tackles all of this. He joins me to discuss how ChatGPT and AI is transforming the way we talk, what words we use, and how we communicate. Follow me:https://www.patreon.com/c/taylorlorenzhttps://www.instagram.com/taylorlorenz https://www.instagram.com/taylorlorenz3.0 https://www.tiktok.com/@taylorlorenz

At least our perception that chat GPT is kind of like a human is affecting the way we communicate to it and with it. Since its launch in 2022, chat GPT has become the fastest growing consumer application in history. Hundreds of millions of people have used the tool, and it's increasingly embedded in more and more areas of daily life. It's transformed how people discover and learn about the world, and now new research shows that it's also transforming how we write and speak. ChatchipT and other large language models are altering the vocabulary choices that appear in written communication like books and research papers. And now we have proof that it's shaping speech communication too. A group of researchers analyzed more than 360,000 YouTube videos and 7771,000 podcast episodes from before and after ChatchipT's release to track the use of ChatchipT-affiliated words like delve and examine. What they found was a massive surge in ChatchipT-affiliated words in the 18 months after the product's release.

The words didn't just appear in formal scripted videos or podcast episodes either. They were even included in more random, off-the-cuff, spontaneous conversation. The studies co-author Levin Brinkman at the Max Planck Institute for Human Development said, quote, the patterns that are stored in AI technology seem to be transmitting back to the human mind. This is creating a new type of feedback mechanism. between AI and humans. We train AI on human writing. It parrots a statistically remixed version of that text back to us,

and then we pick up on those patterns and unknowingly start to mimic them in our own offline day-to-day communication. All of this might have major implications. As Hiromu Yakura, the study's other co-author said, quote, word frequency can shape our discourse and arguments about situations. That carries the possibility of changing our culture. Adam Alexic is an etymologist who studies the way that internet is reshaping language. His new book, AlgoSpeak, which is so good, just came out and it

tackles a lot of this. I thought he'd be the perfect person to discuss how chat GPT and AI is transforming the way we talk and communicate. Hi, Adam. Welcome. Hi, Taylor. Excited to be back on here. Okay, so I'm so excited to talk today because I feel like chat GPT speak or whatever you want to call it is everywhere. And I feel like I have started to notice it myself where I read certain things now and I'm like, that was chat chach. T. Like, I know that that person is like, written something with chat chabit, or the way that people talk. You've been talking about this for a while. And I guess just to start off, I'm wondering, when do you think that chatubes T and AI really started to affect our language? Yeah, you can actually kind of directly see this between 2022 to 2024,

there was a 1,400% increase in the word delve in academic research. And delve for those who don't know is one of those words that is disproportionately used in chat GBT, and we can get into like why words are disproportionately used. But we can sort of see the exact graph of when this happened. And this is in academic research, mind you. There's so much of this being used outside of it. There's so many other like moments where people sound like AI. And there's also like the data to back it up that around when, I don't know, Chagipete 3 came out,

that's when people really started like using it more in their workflow, using it more to cheat on tests or to write articles or whatever. But the reality is that the internet is flooded with stuff written by LLMs now. Were there any other words that stood out aside from DELV that you really started to see rise. Yeah, commendable, meticulous, tapestry is famously one. The M-Dash, people say if you use the M-Dash too much, you're writing like chat GPT, which is, which really sucks for people who like, you know, the Victorian-style writing. I use a lot of M-dashes. I don't do it because of Chad-G-T, but nevertheless, like, maybe we're more likely to get accused of sounding like chat-GPT, and maybe we're going to over-correct or something to use fewer M-Dashes. I cannot use the M-Dash anymore,

because I feel like people just immediately associated with AI, and I also associate it with AI. I was reading this great Verge article titled, You Sound Like Chatty BT. And the subhead is AI isn't just impacting how we write, dash. It's changing how we speak and interact with others. And there's only more to come. And I think that they actually intentionally phrased that subhead. They definitely put the dash there because that sounds like exactly the way an LOM response to, yeah. Exactly.

I feel like we all kind of. kind of have this like understanding of how these chat GPT's AI models talk. How can we kind of describe this very specific way that AI communicates? Yeah, I mean, I've seen it referred to as LLM speak or chat speak. There's there's really no like one unified term for maybe we could invent it right now. But definitely like this is something people are aware of. The forefronts are in academia education and high schoolers are they're still using chat but they're changing words or changing the, you know, the well-known stuff like the M-Dash to make it look less like it's written by an LLM.

