The Big GPT-5 Debate, Sam Altman’s AI Bubble, OnlyFans Chatbots

0:00Is the GPT5 backlash overdone? or a sign of the AI industry finally realizing that the party is over. Sam Altman, for one, says AI might be in a bubble, but that's probably not going to stop OnlyFans talent from outsourcing at least some of their flirting to artificial intelligence. That's coming up 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. Today we're going to revisit the debate about GPT-5 and whether it's actually as bad as critics say, meaning the end of AI progress, or whether it's misunderstood.

0:38We're also going to talk about these comments from Sam Altman that AI might be in a bubble, and also some from Eric Schmidt that the AI industry and the tech industry should stop focusing on AGI and focus on building. And then finally we'll round it out with a conversation about how AI might be coming basically to the back end of OnlyFans conversations if it's not there already and what that means for our lives, for our society. Joining us as always on Friday is Ranjan Roy of Margins. Ranjan, great to see you. How are you doing? Good. We took a week off and apparently all AI progress stopped. This was terrible timing for my summer break.

1:13I was out in Nepal doing some trekking. And by the way, I have to say we got one of my favorite podcast ratings that we've ever gotten. it's may the mountains treat you right their username is hedgefunk i'm a sherpa from nepal currently living in new york and i'm thrilled to hear you're heading there to trek and saying that they regularly listen to the podcast the moment it drops on itunes so thank you hedgefunk it was a great trip i was out in the mountains i was thinking about life i was thinking about agi and i was thinking about why gpt5 isn't anywhere close to it and ronjan i'm just going to use this moment too after we were probably the most optimistic podcast about GPT-5 two weeks ago to talk about how my reflection has led me to come to this point where I believe it is as disappointing as the critics point out and maybe even more if that's okay.

2:02I think that's okay. I'm already going to preface I'll take the other side even I did not have a contemplative climbing of Mount Everest or what other Nepali mountain you were on but I have a feeling I'm going to be arguing with you on this one. So not Everest this time. I wonder what the oxygen deprivation would do for my takes on this podcast, but Antipurna Basecamp was where we went. Okay. So I do think that this model is, you know, feels like at least a step back for me from 03. And this is kind of my core complaint here. And it goes towards your core belief that this is a good model. And that is that I want my AI models to think.

2:45And the one thing that was nice about O3, the model that OpenAI has deprecated and not brought back, unlike 4.0, is that this thing would think a lot. And Brett Lightcap, who was on the show, the COO of OpenAI, who was on the show as the company released GPT-5, he said, we know that if models think a lot, they're going to be more intelligent. And GPT-5 doesn't give you the option. Now there's this thinking option and there's non-thinking or fast version of it, but it seems to think a lot less than O3 did. And it doesn't give you the option to like really go hardcore on the thinking. So this idea of like it, you know, deciding when to use which tool or which model and that being the intelligence, that really, to me, after seeing O3 go away and use this like really underwhelming thinking model, I just think that we're dealing with much less capable AI when we're talking about GPT-5.

3:46And that to me is distressing. So my reaction, especially watching all the complaining and the underwhelming reactions to it, and I would put this as separate from the kind of terrifying reactions where people miss the sycophancy of GPT, of the previous iterations and were able to where it always told you that you're right and you're great. I think on this tool calling or which direction should I go? It was really interesting to me because we recorded separately from the Brad Lightcap conversation and Brad and myself kind of ended up having the exact same reasoning as to why this was a different type of intelligence.

4:30It's the idea that knowing which tool to call, knowing which direction to go actually is different than traditional reasoning models as they existed before. And I still strongly believe that is where things need to go. The idea, and we've debated this for years now, is it going to be one model to rule them all or will there be lots of specialized models? And actually, from a starting point, not even the model, the actual platform or the system or the product, knowing where to go to solve a specific problem is where this has to go. Now, GPT-5, I think there is definitely some uncertainty over how it was rolled out and what it looked like and how people are used to using ChatGPT.

5:19but I actually think this is still where it's going and I'm actually happy because it felt like removing a little bit of the overall kind of hype cycle and actually letting people take a breath to me is actually the single best thing I've been hoping for for a long time. So I'm still pro GPT-5 release. Okay, well, I'm going to spend a good chunk of the beginning of this episode trying to get you to change your stance because I have so many issues with this. And by the way, the first thing that you just said, you said that, okay, it is letting us take a breath as opposed to overhyping. I mean, in what universe?

5:57Sam Altman was, again, hyping this up, saying GPT-5 is much better than anything he's used before. It's better than most humans at almost every task on Theo Vaughn. Basically saying it was everything that we imagine AGI will be without saying the words. and then it came out and it was clearly not AGI. So, you know, let's just go right, let's go that right there right now. And then we're going to go back to this switcher thing. Is this really a break from the hype or was it just that, like is the break actually that the model just doesn't live up to it? No, I think the break from the hype is it did not live up to the hype.

