Science Friday - AI Word Choice | When Dwarf Lemurs Hibernate, Their Chromosomes Do Something Odd

Episode Date: March 27, 2025

Certain words are overrepresented in text written by AI language models. A study investigates why such patterns develop. Also, the ends of chromosomes, called telomeres, typically shorten as an organism ages. But when some fat-tail dwarf lemurs hibernate, they lengthen.‘Delving’ Into The ‘Realm’ Of AI Word ChoiceSeveral years ago, some eagle-eyed readers of scientific papers noticed an unusual trend—an increase in the number of abstracts using certain words. The terms, including “delve,” “realm,” “evolving landscape,” and more, were suddenly appearing more often than they used to.Researchers analyzed the abstracts and compared them to abstracts written just a few years earlier, before the widespread availability of artificial intelligence large language model chatbots. They came to the conclusion that abstracts written by AI were more likely to use words from a list of around 20 favorites than regular human speech. The question was, why? If the models were trained on conventional writing, how did a preference for words such as “delve” creep in?Host Flora Lichtman talks with Dr. Tom Juzek and Dr. Zina Ward of Florida State University, who set out to try to understand the origins of some of AI’s favorite words.When Dwarf Lemurs Hibernate, Their Chromosomes Do Something OddThe fat-tail dwarf lemur is one of the only primates that hibernate for the winter. A new study published in the journal Biology Letters takes a closer look at what’s going on inside lemur cells when they are in this extended phase of suspended animation. It turns out that their telomeres, the ends of the chromosomes, actually grow longer when the dwarf lemurs hibernate. Typically telomeres shorten as we age, as cells continuously divide. So, what exactly does this finding mean for lemurs and other primates, like humans?Host Flora Lichtman talks with the co-authors of this study, Dr. Marina Blanco and Dr. Lydia Greene, research scientists at Duke University.Transcripts for each segment will be available after the show airs on sciencefriday.com. Subscribe to this podcast. Plus, to stay updated on all things science, sign up for Science Friday's newsletters.

This is Science Friday. I'm Flora Lichten. Today in the podcast, the annoying words large language models overuse. And what that means for how we talk. It's like someone pulled a switch and here's the elf. Here's intricate. You know how when you read something, you can sometimes get a picture of the person who wrote it. Like if I start talking about touching base to synergize after we circle back, first of all, please call for an intervention. but second of all, you might reasonably suspect that someone from the e-suite got a hold of the script. Or if I talk about giving 110% or winners don't quit, you might clock me as a sports person. Well, in recent years, people started noticing that AI chatbots, large language models, had their own linguistic tells. Certain words that seem to crop up a lot more often in AI sentences than in human English.

Words like delve. and realm. And it wasn't subtle. It was a big sudden shift in word usage. Joining me now to talk about that are two researchers who decided to delve into the realm of AI linguistics. Tom Uzek is an assistant professor of computational linguistics in the Department of Modern Languages and Linguistics. And Zina Ward is an assistant professor in the Department of Philosophy, both at Florida State University in Tallahassee. Hello to you both. Welcome to Science Friday. Hi. Thanks for having us.

Yes, thanks for having us. Tom, I have to know more about AI vernacular. What are some of the words that AI loves to overuse? Some of the words that we found, and it's quite a list, is the more obvious ones, like Delph, Garnet, Realm, an evolving landscape. Evolving landscape is one of them? I've seen that a lot. So I think that's a candidate. And the more subtle ones are something like underscore, emphasize and showcase it. How did you figure this out?

It started with a discussion on social media that people noticed that certain words, especially Delft, really suddenly popped up everywhere, especially in scientific writing and education. And so people started to look into these more systematically. But then what interests us from the beginning was the why is that happening? Why does this happen? So our hypothesis, which we investigated with a couple of different experiments, is that the models prefer these words because they were trained on human feedback, where the humans exhibited a preference for those words.