So we have like also like this sort of avoidance speech that's happening. What's avoidance speech? Avoidance speech is this phenomenon where we tend to circumvent a topic. And it's usually for some sociological purpose. There's some kind of cultural reason why we tend to be disinclined to speak in this topic. Other cultures often do it for religious purposes. For us at this point, chat GBT is so culturally present that we are using avoid in speech to avoid sounding like this cultural taboo of sounding like an LLM. Got it. So what you're saying is that people are actually sounding less like AI because they're so nervous of sounding like AI?

I think it's two functions. It depends on who's speaking that way, right? There's the surface level stuff that's very obvious and commented on and well known. So we all know at this point that delve is an AI word, that tap is an AI word, plethora, the M-Dash. We know that and we actively reroute around it. On the other hand, there's the stuff that we aren't commenting on as much, the stuff that isn't as obtrusive doesn't stick out as much. And those words probably are worming their way into our vocabularies without us noticing it or without us being able to really identify it. I feel like it's this really specific sentence structure too. And I notice it maybe because

I already write poorly like this. It's kind of like the way you learn to write when you have to meet a specific word count and you're kind of like intentionally elongating sentences? Right. I definitely think that's the thing. There's no word for that kind of speech. Again, I think there should be more people writing about this and coining that, but I think you're absolutely right. Perhaps there's something to the tone of how the AI is trained

because it's meant to sound like this robotic servant or something. It's probably been engineered to sound a little sycophantic, a little responsive and informational. So all of those things probably play a fact. into the ultimate tone of the sentence that it's spitting out. You mentioned that this really kicked off 2023 and obviously 2024. I feel like that was the year that all these kids started using it in school. But hasn't our language always been shaped by technology?

I mean, even things like autocorrect or spell check or other things where like Google's suggesting, you know, the rest of a sentence. Like, I feel like this is maybe just a continuation of sort of the way that tech has shaped our language for a long time. Yeah, language is always going to reroute around the tools that we use it for. I've noticed that my own handwriting, I don't write as much with pencil and paper. And I went back a few days ago and I really wrote something out with pencil and paper. And I noticed I was misspelling more words than I used to, even words that I knew.

And that's because in my workflow, when I type things, the word automatically corrects, or I type it a certain way, or there's predictive text sometimes. So I don't even rely on the same mechanisms in my brain for spelling words correctly, and our spelling is actually rerouting around spell check. That's definitely happening. Like even before the internet, why don't you? we adopt certain words, the word hello was popularized because that was the standard greeting to use on a telephone. It wasn't really around before then, but around the 1880s, it just happened to be a trending greeting.

It was recommended in phone books as the correct way to start a phone conversation because people didn't know how to do the back then. And so hello spread purely because it was part of this new medium of a telephone existing. So our language always reroutes around new mediums. That's not new at all. Yeah, and you've written and talked a lot about the rise of Algo speak, the title of your book. Yeah. Which is also this idea of algorithms shaping our language, right? And like content moderation filters and forcing us to use different words and things like that, right? Well, you can't separate algorithms from what we're talking about with chat GBT here,

because they all rely on LLMs. The algorithms process words the same way by, you know, tokenizing and embedding them. And now we have this representation of numbers about what a video is about when I upload the video. In the same way, chat GPT turns a sentence I type in into a token of numbers and figures out what to do with that. And there is an algorithm there too, right? So it's all kind of... of the same thing happening in different fonts.

There's been all these headlines since I feel like the concept of Algo speak entered the public consciousness and also just like articles about chat chachy-tee-tee-being language that I think are kind of like panicky. A lot of people are like, this is the end of human expression or like it's reshaping our language to, you know, neuter us in some way and we're all gonna just end up speaking like LLMs. Do you think that the rise of Chachy-Bt and LLM use is constricting the way we express ourselves? I'm not really that word. Back when we were using Stone and Chisle, that was of years ago. We didn't have any spaces between our words because it's too much effort to like it's you're constrained by the stone

It's like takes up unnecessary stone space to put a space or a new paragraph But once we like came around with paper and we started writing things on paper we introduced the idea of chapters and spacing and paragraphs Did that ruin the medium of writing? No, we found a new evolution for it. It's a separate question whether or not it's affecting things like creativity Language however remains a flexible tool that humans use to relate to each other. At the end of the day, that's what it is. And we're always going to have ways of relating to each other. So I'm not that concerned. It reminds me of a lot of panic in like the early 2000s when I was a kid of like people using AIM text speak or whatever or like T9 speak. I don't know, even know if you remember this, but like, you know, you could only text certain things, right?