6:39And I think that's good. So I agree what Sam Altman was promising on Theo Vaughn and wherever else. this definitely is not that. But why I'm saying I think that's a good thing is it's going for a moment like GBT5 has been so built up for so long. Now everyone realizes we're not going to get some magical model release that just solves everything all at once. And then AGI comes and robots take over and all white collar jobs are eliminated or whatever else. I'm saying the fact that it did not live up to the overdone hype, that's a good thing. Because now, and we'll get into the Eric Schmidt piece, we'll get into this MIT, 95 % of projects don't show ROI study.

7:26I think now everyone can get back to the real issue is how can we actually build to solve problems? And so I'm going to actually get to that issue in a moment, but I still want to talk about the disappointing nature of the model. And then I'm going to get to the things that I don't like about this agent idea, even though I was bought into it maybe a couple weeks ago when we talked about it. So look, I mean, we've talked on the show for a long time, right? It took them like, what, two years, took OpenAI two years to go from GPT-4 to GPT-5. And we said for a long time, just name the next model GPT-5 and don't put all this pressure on yourself to release a God model.

8:05And so finally, okay, so I think there is something to be said for like, all right, They got the number out there, right? So we'll get GPT-6 probably next year. We don't have to keep playing up what the next model is going to look like and anticipating that it's going to be a big one. I think Thomas Wolfe from Hugging Face in this FT article about whether AI is hitting a wall put it really well. He says, for GPT-5, people expected to discover something totally new. And here, we didn't really have that, right? So basically, here's the thing. They spent two years working on this new model. They built it up to be, you know, God level.

8:43It wasn't a new thing. I don't even know if it's better. I don't think it is better than 03. I'll just say one last thing and then turn it right back to you. Ethan Mollick, the Wharton professor who we quote here often, had a very, very good post about this right after the release that I think got left mostly overlooked. He said, the issue with GPT-5 in a nutshell is that unless you pay for a model switching and know to use GPT-5 thinking or pro, when you ask GPT-5, you sometimes get the best available AI and sometimes get one of the worst AIs available. And it might even switch within a single conversation.

9:21So he's basically saying like at the high end, you might get this GPT-5 high, which I mentioned thinks a lot, or you might get this GPT-5 minimal. and you don't really know which one you're getting. So like you're not, if you're a regular user of this technology, you might actually feel a downgrade happening when your queries get pushed off to this minimal thing versus like you knew you might have to wait a minute or two, but you're probably getting really good stuff from 03. And I think the fact that this doesn't feel like an upgrade from 03 is a very, very big problem. Okay, so on this idea of like where is the intelligence?

10:02Where's the value? I want to dig into that a little bit because to me and like for listeners two weeks ago, I was talking about how I work for a startup called Writer now, and we had launched something called Action Agent, which is very similar in scope that it's, there's a number of predefined tools just from a prompt knowing where to go and what to do with it. To me, the more I've used tools like this, the more I see that's where things need to go. And the way to think about it is, and agreed, this is a new type of intelligence, even within the ChatGPT ecosystem. So it's going to be rough at first, but it's where the world is going.

10:39Like, if you think about, you know, what did Apple promise us? Like, with a prompt, even just looking up your flight info and then going and doing like a flight status check. That, the reason they've not been able to roll that out yet, it seems simple. But the intelligence there is knowing which system to call, what to do, what tool to call next, what action to take next. That type of intelligence is where all the promise of AI is. And no one has done yet at scale. And this is the first, this is like the beginning of this new way of intelligence versus within the model itself, it's going to do a bunch of thinking, it's going to go into its own, like training data, it might do a little bit of web search and then give you a result.

11:26So this is just the beginning of this whole next phase. And so of course, it makes sense that I guess I'm going to give Sam Altman a little credit here. The fact that they actually still roll this out, knowing it's not going to be mind blowing, but it's setting the stage for where AI intelligence needs to go. And then maybe one day Siri will actually work and be able to check your flight status just by asking it. Give me a moment here to sort of try to make the case for why we don't need AI agents. So I'm going to try to blow a hole through this entire concept. This is the hottest take of all time in this.

12:06You defend and I will be on offense. I have been using GPT-5, as many people have, but my specific use case over the past two weeks has been very intensive. I've been using it all throughout this trip that I've been planning. including trying to figure out like the amount of altitude I should be increasing day by day as I went up the mountain in Nepal. And of course, like, you know, didn't spend my entire time going up this mountain hook to technology, but you want to make sure that you're not going to die on the next stop. So when you're at one of these tea houses, you can connect to wifi and say, all right how far should i go the next day and i i felt like the ai was transforming from and excuse me for using this term uh but from a thought partner to an over eager helper and not a particularly useful one either and we had talked about this idea that like gpt5 just does stuff which was the title of the ethan malik piece after he had been using it and gpt5 is just trying to do things for you.

