How We Use ChatGPT, Meta’s New AI Glasses, Can Jimmy Kimmel Be Canceled?

0:00New data from OpenAI and Anthropic reveals how people are actually using their chatbots. Meta's new AI glasses are here. Are they the road to super intelligence? Plus, can you actually cancel Jimmy Kimmel? We'll talk about it on a Big Technology Podcast Friday edition right after this. Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional, cool-headed, and nuanced format. We have a great show for you today because we have a boatload of new data from OpenAI and Anthropic looking at all the different uses of chatbots, some data that may settle some debates we've been having on the show for quite some time.

0:35We'll also talk about the rollout of Meta's new AI glasses, some live demo trouble, but we'll talk about where the product is going, and we'll also discuss the canceling or the pausing of Jimmy Kimmel's show, and really whether cancellation has the same meaning these days, given the outlets available online to continue the conversation. It's going to be a great show. Joining us, as always, on Friday, is Ranjan Roy of Margins. Ranjan, great to see you. Good to see you. I can't believe it. OpenAI just lifted your blog post. Three ways people are using generative AI. We're going to get into it, but when I was reading that research, oh man, I was like, did GPT-5 just ingest the big technology blog and come up with, here's how we want to structure this?

1:23I have to be honest. I read that and I was like, that is the three faces of generative AI post just coming out in OpenAI research. So I'm so excited to be able to talk about the data because we finally do have some real data. And let's get right into it. So OpenAI releases this post this week, how people are using ChatGPT. I, of course, was like, well, what's going on with the companionship side of things? And we actually have real data on that as well. But here's directly from the post. It says, ChatGPT consumer usage is largely about getting everyday tasks done. Three quarters of conversations focus on practical guidance, seeking information, and writing, with writing being the most common work task, while coding and self-expression remain niche activities.

2:09Practical guidance, seeking information, and writing are the three most common topics and collectively account for nearly 80 % of all conversations. Practical guidance is the most common use case and includes activities like tutoring and teaching, how-to advice about a variety of topics, and creative ideation. So Ranjan, I just want to turn this over to you. What do you think about the fact that these are the most common use cases of ChatGPT? And I'll just add on another question because why not? When I saw this, I said, you know what? But chat, like chat with generative AI in search, have always felt like completely different things to me.

2:50And maybe this is some evidence for like this bigger overriding question of is Google toast? Why Google is seeing the success it is and why generative AI might not be the direct threat it was portrayed as for so long. I'm going to push back on that. How do you see practical guidance in seeking information is distinct from traditional search? like wouldn't that be the main or at least a very large component of overall traditional search okay good point seeking information definitely is search right that is without a doubt search but the fact that practical guidance is number one like i think there was a a misconception that ai was going to be just like search redone a refactoring of search and it's probably led by publishers who are like, where's our traffic going?

3:38But the fact that practical guidance is number one, where you can go to, and we're going to talk about it more, but you can go to ChatGPT and say, I'm facing this situation. What should I do? Whether that's a work situation, a health situation, a relationship situation, getting fit, which we're going to talk about. That to me is a brand new use case. You just have not had that online at all. No, no, no, no. If you think back, there was, I remember in the like 2000s or early 2010s media startup days, there was entire, there's one howcast where it was just all short form video that would just do how to videos, like how to do basically small elements of practical guidance.

4:22So, so there was, this was a core component, especially like very SEO optimized traffic. Again, every publisher that was trying to like pull in traffic. It was around these how-to questions. So I think I would still push. Yeah. That's exactly the point I'm making here, which is that this has been a, the belief that AI is the natural inheritor to search has been driven in part by people who work in publishing, which is a place that the narratives tend to come out of. Who've said, we've done this before and this is search and this is taking search. The difference is you had to create all that content on Howcast.

5:02This is a technological solution to practical guidance. And that's emergent. That's something that we haven't had previously. It's brand new. And so to me, this idea that it's a technology replacement to search comes second to this idea that this is a brand new technology layer on the practical guidance front. And in fact, I would say, maybe this is controversial, but I think this does a better job than those Howcast and eHow. and no offense to anybody who's worked on those websites before, they are a creation of SEO. They would not exist without Google. And they're not very good. Like I don't think anyone, you know, going, giving their, you know, coming back to a college, giving the commencement address, somebody super successful, you know, says, hey, I want to just thank the people at eHow.

