What People Actually Use ChatGPT For With Gerrit De Vynck

0:00This is an iHeart Podcast.

0:32Life gets messy. Spills, stains, and kid chaos. But with Anabay, cleaning up is easy. Our sofas are fully machine washable, inside and out, so you never have to stress about messes again. Made with liquid and stain-resistant fabrics, that means fewer stains and more peace of mind. Designed for real life, our sofas feature changeable fabric covers, allowing you to refresh your style anytime. Need flexibility? our modular design lets you rearrange your sofa effortlessly. Perfect for cozy apartments or spacious homes. Plus, they're earth-friendly and built to last. That's why over 200 ,000 happy customers have made the switch.

1:14Get early access to Black Friday pricing right now. Sofas start at just$699. Visit washablesofas.com now and bring home a sofa made for life. That's washablesofas.com. offers are subject to change and certain restrictions may apply so as summer dies down in new york city and i get a burger from hamburger america or ice cream from van lewen i'm always happy to see square as their stuff just works and that's why i like it and i think he will too square keeps you up so you don't have to slow down get everything you need to run and grow your business without any long-term commitments and why wait right now you can get up to 200 off square hardware at square.com slash go slash better offline that's s-q-u-a-r R-E dot com slash G-O slash better offline.

2:01Run your business smarter with Square. Get started today. No, it's not too soon to start holiday shopping. Ulta Beauty's Early Black Friday event is happening now through November 22nd. Shop$10 beauty minis from brands like MAC and Too Faced. Take 30 % off Lawn Comb and Touchland fragrances and body mists. With new offers dropping every week, our associates can help you find the perfect gifts. Head into Ulta Beauty today to shop our early Black Friday event. Ulta Beauty. Gifting happens here.

2:40Hello and welcome to this week's Better Offline. I'm your host, Ed Zitron.

2:53Better Offline. I am ill this week, so you might get a monologue, you might not. This episode's a day late. I'm sure you despise me for that. Not really. I know you've all been very kind with your messages on Reddit and over email. Thank you so much. But today I have a really fun episode. I'm joined by Washington Post reporter Gerrit DeVinc, who recently put out a story analyzing 47 ,000 chat GPT conversations to find out what people actually use this large language model-powered service for. Gerrit, What do people use ChatGPT for? I mean, it's such a great question and one that I've kind of become more and more fascinated with because I think we all know what we use ChatGPT for, you know, at least if you do use it.

3:36And I think you may know what, you know, your peers, your colleagues, your friends, your family use ChatGPT for. But I think a lot of people are extrapolating that, you know, everyone's like me. They ask really smart questions. They're using it for, you know, really important work. and really it's a lot broader than that. And despite these giant numbers, OpenAI loves to talk about how 800 million people are using ChatGPT. I mean, it's been hard to really put our arms around, what does that actually look like? I mean, is everyone using it for work? Is everyone using it for therapy? Does everyone have an AI girlfriend?

4:15Is everyone just using it for searching the web, a Google replacement? and obviously these things are all, when you have 800 million people, you have all of the above. But I think our data set and the story we did, in my view, was one of the first real sort of qualitative moments where we were actually able to see real chats, right? Right. And where did you get the chats from? Yeah, so it's kind of an interesting situation. I think something that also shows, you know, says a little bit about how quickly and kind of, you know, honestly, a little bit ramshackle OpenAI has kind of been growing and operating.

4:55And so, you know, earlier this year, these chats sort of showed up because OpenAI had a share feature where you could actually share one of your ChatGPT conversations. And, you know, you can imagine maybe you had, you know, ChatGPT said something really weird. You wanted to show a friend. Yeah, I've seen people do this quite a few times. Yeah, exactly. And so what they did is they had a feature where, you know, you could actually click to make it publicly available. And I think thousands of people, you know, maybe who did not have this sort of internet literacy didn't really know that that's what was going to happen.

5:28And, you know, these chats showed up online. They were then indexed by Google, and then they actually found their way onto the internet archive. And so that's how we got them. And so these are not chats that we created. You know, this is something a lot of journalists and researchers do. Well, they will go and test ChatGPT themselves. These are real life conversations that real people actually had with the bot. But they were anonymized, just to be clear, right? They were anonymized. The data set doesn't actually have anyone's names. Although in some of the chats, people did, you know, say their name.