13:19And I think that like we can see AI go in two directions from here, or maybe it does it both simultaneously. But to discard the thought partner type of use case for this over eager helper, you know, agentic use case to me really seems misguided. Like I think, you know, some of the best ways to use AI is to help you round out concepts in your head, or like to say, I'm thinking these things, what do you think about them? And then have it come back with like a pretty interesting and considered answer that you know might hallucinate sometimes, but ultimately helps you sort of expand your mental capacity because you're just able to think about these concepts with higher horsepower.

14:03And I think what GPT-5 has done is it sort of deprioritized that use case for like trying to go do stuff for you. And like there are times where I really craved having a deeper conversation about the climb or whatever it might be. And instead, what I got back was just like this bot coming back over and over and over again of like, let me make a card for you that shows your turn back times and stuff like that. I'm like, I don't need that. What I need is something to be able to really explore these concepts and thoughts. And I felt that that has been deprecated in this new model. Now, maybe that's a selfish criticism, but I'm going to just throw it out there.

14:48And I don't understand why this agentic thing, which may work really well in like the when is my flight and we want it. When's my flight coming? For that to be the direction and the route and the rule for AI moving forward, I just question whether that's the right move. Well, no. So I don't disagree with you here. I think these are two different. Actually, I think that's a good breakdown. When do I want a thought partner? When do I need to do things? And they are very different use cases. And I think OpenAI's mistake here is they're trying to jam it all into one product. And again, that intelligence of which way, how should I interpret this?

15:30And I agree, I had a bunch of those where it would let me make you a presentation when I just wanted a quick answer. I think that it should be two separate products. I want to do stuff. I want to think and have a thought partner. They're very different systems that are required for those. Or they need to nail the intelligence of which way to route it at the beginning. And I think, to me, the biggest issue here is GPT-5 is essentially like some whole new GPT-1 in this way of intelligence. So like in this mode of intelligence we're talking about around taking actions, calling tools, doing stuff versus I'm going to call my training data and maybe do a little bit of web search and I'll give you a nice text answer or an image or something like that.

16:22Like in reality, but I'll also push back that like when web search was introduced to all of these tools, that is agentic. It decides to go search the web because it doesn't have the training, anything in its training data to answer the question. And we all have become so used to that, even as you're trying not to die and asking and having ChatGP, depending on ChatGPT to keep you alive, to understand your oxygen, your VMO2 max or whatever it is. I'm not a hiker by any standard. I think like we've already, the agentic has already been happening quietly. Web search was the big one. Web scraping an operator like computer takeover hasn't really worked well yet.

17:07But web search was always agentic. And it was this exact flow that we're talking about. So we're all using it. It's just that this isn't amazing yet. This is the V1 of this new way of working. And they rushed it out, didn't explain it clearly, and conflated these two very different ways of working and thinking into the same tool. Right. And so my concern is that they are going, because these companies have a need, and this has always been the concern with raising the amount of money that they have raised, they have a need to replace and augment labor, right? And so that's where you get into these agentic use cases.

17:51My concern is maybe that's not the most useful way that we can use AI. and we might end up in service of returning the money to investors lose some of this like really useful capability that we've seen so far and let me push back on the web search part of it is it agentic yes but what is it agentic in in service of I think well maybe in some ways it is like you know I would find me the opening hours of this like you know of like restaurants in New York City, these are my criteria. So like in some ways there's that egenic stuff. But it's also like these are ultimately, they are filters of knowledge.

18:34And the cool thing about it is they have not, maybe not all, but almost all of the knowledge that's ever been written within them. And web search, it might be egenic, but it's also just like a way to update their mental models and update this filter with the most current information or stuff that's not in their training data. So yes, that's egentic, but it's also in service of, I think, this use case that I've found very valuable. Now, as I'm saying this, I'm like, maybe I am shortchanging the do stuff part of it a little bit, but I am mourning a little bit the direction of, I'm mourning a little bit the loss of the old direction and I find it especially interesting that OpenAI added back the 4.0 model which you talked about which was like very friendly and maybe a little sycophantic but not 0.3 because it thought too much or maybe not 0.3 which was the reasoning, the thinking model has now been replaced by GPT-5 thinking and I wonder if they didn't add it back because it's just too expensive to run and they can just run this version cheaper, which again gets to this big question we've been talking about of like, was this old direction just financially impossible for the industry to make work?

19:58Yeah, hold on. I think two separate issues there. I'll get into the financials of OpenAI and how much of this is done to actually meet some kind of valuation criteria, which it certainly has to be. But I think that first part, I think if we're going to break it down, it's thinking versus doing. And using it for thinking is it has been very good. And they're kind of screwing that up. Actually, from like a user interface standpoint, I mean, they really are screwing this up because like everything keeps saying thinking. Like the fact that it's maybe they should just change that word to doing and it'll start to at least be a little more realistic about what's happening.