5:48You know, I had all these problems. And I figured out everything on eHow. No, it was just like it could get you maybe two-thirds of the way to fixing your sync. When I look back at all the success I have achieved in my life, I would like to thank eHow, Howcast, Ask Jeeves, and all the above. Yeah, I think... Students, students, you may be asking how, but you should be asking eHow. We're dating ourselves for any Gen Z listeners here, but there was a time that these websites were large and dominant. So, okay. So one thing I'll start to kind of, if we dig in further, practical, the way they start to distinguish these things, I still found a little difficult or problematic.

6:37So again, what is the difference between practical guidance and seeking information? Like there's going to be a lot of blurry lines in there. Writing, I think is fairly, fairly consistent, but to me, they had gone another level deeper. So they had, those is one categorization. But then to me, going back to your blog post, they had three patterns within messages, asking, doing and expressing. About half of messages, 49 percent are asking and then expressing doing is 40 percent expressing is 11 percent. These fit very neatly into our discussions, your blog post thinking, doing and companionship.

7:18um so i think they start it's kind of amazing that they're starting to see the world in the way that we've been talking about this thinking doing companionship or asking doing and expressing the only thing for me i want to start kind of like i want to hear your thoughts on the way they define doing is not how i define it they define doing as getting help in doing some kind of process, like getting help, drafting text or planning or programming. The way I've been discussing this is to me, doing is actually going out and doing it. I had a complaint with Delta about luggage a few weeks ago, and I had ChatGPT write the email and tell me where to go, but it didn't send the email.

8:02So it's almost there, but to me, that is the real doing. But otherwise, I'm glad they're starting to look through this in the same way we are. Well, that's a great point. And I'm going to take it one step further. Right. So these are the tasks that people are coming to generative AI or Chatshift for. 49 % is asking. That would be my thought partner category. 11 % is expressing, which OpenAI says is captures uses that are neither doing, asking nor doing usually involves personal affection or exploration and play. To me, that's companion, obviously. Now, doing should be agent, right? And you're saying that their categorization of doing is actually much broader than you would use for agent.

8:47And I would agree. It seems like some of these things like using the chatbot for drafting text. Let's just go with your definition. If it's not calling a tool, it's just like normal LLM behavior. So I think that what they're doing in this study is expanding their agent category to encompass more than it really should. And I wonder why they're doing that. I think I have an answer. I think what we're seeing in the numbers here is the vast majority of activity within ChatGPT is really thought partner. It is a thought partner tool. And the agentic stuff, of course, is super early. But they didn't want maybe the numbers.

9:27and this is maybe somewhat conspiratorial, but I stand by it. They didn't want the numbers to show how deeply into thought partner they are when they just released a product, GPT-5, not only released, tuned their product to the agent use case. To me, this explains why there's been so much uproar and disappointment and disconnect between the people who've been using the previous models and the people using GPT-5. OpenAI had a thought partner product, and they turned it into an agent product, which is, by their definition, a minority of the uses. And as you look a level deeper, a very small percentage of the uses.

10:13And that to me is, I think it's a problem. And that's why I think these companies are going to have to have clarity over which one they're going for. You cannot really do all three. Maybe you can do it in the same product, but you definitely need a switcher. Yep, okay, and I like this. I don't think that's even conspiratorial. Actually, I will say, I do wonder, because this was presented as a research paper, as a PDF in that Times New Roman font, and it just looks like a research paper, you assume there's no marketing hand overseeing it, Whereas if this was produced as a slick PDF, like from a Google, you would just be like, okay, whatever, this isn't real.

10:54Like, or it's just marketing. And I actually do wonder how that gets like politically divided within an organization like OpenAI, because all of the findings really do help push their business case. Like there's, again, like as you said, giving that looser definition of agentic with doing makes it seem like they have very distributed use cases and they are moving towards this world of agent, but where in reality is we're discussing they're not. They also have nearly half of messages come from users aged 18 to 25. So guess what? We're capturing the younger demographic. They have demographic gaps are shrinking that now women are using it as much as men, which I kind of found interesting.