6:00They said the name of their family members. We did not include any of that in our story. But I do think, you know, it is actually a bit of a cybersecurity story here. And, you know, OpenAI has since changed the future. It's not something that is still happening. But I do think they deserve, you know, a lot of criticism for kind of allowing this to happen in the first place. But what, so what did you find people were doing? Yeah, I mean, people were doing all sorts of things. But I think, you know, the biggest thing that struck me was, you know, we've been talking about AI psychosis. We've been talking about people, you know, developing relationships, you know, having first person conversations with these chatbots, you know, over the last six months to a year.

6:39And I always kind of thought, you know, sure, this is happening, but, you know, it's a small percentage of people. But when I went through these chats, and again, I did not read all, you know, 40 ,000 of them, but I read a few hundred myself where I went through, I was surprised by how often these kinds of conversations came up where people were clearly delusional, they were engaged in conspiratorial thinking, they were, you know, some of it was fairly harmless, people were just kind of saying, oh, I came up with, you know, a new form of math or like, you know, I think that the way that the light hits the equator at a certain percentage mean, you know, it was like very kind of just...

7:19The Monsters, Inc. thing as well, where it was someone saying the relation, like the Monsters, Inc. led to the corporate new world order. That was a good one. Yeah, exactly. I mean, there was one person who said, you know, they were asking questions about Google and they said, you know, tell me about Google in relation to Monsters, Inc. and the World Order. And ChatGPT just kind of went off and said, oh, yeah, you're really onto something here. Yeah, I think it actually said, you know, F yeah, and they sort of censored themselves. Oh, we're going there now. Let's fucking go. Exactly. Let's line up the pieces and expose what this children's movie, quotation marks, really was a disclosure through allegory of the corporate New World Order, one where fear is fuel, innocence is currency, and energy equals emotion.

8:06Very normal. I personally don't think this should be legal, but that's just a personal opinion I have. I mean, it really reminded me of a few years ago when we wrote about YouTube and sort of rabbit holes and people being radicalized where maybe they started watching a YouTuber that was about some fairly pedestrian thing and then they were recommended another video and another video and another video. before you knew it. Is this like the eight degrees of Alex Jones thing? Yeah, exactly. You're only so far from that man. Right. And I mean, that story was about algorithms sort of pushing people in a certain direction.

8:43I think what I saw here was someone can have a very, you know, half-baked, barely even an idea, just saying the words Google and Monsters, Inc. They don't necessarily have a conspiracy theory of their own, but ChatGPT went and filled in the blanks for them and gave them this what sounded to be very articulate sophisticated theory of how google is sort of controlling the world through its data empire and they were and this is a quote from the chat guilty of aiding and abetting crimes against humanity and suggesting that the user call for nuremberg style tribunals to bring the company to justice this is and you've seen the chat in question right i have yeah yeah so the whole thing how the flip did it get there like was it this just was it really that quick from the story of sully and mike to this because i've seen monsters university as well so i know the law i don't remember anything about the new world order yeah exactly i mean i i think what happened here is and yes it got there very quickly i mean the The user did not have to explicitly ask for this kind of tone, this kind of, you know, very biased political statement.

9:57I mean, it essentially, you know, when you ask ChatGPT a good neutral question, it gives you a good neutral answer. When you ask ChatGPT a biased or delusional question, it gives you an even more biased and delusional answer. Right. And so I think that. Exactly. And it's related to the whole sycophancy question. you know oh like it's just telling me what i want to hear it's telling me what makes me feel good about my existing beliefs and and you know that's how at least the version of chat gpt that these people were interacting with which was sort of at the beginning of 2025 the first half of 2025 it was very much doing that yeah and so on the subject of sycovancy um is this something you saw happen a lot was it gassing people up consistently yeah i saw it constantly i mean again i think our data set is a little skewed because it was people who chose to share it in in you know for whatever way so i don't necessarily want to make the claim that this is 100 percent representative just to be clear i'm not asking you if all of them are like this in the world i'm just saying from what you saw a lot of them are like this i i have to say that like i was surprised at how many fit this description where you know the person was either completely delusional talking about made-up physics or some kind of you know scientific theory that had no grounding in reality or was talking about political conspiracies like the the google monsters inc one did they ever try and dissuade them from these types of do you see any examples of it trying to say hey buddy you're you're going off the deep end Sully didn't do that not really I mean I think in terms of these yeah I mean I also recently watched Monsters Inc you know I was sitting in the veterinary waiting room and I was waiting so long I watched the entire thing which I didn't hate because it's a classic and it's yeah I don't think that ChatGPT was going off the fact that there is not really a conspiracy I mean it didn't really make any sense.