This is where like LOL or like IGTG, whatever, where like parents would be like, please write it out. Like you are, you know, you're censoring yourself or this is some sort of like bad thing that teams are using these acronyms. ultimately, I think it ended up being sort of overblown. But I don't know. I mean, with LLMs, I think one thing you hear a lot is like chaty p.D helped me express myself. Like I was talking to a friend recently who used it actually for an apology message. And there was a lot that she wanted to communicate that I feel like she couldn't put into words without engaging with one of these large language models. And I did end up reading her apology and it was pretty heartfelt. Like I didn't

perceive it as being written by AI. I don't think the person she sent it to perceived it that way. And And it ultimately helped her express herself more and sort of like widened, I guess, like, the vocabulary that she was using to talk about her feelings. So do you feel like this can be helpful in any sense? Certainly, I think when, you know, these tools are used not for cognition, but for guidance and for editing, they can be very useful for finding new ways to express ourselves. I think maybe the concern comes in when you let the AI think for you instead of you having your own thoughts and where you want to guide this.

Like, ultimately, it's a new canvas for us. to use, we shouldn't let the canvas dictate what kind of art we make, but that does end up happening a lot. I think maybe another big concern is a homogenization effect happening through the algorithm. These models fundamentally are predictive. They generate some amount of randomness, but the word that is generated in each part of the sentence is what tends to be a more likely word. And human language, there's more randomness, there's more entropy, there's more maybe poetry in how we communicate because we find these weirdly little cross connections that N LLN might not generate as easily. Yeah, it reminds me of Kyle Chaka's great book, Filter World, which is all about kind of

algorithms and homogene and, you know, kind of how things, we sort of all end up in this sort of like bland neutral if we just keep following the algorithms down the rabbit hole. And that got me kind of wondering, too, of like, if we'll come to appreciate more like artisanal bad writing or weird writing, like, I think of the, did you ever read the book, boy? No, I haven't. It was this like novel that was kind of like a viral novel. It's written from the perspective of this guy who's like a boy kind of.

It's hard to even describe, but it's written in this really distinct writing style that is very human. I wouldn't be surprised that there's like a cultural backlash and people move. Like, what was the, why do we have that trend? ICLTSPMO. We had this slang abbreviation trend popularized earlier this year. At this point, LLM's already widespread. ICLTSPMO stood for I can't lie, this shit pisses me off. But there was like this whole host of abbreviations similar to that that just sound ridiculous that people were.

spreading and why are they spreading them? Why are these things funny? It's because it's sort of a reaction to this overly ordered world we have. It's something that only humans can understand once the machines catch up, then the meme is dead. I think we're very good with our language evolving in reaction to these new technologies. Yeah, that's what I think too. It's like I think that once things start to become boring, then we're drawn to other things that are new and interesting. And so I'm not as panicked, I guess, about the blandness, at least not in like day-to-day communication. But, you know, you mentioned academic papers and sort of the consistency of certain words used across that. I definitely notice it more in like articles and papers and things like in that context.

Do you think that those sorts of things will be subject to kind of like the LLMification of language? Like if all of that training data is pretty homogenous and then people are writing more research papers and more articles that are also homogenous, it'll just sort of like... And the models are trained on themselves and they end up recommending things based on previous things that they've already output and then we end up in this positive feedback cycle. it's definitely concerned. And when I was saying earlier that human language operates kind of on this serendipitous way of finding ways to express ourselves. We engage in lateral thinking our creativity comes out of finding cross connections across things. LLMs are deterministic. They find the most likely thing and you can turn up the randomness, but they're always going to be finding the most likely thing. Which means that, let's say you write some papers that are pushed higher in the Google Scholar algorithm and now ChatGBT, he recommends them higher. And we were several degrees abstracted from, let's say you just find a random book. in a library that helps you better articulate your thesis if you're writing an academic paper. Now, this random thing might not show up in the super-filtered version of this homogenized space of academia we're engaging with. And so maybe it's harder to come up with true cross-connections that help us think creatively. And beyond thinking, maybe it works the same with generating poetry,

generating art. Speaking of how AI is shaping language, you've also talked about this sort of swap communication thing happening. Can you explain that phenomenon? Yeah, so it's directly in the interest of these social media platforms to be pushing slop because you can edge out human voices and creators and therefore pay them less money in the end. And Slop is part of their grand plan for social media. We saw meta loosening their guardrails for AI actively incentivizing AI generated comments, AI generated profiles. There's a lot more of that. We're in this period of epistemic uncertainty right now where we're not sure what's real and what's not real. And when we're talking about LLM's affecting language, we have to also be thinking about the visual communication we're getting,

what's actually dominating the rectangle of our screen on our phone when we're looking at things, and whether it's real or not real. And now that we're not sure what's real, we maybe face an extra need to distinguish ourselves as humans. So I do think there will be an increase in people trying to sound less like slop, sound less like chat, GBT, reclaiming human kind of communication. There was a big boom in lavalier microphones on YouTube back when YouTube started getting more professional, and you'd hold the lavely and communicate authenticity, even though there's no actual, like, it's meant to be clipped on.