11:43More than men. That was the lead finding in this study. Yeah, but they were very clear that it was like with using traditionally feminine gendered names. They were like, we don't actually have data on you. Don't worry. But we're using just the - But we're going to tell you exactly how people are using our product down to the minute, like tiny little percentage. Okay. Exactly, exactly. And then they have like, they said emotional companionship are rare. They said only 1.9 % of messages are about relationships and 0.4 % role play. That's the companionship we're talking about. It's one level deeper than companionship.

12:22One asking for help, the other getting into the role play. But overall, they even had geographical lower income countries are starting to use this more. So it's positioning it as this beautiful democratizing force, like bridging the wealth gap between nations. So overall, like the more I was looking at it, I'm like, the greatest form of content marketing you can do is make it look like a research paper. And then suddenly just add so much credibility to it. Yeah, look, I have no doubt that researchers worked on this. But we cannot. It's just one of those things when a company releases research, you have to look at it with not.

13:04I don't think you don't believe anything you read. You just have to like read between the lines a little bit and you can get some good data. Like I'm sure that part of this really does reflect the way that ChatGPT is being used. Like the idea to me that most of it is this thought partner use case, that makes total sense to me. It's totally tracks with what I've been thinking about about the bot. But you do have to say there are certain narratives that they want out there And you're right, if you put it out as this research paper, it does do a better job of advancing those narratives than a thou does protest too much blog post saying only 1.9 % of ChachiBT messages are in the topics of relationships and personal reflection.

13:49And in fact, this study, I initially found out about this study because I think one of our listeners tagged me on Twitter and said, oh, hey, look, only 1.9 % are companions. And I had been sort of beating the drum saying that companion is one of the leading use cases. Maybe I even said it was the number one use case. I might have to revise that after this show. Well, it was an HBR article that seemed to suggest that. And that HBR article was also cited in this OpenAI study. So I think they give some credence to it, even though they found something completely different. But I would say after reading the articles that have come out about OpenAI and listening to the podcasts about OpenAI and how people are building relationships with chat GPT and that's unhealthy.

14:35Let's just say this. If I was a researcher at OpenAI, I would do whatever I could. And maybe this is me getting it completely wrong. I'm open to that. I would do whatever I could to minimize that use case. I would say there's only a small percentage of people that are falling in love with ChatGPT. And by the way, let's just do our building relationships with ChatGPT. And let's just do the back of the envelope math here. Because we have 700 million users of ChatGPT. And so this would say 1.9 % of people have some sort of relationship with them, I think. So that would still leave us with 13.3 million people weekly having a relationship style conversation with ChatGPT.

15:19Obviously, it's not the overriding use case, but that's still a shit ton of people. Oh, actually, if you look at it, it's 18 billion messages per week. So I just did. That's 35 million relationship messages right there per week. can we can we back of the envelope figure out how much it's costing open ai to serve these online boyfriend and girlfriends well actually well they're gonna burn a hundred go ahead in the world it's also funny because like in the world of gpt5 where you could have a simple non-agentic answer that can keep things going instead it's going to do some like multi dimensional multi-layered thinking that's just burning tokens just to say that's a great answer you are so smart.

16:08You know, it is interesting because you do get, I really, I mean, maybe I should just do a test, but I really wonder how this agentic, you know, we talked last week about how GPT five always asks, like, do you want me to do this for you? And like, right. Your friend at the barbecue where it's like, uh, where you said, Hey, are you flirting with, with, uh, chat GPT? And it's like, Hey, I can be more flirty if you want. Like, I do wonder when you get deep into those role plays, like what ChatGPT is actually suggesting as terms of the next step? Someone for research, go out. I just can't. I just, that scares me too much.

16:45Actually, that's where I feel like, yeah, we need that story out there of someone who actually goes down that rabbit hole. Alex, I'm nominating you. I'm not going to say I'm going to do it, but I'm also not going to say I'm not going to do it. If I don't have a story for next week, I may have to desperately start to try to go into a role play with ChatGPT and see what happens. God help us all. There was also another interesting speaking of stats here. Another interesting stat. They say 4.2 % of ChatGPT messages are related to computer programming compared to 33 % of work-related Claude conversations.

17:25It's interesting how they have that qualifier. work-related cloud conversations. But even still, I think what they're saying with this is, hey, we've really been working on coding and we have a lot of opportunity here on the coding front, given how intensely, how intense cloud is used for coding and sort of how it's still emerging as a use case for us. Yeah, I agree that that one jumped out at me. And we have been talking about this for months now. And I think both of us have agreed that like Claude's essentially pivot towards coding as like a core use case in terms of monetization has actually been a seemingly successful one as they've been just like ripping through revenue growth.