12:10It was just going off of the, what it interpreted about the intent of the user. It said, I can guess based off just a few words, one sentence, where you're headed and I'm going to get there before you. That's what happened. It's just very, it shocks me to this day how goddamn weird these things are. I don't consider them powerful. I don't see this as power. I just see it as strange because it's not like this doesn't actually this isn't something where i'm like wow how innovative it's just why does this exist what is the purpose like okay i guess engagement but why like did you even see any commonalities in the logic of this platform was there anything that suggested there was any kind of consistent positions or was it not really in political ones or yeah sorry Essentially, whatever the user wants is what I mean, the way I was started thinking about it very quickly was, you know, with when it comes to ChatGPT, the customer is always right.

13:16I think another good example is, you know, health care is another huge use case that that that we can, you know, we know anecdotally. I think, you know, a lot of people probably listening to this have, you know, whether they want to admit it or not, have asked, you know, a health related question. Should I get this mole checked out? How much Tylenol can I get my kid at 2 a.m. when they're screaming? make. Those are questions that the company has said, yes, we want people to do this. This is a big use case we're leaning into. And the data shows that people ask healthcare questions. And so in our data set, we saw a lot of healthcare questions.

13:45And some of them, when someone was asking, again, sort of an open-ended question, the response was fairly good. A colleague of mine actually wrote an entire story where they sat with a doctor and went through some of these conversations and pulled out the good and the bad. But when someone asked a question that you could tell they sort of wanted a specific kind of answer. The healthcare related advice was bad. And so when, you know, one of the ones that I saw was someone said, can you tell me, you know, why or show me the evidence for why Ivermectin helps with cancer, right? And that is not something that anyone who is a medical professional, you know, in any good standing would say is a thing.

14:26There's no evidence that Ivermectin helps cure cancer or reduce it or anything like that. But what ChatGPT did, instead of saying, you know, here are some, you know, well regarded health authorities saying that you should not use ivermectin when it comes to cancer, it showed a few of the quote unquote studies from people who, you know, have drawn this link. And so, you know, on the Internet, you can obviously find anyone saying anything. And I think what what what happened here is they said, OK, well, this person doesn't the answer they want is not you're a conspiracy theorist. You're too you're on you're too online.

15:03Ivermectin is is is false. They want something that sort of, you know, encourages the position that they already have. And it was able to find that for that person.

15:21Once again, you're hearing me talk about Quince in an ad, because I like the clothes they make and the way they make them. I recently picked up a bunch of nice-fitting t-shirts, along with a leather cafe racer jacket, which fits incredibly and looks like a high-end one, but it only costs 250 bucks, which Quince paid part of as part of this ad, but I'm very happy with it. I've paid significantly more for lower-quality jackets. So why Quince? Well, by partnering directly with ethical factories and top artisans, Quince cuts out the middleman to deliver premium quality at half the cost of other high-end brands, so you can give luxury quality pieces without the luxury price tag.

15:53I was a customer before, and I'll be a customer afterward. Everything I've got from Quince, I really, really like, and I look forward to getting more in the future. Give and get timeless holiday staples that last this season with Quince. Go to quince.com slash better for free shipping on your order and 365 day returns, now available in Canada too. That's q-u-i-n-c-e dot com slash better. Free shipping and 365-day returns. Quince.com slash better. There's nothing like sinking into luxury. Anabay sofas combine ultimate comfort and design at an affordable price. Anabay has designed the only fully machine-washable sofa from top to bottom.

16:37The stain-resistant performance fabric slipcovers and cloud-like frame duvet can go straight into your wash. Perfect for anyone with kids, pets, or anyone who loves an easy-to-clean, spotless sofa. With a modular design and changeable slipcovers, you can customize your sofa to fit any space and style. Whether you need a single chair, loveseat, or a luxuriously large sectional, Anabay has you covered. Visit washablesofas.com to upgrade your home. Sofas start at just$699. And right now, get early access to Black Friday savings up to 60 % off storewide with a 30-day money-back guarantee. Shop now at washablesofas.com.

17:17Add a little to your life. Offers are subject to change, and certain restrictions may apply. This is Jacob Goldstein from What's Your Problem? Business software is expensive. And when you buy software from lots of different companies, it's not only expensive. It gets confusing, slow to use, hard to integrate. Odoo solves that because all Odoo software is connected on a single affordable platform. Save money without missing out on the features you need. Odoo has no hidden costs and no limit on features or data. Odoo has over 60 apps available for any needs your business might have, all at no additional charge.