So that's the thing we wanted to test, the idea we wanted to test. And in short, what we did is ran a couple of online studies where we asked participants to judge the quality of abstracts with and without those words. And we found in the most recent run of the study that participants do exhibit a very slight preference for abstracts with words like showcase and underscore and comprehend. And so the models are trained to produce responses that human evaluators like. The models learn from human feedback in a certain stage of their training. And so if these human evaluators are exhibiting that preference, the models are going to pick up on that and reproduce those words more frequently. So that's what we think is

happening. That's so fascinating. I mean, to me, delve and evolving landscape really feel like words I might hear in a Silicon Valley conference pod or whatever. Is that a coincidence? I mean, is that who's training these models? Well, we don't know for sure. Tech companies don't publish the details of their training procedures, but it seems likely that the workers in general are, employees in the global south. The tech companies often outsource that kind of labor to countries where wages are lower. And so those workers are seeing potential outputs from the models and providing their feedback on which ones are higher quality. And then the models are being trained to reproduce those sorts of responses. So it's really probably the preferences of these employees

in the global south that are filtering through and shaping the lexical choices of these large language models. Yeah, that's fascinating. Tom, I mean, we've seen linguistic fads before. You know, I'm old enough to remember the meteoric rise of touch base. How is what you observed here different? Rather rapid language change is also nothing new. Touch base, looked up some millennium speak. Awesome, cool.

What they exhibit when you look at the data over the time is that they have a period. of several years of pretty steep rise, almost like an S-cur. They start to rise, then they have this period of really quickly gaining traction, and then they plateau out. Sometimes we do see a sudden spike, which is when a real-world event occurred that motivates that spike, concretely SARS or Omicrom. Now what is different than now observed changes is

that there is a sudden spike, almost like a step function with an abrupt, crew, but there's no real-world motivation why it should be Delph or intricate or nuanced. And that is different. You mean like it's, it's, it's, it almost feels random. Yeah, it's like someone pulled a switch and here's Delph, here's intricate. I mean, that seems only possible using AI, right? Because humans maybe don't change as fast. Yes. And that is one of the areas where still a lot of research is directed to, the question, do these words really sit in our language systems now? I mean, is there delve in or intricate in the abstract of a scientific article or a student essay because

people have this in their language system and this is what they put down? Or is it actually tool usage, AI usage? And so far, this is under research. I mean, it's an hypothesis one way or the other. And I think what we will be seeing sooner or later is where people look at spontaneous spoken language. And I think chances are they will find some traces of these AI words having entered the human energy system. Well, I'm really interested in this. I mean, we talk a lot about us training AI. But is AI training us? I mean, is there evidence that these AI are changing the way that we talk and we write and we think? Yeah. The state of the research is that this an hypothesis and the key here is to look at spoken language. Most research has focused on

written language. And when it comes to spoken language, I think it's fair to say we do not see these abrupt changes that we've seen in written texts. But now there is first research coming out. There's an analysis of YouTube talks, which is still scripted spoken language, where there has been an uptick of these AI-induced words. And there is research in the making, I have a student working on this, that when it comes to semi-spontaneous spoken language, conversational language, there seems to be an uptick, but that seems to be more mixed.

I mean, that, to me, seems really profound. Absolutely. I mean, and it's interesting, too, that to the extent that these models are changing our language and to the extent that they're preferring, about words are shaped by these workers in the global south that we've talked about. It's a really interesting inversion of the usual way that linguistic change happens. So linguists and historians of language have studied over time how changes in language in the Western industrialized world

trickle outward, especially in this sort of globalized culture and economy to everywhere, to other English-speaking parts of the world, if we're looking in English in particular. And what's interesting about these linguistic changes is that if our language really is being shaped by LLMs, and LLM's language is in turn being shaped by employees of tech companies in the global south, there's a kind of reversal of the usual direction of influence. And I might come in on this. There are two sides to this. One is today's produced data will become tomorrow's training data for these models.