18:09But yeah, it felt like as the new Codex product came out, like this is another one like, hey, everybody, we have plenty of opportunity here and clods already a little bit saturated so so it definitely again this one felt like one of those very uh very convenient uh statistics that was that was put out there i i'll say though actually no this one and i wasn't surprised by it either because again i feel everyone i speak with chat gpt is not the default for coding assistance and coding help so this made sense. That's right. Okay. Let's talk a little bit about work versus not work. 30 % of consumer usage is work-related and approximately 70 % is non-work.

18:57That is interesting to me. Yeah, I think, but again, how this stuff gets defined, I think is difficult, but also I think we've said before that they have to win consumer and they position themselves as they're going to win consumer and they are leading in consumer. So I think this also still kind of helps drive that narrative too. And apparently that it's up from 53 to 70 % that is non-work usage. So they're showing that this is growing in importance. Everyday people are using ChatGPT, which I, again, flirting with it while it's telling you how to grill on a Labor Day weekend. Like that's I'm going to call that non-work usage.

19:42So so overall, I think that one seems clear to me. There's there's one last part here, which I think is worth talking about, because they they particularly call out this decision support side of things. And I think it's very interesting. So they say a key way that value is created is through decision support. ChatGPT helps improve judgment and productivity, especially in knowledge intensive jobs. And as people discover these and other benefits, usage deepens with user cohorts increasing their activity over time through improved models and new use case discovery. A couple of things on this. First of all, they are saying that this is a way that chatbot productivity is actually, you can't see it in GDP numbers because it's not like a clear activity.

20:27But they say that this is actually already making a difference in terms of economic impact. the other side of it is the fact that they're highlighting this and practical guidance so prominently in this study just suggests to me that I don't think we have fully grasped the level with which people already in the year 2025, three years, nearly three years after the release of ChatGPT, we trust these things. We do. We trust it for guidance in our personal relationships in our work world, everything from our health to how to write an email, inviting people to a party. For instance, I just wrote an email inviting people to a party, and I just screenshotted it and dropped it in ChatGPT and said, how's my subject line?

21:22And it suggested three different subject lines. I don't have a lot of experience inviting people to parties but chat chapt has a lot of that baked in and i was like you know what your subject line is better than mine copy paste and away it goes and the rcps are flowing in so um just the amount of trust people have in these bots is unbelievable already is that doing is that doing right there or is that thinking well you're the bot's not doing it i mean maybe the bot is i would put that in thought partner yeah so that's where but i would i'm guessing that would have been That would have been categorized as doing in the open AI context.

21:59I also, that is an interesting point around how GDP and traditional economic metrics aren't capturing any of this because they're only going to capture, I guess, improvements in output. But all this, and this happened, I feel like in the early digital days where there's all this discussion around the time spent posting on Facebook or posting on Twitter or all these kind of things actually created very little economic value. And it was more where people were just spending time and energy. And I think that's actually going to be an interesting as people just spend more and more and more time with AI chat, like what that actually means for overall measurement of economic activity because none of it is going to be captured yeah other than your other than your better converting subject lines to get people to the party that'll be captured certainly but that will be captured yes and drink served but um what do you think about this idea that people trust it already i mean for all the it and and is it trustworthy Go ahead.

23:08Okay, two separate questions there. I think, do people trust it? Yes. I mean, undoubtedly, everyone. I think we definitely have crossed the inflection point of all my normie friends using ChatGPT regularly or using some kind of chatbot, Gemini, Claude, whatever, in day-to-day personal life and just asking it questions. Is it trustworthy? I think, I don't know. It's still tough because like hallucination, I still very clearly get hallucinations on factual information if the quality of information around the query is not good. So it's certainly not trustworthy around more niche topics or things where there's kind of like conflicting information out there because it shouldn't be trustworthy.