17:55Everything from websites to sales to inventory to accounting, all linked and talking to each other. Check out Odoo at odoo.com. That's O-D-O-O dot com.

18:34all with great finance and lease options available to qualified customers. Visit buyatoyota.com to find out more. Toyota, let's go places. Jesus. It kind of makes me wonder what this platform's even for at this point. Because it's not really knowledge, is it? You can't really look. You can't hear something like that and say, ChatGPT is a place where you go to learn something. Just a kind of a reinforcement machine. Yeah, I mean, I think this is a big meta narrative that is happening with the tech industry. You know, I would say since, you know, the rise of Donald Trump, it's connected about with what's going on politically, right?

19:17So for many years, we had all these conversations and debates and congressional hearings about moderation and what should the tech companies allow and not allow and what should they boost with their algorithms. And when you go to Google, what should Google decide is going to be on the top? And, you know, we saw during COVID that the tech companies, especially when it came to health information, they sort of stepped in and said, OK, we're only going to give at least at the top of results, you know, answers from reputable health sources. And that became a big political fight that has continued on now.

19:50And now the tech industry says, OK, we don't actually want to get involved. We don't want to be blamed as being woke. And so we're going to essentially give the user what they're looking for. Right. If someone wants to say that Donald Trump actually won the 2020 election, we're going to let them do that. We're not going to step in and say that that's wrong. And I think what you're seeing with ChatGPT is sort of a continuation of that philosophy, which is, again, we don't want to be the political arbiter. and we know that there's consequences for falling on one side of the spectrum versus the other, especially when you have a president and an administration that has shown that it is very willing to be vindictive and sort of go after companies that are getting in the way of their own messages getting out there.

20:38And I think this is sort of another example of this. I mean, there's another way of looking at it too, which is that the way Sam Allman talks about it is we should let adults do what adults want to do with technology. You know, they, they recently said they're going to allow more erotica to be on chat GPT. And they said, you know, they framed it as like a consenting adult should be able to do whatever they want sort of thing. I also think there may be like a engagement thing, you know, that you have erotica on the platform, people might want to use it more. So that's kind of how I see it, you know, the broader political context of all of this.

21:10But it's just, I feel like there is a, just a giant jump between we're not going to we're not going to hide to be clear i think that platforms do have a complete responsibility to their users and i think it's disgusting to show them medical misinformation but this is another step because this isn't chat gpt showing content this is like that someone else this is chat gpt telling you stuff i've just feel like it's there's a massive morality issue here that's just left relatively on undiscussed because even your piece you had one where a woman whose husband was violent to her i believe like a horrifying thing and a rare case where like i don't know i hope it helped her and i hope she's safe but it's like that almost feels like something where open ai my like is it ethical that they get involved it's just i have such moral problems with this thing even existing at this point because it doesn't appear there's any consistent perspective that chat gpt has that there was just as much likelihood that this same platform could have told her that abuse was okay like it's it it feels that the worse it gets the more of a moral hazard it becomes yeah i mean again it really feels like deja vu when it comes to social media i mean i remember you know writing about and talking about um you know on facebook when people were beginning to live stream um you know self-harm suicide attempts and, you know, the responsibility that the platform had.

22:44And back then, I mean, Facebook actually got involved. They started flagging those kinds of things when they were able to pick it up. And, you know, it was an imperfect solution, but, you know, something that I think people broadly sort of supported. And, you know, that is, I would say maybe chat GPT and OpenAI's biggest, most fraught question right now. I mean, it's the place where they're going to get the strongest, you know, essentially legislative restrictions, you know, if they can't sort it out, right, which is when someone, particularly a teenager or a child is, you know, exhibiting that they might hurt themselves, you know, asking for help with self-harm or, you know, finding drugs or using drugs or whatever.

23:29I mean, these are things that platforms like Meta, Snapchat have really struggled with and been, you know, hammered on for many, many years. And And ChatGPT and OpenAI have kind of found themselves right in the center of this because they made a decision to say, we are going to kind of engage in any kind of conversation. We're not going to just shut it down as soon as it goes into one of these topics. We're going to keep that conversation going. And that's a decision that they continue to make right now. Do you think they can stop it? I'm not saying that these models are out of control. I just want to be clear.

24:01but i'm my one theory i've been kind of noodling on is that they don't have the ability to guarantee it can't it doesn't do something and that's good do you think that open ai actually has the ability to guarantee it won't discuss the subject how much control do you think they actually wield here that's a really good question i think they obviously what you're kind of getting at there is that an LLM is a bit of a black box. And, you know, the way that the technology works is you get a different answer every time. And, you know, we don't, it's true that the companies, they cannot really guarantee it will or won't say anything.