There's no reason to be holding that microphone. There's the internet ugly aesthetic that's been commented on since I think like a 2014 Atlantic article. But it's a consistent thing that we communicate authenticity online by appearing more like real humans. And that will be reflected in our language and in our aesthetics. I think that's such a good point because I have even thought about this myself of I'm taking this YouTube thumbnail editing course.

online because my YouTube thumbnails suck. I know everybody watching this knows that. I make them myself. That's why they're bad. And they're basically just encouraging us to use as many AI tools as possible. Like that's like what the whole thing seems to be. And I was thinking like, I feel this like desire to kind of actually make it look even more human. I wish there was like a stamp that I could put and be like, I actually made every part of this, you know? And so yeah, it sounds like you think that there will be this desire from people to consume maybe more organic and original content that's not tainted by AI. And so speaking in a really weird a different way or using a very sort of like quote unquote authentic visual communication will entice more people.

I mean, wouldn't that just sort of also incentivize the algorithms and the AI to copy that to replicate those formats? Obviously, I think that's also what happened with the Italian brain rot trend. So Italian brain rot was this like series of AI generated animal images, animals hybrids with like difference like a shark wearing sneakers or a fish that's also a wheel. And so this is like the absurd genre of Italian brain rot. And that emerged as a genuine human use of AI. that felt real because even though we know this is AI, we acknowledge that, it's nevertheless humans reclaiming power in how they use it by defining the categories that are expected of them, by defining the expectations.

Then later on, we get these AI slop accounts, which are just churning out Italian brain rot after Italian brain rot meme, and now the trend is dead because we feel that it has lost its authenticity. A similar thing happens with every single internet trend. I think it's happening faster and even faster because of AI, but the algorithms kick started it, and AI is pushing it even for What trends are you seeing in terms of language with the rise of AI? Like you've written so much about social media and how social media algorithms are impacting speech. Aside from words and sentence structure, how else is AI upending the way that we speak and communicate?

I've seen a few examples of prompt speak where people speak to each other as a joke the way they would be prompting chat GPT or something. A lot of the slop on Instagram Reels right now is jokingly playing into the idea that it is AI generated. So there's characters that are AI generated that might say, oh, no, the prompt is making me drink more alcohol. I don't want to be doing that, but the prompt's making me do that. And then there's also real videos that replicate the AI generated videos. And that's sort of playing into them. These videos are making me lose my mind because some of them, like, ever since people started recreating the prompt videos, IRL, I feel like I have to like zoom in on the text in the video to know sometimes if it's AI generated or not.

But I think it's really funny that, but I think people are really interested in this idea of like, the AI becoming self-aware and creating these like self-aware characters within AI and then I guess replicating that in real life. I think there's a real trend of people treating chat GBT or AI in general like it's a god, like it knows everything and it is in engineering a black box we don't know what's going on. It's like the human mind at this point where there's inputs but we don't understand what happens up here. The best neurologist can't actually tell you what happens up here and then we have outputs. Same with AI. There's inputs. We train them and at a certain point it's deep learning that we don't know

what's happening. There's billions of things happening at once. It just spits out an output and we hope for the best. So we can't tell a question, like, whether this is conscious or not, whether it's sentient or not. I just want to hold up right there because I did just make a video about this and I do want to be clear because a lot of people are falling for these conspiracies about like they've awakened their AI, it's sentient. Ultimately, these are large language models. So they don't have the kind of like sentience or souls or whatever you want to call it. I think the prompt thing is interesting too, because when we think about how chaty p.t is affecting language, I also think that the way we're speaking to it changes.

Like a lot of older people will speak to it as if it's another person or human, and a lot of younger people treat it like a Google search bar or they're just a lot more direct. And you saw these conversations come up where actually these AI companies were saying that they're wasting lots of resources on people saying, please and thank you to the AI, and they'd rather you be more direct. So I feel like even just the way that we ourselves talk to the large language, which models is evolving. There's been a rise in young people, especially Gen Z, Gen Alpha, referring to chat.