23:58That's not how an LLM is meant to work. So I think that could be more of a problem going forward. But still right now, everyone's in the kind of honeymoon phase. What about you? That's right. I do think it is very interesting. I can't tell you whether it's good or bad that so many people trust it. I don't know yet. I don't think we have enough data. I have found it to be trustworthy to a certain extent. for instance like in situations where I've been sick I've said give me a you know a day-to-day of where my health is going to be on Monday, Tuesday, Wednesday, Thursday of this week and it's been pretty accurate and it's been able to sort of help me figure out what to do how to plan things you know when to take it easy when not to and that's just one small use case I've found it to be, obviously it hallucinates, but it gets stuff right so often that I found myself trusting it more and more.

25:00And it also goes back to this Matthew Prince thing, the CEO of Cloudflare that came on a couple of weeks ago. And he said, people used to click out to the links. Now they trust the bot so much, they don't even care about the footnotes anymore. I think we're getting to that point. Yeah. And I think also, is the web dead? Is the web in secular decline for any longtime listener will know this is something we've debated regularly. I think just so much of the web actually became less trustworthy or just overly SEO optimized or just unusable that people are actually just excited that there's something that cleanly gives you an answer that's at least seemingly or is fairly accurate.

25:42So people are just excited about that. Yes, but the one tragedy here is that eHow is relied upon less. So I expect us to produce a generation of less capable leaders because of this. Let's bring it back, eHow.com. Maybe they should merge with ChatGPT. By the way, one last thing. The usage numbers, which they included in this report, with some really kind of key data points in terms of how people have used ChatGPT, it is fascinating. So we talked on the show a lot in 2023, 2024 about like, where's the growth of ChatGPT? Is ChatGPT flatlining? And you actually see that getting into mid 2024, it had not reached 200 million users.

26:27after people said it hit 100 million early 2023, which you look at the data and it's not quite clear that it did, at least weekly users. But then around mid-2024, you see a spike. So it crosses 200 million users in around July 2024. It goes up to 300 million in around January 2024. This year, it's on track to hit 800 million weekly users by the end of the year, maybe even more, maybe 900 million. That is astonishing growth. And I have never seen this in any product ever before. And to be honest, not even close. No, I agree. And again, you see it, I think like with a lot of technology, when it really is in front of you day to day, in front of not just early adopters, but even the iPhone, I remember I stood in line for the first iPhone and I was made fun of at my work by people who are like, why are you standing in line for a phone?

27:30It was a solid like five or six years before you saw it just ubiquitous. And this has become a lot, lot faster where it's just fully immersed in everyone, my parents, pop culture. It's just out there. So I'll give, and it's a testament to just how revolutionary the actual products are. That's right. Yeah, it's in South Park. It's in South Park, which was canceled this week. South Park was canceled this week? No, no, no. It was canceled this week? They removed, they did not air their new episode. And they said that they did not have it done, which is just odd. That never happens. Yes, okay. No, I mean, yeah.

28:19Anyways. We'll talk about more media weirdness after the break. And if you enjoy data about the chatbots, we have more coming your way as we get into the numbers that Anthropic shared this week. It seems like it's AI usage data week here on Big Technology Podcast Friday Edition, and we'll do more of it right after this. Did you know your credit card points and miles can lose value to inflation? Credit card companies often reduce the redemption value of your points and miles. Now, imagine a credit card with rewards that can grow in value. With the Gemini credit card, you can earn Bitcoin or one of over 50 other cryptos instantly with no annual fee.

28:58Every swipe at the store or gas pump earns you instant rewards deposited straight to your account. Plus, sign up now for a$200 Bitcoin bonus to kickstart your rewards. Visit Gemini.com slash card today. Check out the link in the description for more information on rates. Again, if you're looking to invest in Bitcoin but don't know where to start, the Gemini credit card makes it easy. The Gemini credit card is issued by WebBank. In order to qualify for the$200 crypto intro bonus, you must spend$3 ,000 in your first 90 days. Some exclusions apply to instant rewards in which rewards are deposited when the transaction posts.

29:36This content is not investment advice and trading crypto involves risk. The Gemini credit card cannot be used to make gambling-related purchases. Shape the future of enterprise AI with Agency. A-G-N-T-C-Y. Now an open source Linux Foundation project, Agency is leading the way in establishing trusted identity and access management for the Internet of Agents. A collaboration layer that ensures AI agents can securely discover, connect, and work across any framework. With Agency, your organization gains open, standardized tools and seamless integration, including robust identity management to be able to identify, authenticate, and interact across any platform.