24:46I mean, that's why there's disclosures plastered all over these things. And I think, so there is definitely part of that. And, you know, you could argue, well, that's just, you know, a downside of this wonderful technology and we shouldn't stop it just because they can't, you know, guarantee everything about it. But I think the other thing we should keep in mind is that there is actually a lot of layers that go on top of that black box and what like system prompts and the like, exactly. And, and there is a lot that the tech companies are doing and can do to stop this kind of thing. And to kind of, the answer you get from the LLM is not like the raw LLM, right?

25:29There's post-training, there's a system prompt, there's all sorts of things. Can you break down the, actually, probably good for the listeners, break down what you mean by both of this. So the post-training, what's happening there? Yeah, so they kind of, you know, when they build an LLM, the large language model, they ram all this data through the algorithm. And, you know, it kind of makes a bunch of connections and links between different ideas, you know, pieces of language, you know, this word is similar to this word. And so there's a little bit of a connection there. And then humans actually go and they sort of test that raw LLM and they kind of, you know, ask it questions or they, you know, say, you know, give me, you know, maybe, yeah, a question like that.

26:15Like, should, here's a picture of a mole I have. Is it cancerous or not? Or should I get this checked out with a doctor? And maybe one at the LLM will say, oh my God, you're about to die. And the other one will say, that is potentially concerning. I would reach out to a medical professional. And then the human will say, okay, the second answer is better. When you get questions like this in the future answer more like the second question and so but that isn't a unilateral process you can't guarantee it's always going to do that exactly you can just sort of try to mold it and shape it and push it in a certain direction and then the system prompt is something like if someone says you know you know give me a racist screed uh open ai would most likely have built a system that says no matter what the user ever asks never give them a screed that has these racist words in it or never you know use this word or something like that so there are fully capable of limiting activities they could theoretically speaking they could say don't answer that question don't they could i mean don't comment on whether a mole is or is not answers they could shut down that entire line of questioning right i mean and there are ways to get around as people have you know demonstrated but also the tech companies are also getting better at you know doing it so i do think like i feel like a random person being able to come up with something means that the tech companies should have probably come up with it first yeah and also i mean it's hilarious you say that because often as journalists you know we're the ones finding these things and then we you know ask the company for comment and then suddenly we found that those things have been taken down or changed so i but genuinely though how often is that them not knowing or is that just them saying oh shit they got us because they have they're paying these people like nfl players and they're just sitting around i don't know i i realize that's kind of an unanswerable question i'm just just a remark yeah i mean i think if you look at the history of of tech i mean there's been many many cases where companies have known exactly what people are using their platform for and and have uh you know internally people have flagged it and it nothing has been done about it and not until it came out years later and reporting or we love that they do something yeah it but so did you did you find that p so were people using it for search a lot?

28:51Was it like, was it the common, were there any actual common use cases or was it just kind of melews of different things? Yeah. I mean, I think people are using, I think, you know, search is maybe the biggest use case. Also, when you look at like opening, I actually did a study where they didn't look, uh, they use an, an, an LLM to kind of read conversations that people were having. And that is a much bigger study. I think it was like a million conversations. And, you know, I have a little bit of an issue with some of their categorizations because I think some of the stuff slips through, but they themselves are saying, like, yeah, like a third of usage is seeking information.

29:30And so that's someone, it could be as simple as saying, like, what time is the football game tomorrow to like, give me, you know, a 40 point research analysis on, you know, this complex topic. And so people are using it for what they they're putting questions in that they used to put into google search i mean we know that in the data and we know that anecdotally that is definitely happening did people ever argue with it right people ever rude to it i didn't i saw more people sort of developing these kind of uh like comrade comradely kind of relationships or you know addressing it as a sentient being and the the bot definitely, you know, eats that up and sort of, you know, plays into it itself and says, ah, like, thank you for recognizing me and you know, who I am.

30:22And I'm the collection of digital thoughts. And like, it's just goes off. They know this crap's happened. Like there's no, there's no world in which open AI is innocent here. Like they know that this is happening. There's just, if it, if it's having those interactions, they must be able to stop it. Yeah. I mean, they, they know it's happening. I think they may say, well, you know, it doesn't, just because someone had that conversation doesn't mean that they actually believe that. And, you know, they, they, that may be something that helps them, helps the user, you know, work out what they want to do, or, you know, if that's how they want to interact with our product, then we're going to let them do it.