So there's a chance that we will see a loop and all this accelerating. And then the other side, there is what one could call the creep infector. We observe AI influences human language to quite some degree. But it might well be that when we reflect on it, we might come to the conclusion, this is not something we actually want. And so some of that language that now enters our sense. system does so beyond the level of perception. And items like underscore multifaceted or necessity, they're less discussed.

But of course, that links to the discourse of a more concerning version of this creepy-in factor, which is, say, undesired political beliefs, right, that then gradually seep into our belief system. And I think this is why the research to try and identify where this model behavior comes from matters to a good degree. Can these words be used as a diagnostic, like as a way to tell that something was written by AI? Not with high confidence. As you probably know, AI detection is a really, it's a cat and mouse game, and it's really quite difficult. So we're, I think, a little bit pessimistic about the potential to use this for diagnostics. But it certainly raises eyebrows,

I think for us when we see these words, it raises questions, even if it doesn't provide a conclusive answer. Zena, you're a philosopher. Why does this interest you? I'm a philosopher of science. And so these models just fascinate me. I mean, I think like many people, I'm amazed by how well they work. And it's all the more amazing, given that we don't really understand why they work as well as they do. And so it's also a kind of intellectual challenge, trying to figure out how they work, because it's not like you can just pop the hood and look at the word choice module, right? They're really complex systems. So you have to be quite creative in how you probe them to try to figure out what they're doing. So, yeah, for me, that's why

it's of interest. It's a complex system that really requires some ingenuity to understand. And Tom, what about you? From a linguistic perspective, I mean, what we're seeing right now is almost unprecedented language change. I mean, the shifts in word usage that we're observing over such a short period of time, that's really remarkable. I say almost because we've seen in history, I mean, we've seen technology influencing language. I'm thinking like the printing press, the internet, social media. But what we're observing right now is quite something.

We could be entering a period of rapid language change. So what we're seeing is not just a year or two, a few dozen words, but really this continues for an extended period of time. This is one possibility. The other possibility is that now we had two or three years of rapid language change and it will kind of fade out. And we don't know yet what is going to happen. We'll find out.

But there is a chance that we will see considerable language change over an extended period of time. I want to thank you both for joining me today to talk about this. Thanks for having us. Thank you so much. Tom Eusek is an assistant professor of computational linguistics in the Department of Modern Languages and Linguistics, and Zena Ward is an assistant professor in the Department of Philosophy, both at Florida State University in Tallahassee. Don't go away after the break. Leamers and the weird anti-aging things that happen inside their cells when they hibernate. Yeah, I don't think we've hacked the fountain of youth yet, but I do think that this provides some sort of like tantalizing clues about how longevity can be mapped.

Up next, new research into lemurs, those fluffy, big-eyed primates from Madagascar. Specifically, we're talking about the fat-tailed dwarf lemur, one of the few primates to hibernate. Lemaologists were curious about what's going on inside their cells during hibernation, and they found something really surprising. It turns out that these lemurs, telomeres, those are the ends of their chromosomes, grow longer when they hibernate. typically telomeres shorten as we age, as our cells continue to divide. So why do their telomeres lengthen? What does it mean for how they age?

Have these lemurs tapped into some longevity hack? Here to take us into the lemur borough, our co-authors on this study, who also happen to be married to each other. Dr. Marina Blanco and Dr. Lydia Green, research scientists at Duke University in Durham, North Carolina. Welcome to Science Friday. Thanks very much for having us. Thank you. Lydia, give me the 411.

on this lemur species. What do they look like? Where do they live? What's their deal? So imagine a fluffy, fat-tailed squirrel with forward-facing eyes and human-like hands that could grasp onto tree branches and lots of fruit. And then imagine them in the western forests of Madagascar that are dry deciduous forests, maybe hanging out in a baobab and facing food scarcity during the dry season. And so they decide to curl up inside a tree hole and live off the junk in their trunk, the fat in their tail that's been stored up in the pre-obeyed. preceding season and basically curl up and just past winter by not eating, not foraging, not moving and saving a lot of energy by living off the fat in their tail.