GBT is just chat. And they'll say, let me ask chat in the same way as we might in the past say, let me Google it as a verb for referring to this company. It's sort of crazy to me that chat has just become this regular term in the vernacular. I still say chat GBT. I like kind of saying that because it reminds me that it is this foreign entity. Because we'll also use chat as like this pronoun, you know, for this omniscient entity that's watching us like a Twitch chat, right?

And using AI and analogizing it to human communication is potentially changing things. At least our perception that ChatGBTGPT is kind of like a human is affecting the way we communicate to it and with it. Yeah, I think it also affects how it delivers results. And I think this is where people can end up going down a rabbit hole where they end up giving it a name and sort of conversing in an overly personal manner, which then prompts large language models to respond in an overly personal manner. Whereas if you're just putting in, you know, date of civil war, or whatever, it'll just spit back pretty much raw information. I think there was a real problem, and this was true with algorithms as well in social media,

that we tend to personify them, simplify them, and it anthropomorphize them. When we treat chat GPT as a single human friend, now we're thinking about it differently. We forget about the fact that it turns all of our language into embeddings, throws them through this black box of understanding how tokens and vectors work, and then spits out an answer. We forget that there's this complex process happening, and in treating it like a human,

do perhaps understand things differently. Another thing that has been interesting is AI editing tools, and I know these aren't exactly the same as large language models, but for instance, I use an AI tool to edit out a lot of white space in my TikToks or other short-form videos. I kind of started it to make them shorter, and now I just do it as a habit. I know there's these products like Descript as well that also edit out ums and ahs and these weird ways that we speak.

And I think it's already affected me where now when people don't do that, it seems annoying to me. even though they're communicating in a normal human way, I'm kind of like, why are you taking so long? Why are you taking these breaths between takes? It seems like we're just, we want things to sound almost more text to speech and robotic. Yeah, maybe that's part of the rise in prompt speak. And maybe where communication will be more concise in some contexts,

but in others where there is this need for authenticity, maybe it is good to throw in a few ums because the, yeah, the LLM would never do that. There's just been a general trend with algorithms and making language more rapid and concise, because that's what goes viral. And LLM's AI are contributing to that through, you know, their own angle. All of this discussion about AI's impacts is, I think, filtered through the media that ends up pushing a lot of scary headlines. And I want to read you this one quote from an article in LA

magazine. It's talking about chat GPT. It says, we are losing the autonomy of our speech and the trust in our communication with others. And soon we may lose agency over our own thinking. Instead of articulating our own thoughts, we articulate whatever AI helps us to articulate. We become more persuaded, Naman says. I'm curious your thoughts on that. There's valid concerns being raised. I think at the same time, I wouldn't have been concerned if we've added spaces once moving to the paper medium from the stone and chisel medium. So we need to differentiate between when is AI doing cognition for us?

That does feel a little bit concerning. I think humans should be thinking. That seems like a good thing. But I don't think it's bad when it just helps us reroute language in a new way. that makes it easier to communicate in a lot of aspects. Maybe one of my biggest concerns is there's a bias in the language that algorithms and AI had in the first place. So we have to ask ourselves, what does it mean to sound like AI

and why do some people sound more like AI than others? Why is AI using the word delve more? And it comes down to how these models were initially trained. The word delve is disproportionately used in Western Africa, where open AI hired a bunch of people to train these models by inputting text data. And so they put in the word delve way more than people in the Western hemisphere. And now it seems weird in the Western hemisphere that the word delve shows up so much.

And so somebody who is just West African could be way more accused of speaking like AI simply because that's how AI was trained. I think that's terrifying. Yeah, it's really concerned to me. And also like the M-Dash thing. I've liked my M-Dashes. I think there's been enough acknowledgement with it that you can use it, but I'm still traumatized by like the association and so I cannot. Well, Adam, thank you so much for chatting with me today. Tell me where people can find your book because it's fantastic.

Thank you, Taylor. Yeah, it's called AlgoSpeak. You can find it in most bookstores online. Just, yeah, search AlgoSpeak. I'm also etymology nerd online, but buy the book first. All right, that's it for the show. Don't forget to subscribe to my tech and online culture newsletter, Usermag.com. That's UserMag.

Where I write about all of this stuff and more. You can watch full episodes of Power User on my YouTube channel at Taylor friends. And if you like the show, please give me a rating and review on Apple Podcasts, Spotify, or wherever you listen. My bestselling book, Extremely Online, is also now finally out on paperback with a brand new, amazing cover. Pick it up wherever books are sold. Thanks for watching and I'll see you next week.