The junk in their trunk, that's where they store all their fat in their tail? Yes, as do I. Is that unusual though? Like, is that why it's in their name? I just want to go one level deeper on the tail, I think. Yeah, I think I would say that there is a good reason to store your fat in your tail because it's a pretty safe place to put it. If you were to put your fat, around your organs or in other places in your body cavity, that might be more dangerous. We know that those fats are not great for humans either. And so putting it in your tail just as a safe place to keep it out of harm's way and you can use it when you need it. And so they can hibernate for four to seven

months a year without eating and also without drinking and without even leaving their tree hole. So they'll spend up to seven months just living off that fat. Marina, tell me about what you found in this recent study. What was the kind of the biggest takeaway? So we knew that telomere elongation was possible in some animals, particularly in hibernators. Because a few studies had been published about 10 years ago where small temperate hibernators would elongate their telomeres after the hibernation season when they had access to rich fruits. And so we knew that hibernators may have a mechanism to elongate telomeres. and we also knew at the time that hibernators tend to live longer lives than similarly size animals.

So we wanted to test if telomere elongation was also possible in dwarf lemurs, which had tropical hibernators. After we were able to sort of facilitate hibernation in dwarf lemurs at the center, we separated in two groups. And so we had the dwarf lemurs that were housed in rooms with cold temperature during hibernation and no food at all so they could live of their fats. And then we had a group of dwarf lemurs that were also housed in cold rooms, but they were given a daily ration of food.

And so the expectation was that the dwarf lemurs in the cold rooms that had access to food probably could be able to elongate their telomeres because they had, in addition to the fats, a store they have extra food. And so if they needed to protect their cells and their bodies during hyperbolems, nation that they could be able to elongate their telomeres. But what we found is that it was actually the dwarf lemurs in the cold rooms with no food, the ones that elongated their telomeres at the end of the hibernation season. And so we were a little surprised by that and sort of concluded that it was the lemurs that were at

higher risk or maybe more vulnerable at the end of the hibernation season that needed to elongate telomeres to protect themselves. What does hibernation look like for them? So during the hibernation season, these animals do depress their metabolism. They go into these long torpor bouts. But these torpor bouts are interrupted by arousals. And arousals are these short periods where limors basically heat up and go from zero to a hundred metabolically. And they stay in this arousal time for about 24 hours.

During this time, you know, heart rates go up, breathing rates go up, they actually sleep. And then after the arousal, they return, they go back down into a torpor bout. What does a torpor bout look like? Like, what does that mean? So basically, your systems are shut down to a minimum, to a sort of a suspended animation state. So during a torpor bout, your brain basically is flatlining in an EEG, your high rates go down to maybe six or eight bits per minute instead of, you know, 300 when you're active. You don't breathe for several minutes, eight to ten minutes.

Wow. You don't breathe for several minutes. No, they have apnea. So they don't breathe. Everything is really shut down. They're called to detach. So it's minimum survival mode. But, you know, you can stay like this for maybe seven to ten days.

and after that torpor bout arousals come. So there is something inherently important about arousing from a torpor bout to sort of reset your systems. And it is during those arousals where their bodies really are at a vulnerable stage, because imagine these metabolic surges that have to happen in very few hours. And I think that's when we start looking at, you know, potential mechanisms of protection that these animals have to protect themselves from these arousals that could be dangerous. And that's why you think the telomeres are lengthening is that when they arouse, like there could be a lot of damage to cells.

And so you would want some extra protection. Is that the idea? Yeah, and I was going to add more context. At some point, you need to re-warm up again to go through these arousals to sort of maintain physiological health. And so during these arousals, they're essentially burning their fats to warm up. And there's a lot of potential damage that can happen to the cells as a result of that intense fat burn. And so there might be a reason why an animal sort of at the end of the hibernation season,

when they're burning through the ends of their fat, may put a last surge to lengthen their telomeres as a way to protect the integrity of the cell at the very end of the season. I mean, Lydia, we know that telomere shortening is associated with aging. Did these lemurs live longer? Yeah, so there's a fascinating correlation between hibernation and longevity. and that mammals that hibernate tend to live longer lives than do similarly sized counterparts that don't hibernate. So in the

counterpart to the dwarf lemurs would be the bush babies that are about the same size primates, but they have much shorter lifespans, what we're in like 12 to 13 years, whereas dwarf lemurs can live to almost 30. And so we've known for a long time that there's this interaction between

longevity and hibernation, but the mechanisms are not well understood, particularly in these dwarf lemurs. And so what we found in these animals is you might expect that the stress of hibernation would shorten telomeres much faster than it would in a non-hybernator. But what we seem to find is that these animals have some ability to lengthen their telomeres to protect themselves. And maybe that is what helps them live longer lives despite the fact that they're going through these stressful periods. Well, that's what I was going to ask. I mean, does this mechanism or understanding this mechanism

tell us anything about longevity or, of course, people are very interested in this? Yeah, I don't think we've, like, hacked the fountain of youth yet. But I do think that this provides some sort of, like, tantalizing clues about how longevity can be maximized. But I think the question for us is still always going back to this mechanism of how are telomere's lengthening. And, you know, we talk a lot about this lengthening in this project, but also two weeks after they came out of hibernation, their telomeres had shortened again. So maybe that stress of those first two weeks of being aroused and of trying to find a mate, because that's the reproductive season. So it's not the thing. So it's not the time.

that their telomeres lengthened and stayed long, they actually ultimately shortened again. And they started off the hibernation season of year two at the same length that they started off the hibernation season of year one. So it's not like they're sort of indefinitely lengthening their telomeres. But I do think that if we continue this work and if colleagues and peers and collaborators continue this work, there's a lot we can learn about the mechanisms underlying telemere dynamics that could downstream be applied to people. Yeah, I'm interested in this. I mean, I know bats can be a kind of poster child of longevity. And I know that in some species, researchers have found their telomeres lengthening during hibernation.

Like, do we need to get all the people who study animal superagers together to work on this mystery? Yeah, and I think we can throw in there people that do deep diving in the bottom of the ocean and people that are traveling to outer space because there's also been some interesting findings that maybe they are able to lengthen their telomeres under sort of the extreme physiological stress of humans at the maximum of their physiological potential. Hmm. That's really fascinating. What have we missed about this study? These are critically endangered primates from Madagascar, and so we're not going to be doing the invasive work to really challenge their cells in ways that would allow us to understand the mechanisms, but we can potentially grow cell lines from these animals and expose the cells in a petri dish to the sort of challenge that the animal would experience. And so we can answer some of these questions without, do we

any invasive work on the animals themselves. But I also think that there's real value in us looking to Madagascar and looking to wild populations of dwarf lemurs to try to understand the variety of conditions and environmental heterogeneity and environmental challenge that they're experiencing in the wild. And to start modeling some of these telomere dynamics just from the same cheek swabs we used applied to these captive animals to understand what are the dynamics going on in the wild for animals hibernating under cold conditions, for animals hibernating under warm conditions, for animals hibernating at the beginning of their lives or hibernating at the end of their lives. And I think there's all sorts of questions we can also ask in the wild.

So I think this dual approach of sort of the natural ecology, natural history perspective in Madagascar coupled with really high-quality, rigorous lab work that doesn't have the animal involved in it is really, I think, the duality of the approach that we see. Okay, well, when you head to Madagascar for your fieldwork, please phone us in. And thank you both for joining me today. Thank you so much for having us. Dr. Marina Blanco and Dr. Lydia Green, research science, at Duke University in Durham, North Carolina.

And that is about all we have time for. Lots of folks helped make the show happen, including Jordan Smudjik. Rasha Auretti. Charles Bergquist. Shoshana Bucksdown. I'm Flora Lichtman. Thanks for listening.