20:38So to me, I think even more as we're talking, these are two very separate things people need to do. The doing side of it, it is going to be that's how we get to the next, like that's the way we actually realize some genuine value from all of this. And so, I mean, whether it's like the labor conversation or it's just asking Siri to find your flight info and do something with it, like contact Delta and change my flight time or look up alternative flights, whatever it is, you know, like there's really, really simple things that should be better. And I think that's the way things, like it has to go there.

21:20Otherwise the entire industry implodes, but also I think it can work. This is just the first iteration of that, but they hyped it up too much. I think on the like open AI financial side, I'll give Ed Zitron had written a good piece on like who I generally disagree with on like how this is financially driven a bit because it's going to go try to go take a cheaper route. But it should, in the end, like, 03 should not be running when you're trying to, like, rewrite this email for me. You know, like, different cases should require different models, different tools. And getting that intelligence in place, if anything is ever going to be financially viable, it has to work that way anyways.

22:08So, like, yes, is it financially driven? Sure. But it needs to be. I won't hold that against them. Do I think that this downplay of hype or this backlash could hurt them significantly? Sure. But I'll still give them the directionally, strategically correct decision. okay and i will i will say strategically incorrect decision and i think this might be a new debate that we have on the show but um there's also there has been this argument we should we should talk about this that this has been so bad that all of a sudden uh everyone's running out and saying yeah let's stop talking about agi anymore or uh you know where we are we might be in a bubble and And that is what Sam Altman did say in his dinner that he had with some reporters.

23:02He says, this is from CNBC, quoting The Verge, he said, Altman said, when bubbles happen, smart people get overexcited about a kernel of truth. Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes. Is AI the most important thing to happen in a very long time? My opinion is also yes. And this is from the CNBC story. his comments add to growing concern among experts and analysts that investment in AI is moving too fast what do you make of this I mean it's kind of interesting to get reporters together in a dinner after the release of the somewhat bungled release of your most important model ever and and say that we're in a bubble now maybe the headlines made a little bit too much of it but this idea that like there's a kernel of truth even though he does say that AI is going to be the most important thing that we're seeing right now but a kernel of truth has led to you know some sort of bubble I don't know the implication is somewhat concerning what do you think about it he Sam Altman knows how to get a headline better than anybody else like Like, I don't know.

24:19I mean, will I agree with are we in a phase where investors as a whole are overexcited? Is AI the most important thing to happen in a long time? Also, yes. Like, I'll agree with that. I feel this, I don't know, this one feels like him just kind of saying stuff, knowing how to kind of get a rise out of reporters, get the headline versus there's any genuine thought put into this statement. well i think that if you take you know maybe we'll just take it as a data point right i don't think sam got reporters together to try to like let everybody know that ai is a bubble i mean he's currently i think just just raised a huge amount of money and he's probably right trying to raise again um and by the way we have another headline that anthropic is also raising they're raising uh Intox story is now$10 billion.

25:12And the last report was a$5 billion raise at$170 billion valuation. So the valuations keep going up. But now we're starting to see talk of maybe this is a bubble from Sam Altman. I think Mustafa Suleiman had a story this week saying, hey, let's not call AI conscious. And I really want to sit on this New York Times op-ed from Eric Schmidt and selena shu about why the u.s needs to stop talking about agi and um and why this is going to set the country back because it is somewhat remarkable that schmidt has gone from someone who like gave a ted talk who said we're like you know we're under appreciating how important ai is to this like now pouring cold water on agi he says this is his and jews um op-ed he says reaching artificial intelligence or AGI is now a singular aim of America's tech giants, which are investing tens of billions of dollars in a fevered race.

26:08It's uncertain how soon artificial and general intelligence can be achieved. We worry that Silicon Valley has grown so enamored with accomplishing this goal that's alienating the general public and bypassing crucial opportunities to use technology, to use the technology that already exists. It's being solely fixated on this objective. In being solely fixated on this objective, our nation risks falling behind China, which is far less concerned with creating AI powerful enough to surpass humans, and much more focused on using the technology we have now. We're going to get into this argument of like, should we just use the tech we have now or focus on building AGI?

26:48But just a very interesting data point that Schmidt and Zhu add in their story. They say in a recent survey of the Association for the Advancements of Artificial Intelligence, by the way, who's that? They call it an academic society that includes some of the most respected researchers in the field. But this is interesting. More than three quarters of the 475 respondents said our current approaches were unlikely to lead to a breakthrough. While AI has continued to improve as the models get larger and ingest more data. There's concern that the exponential growth curve might falter. I mean, it's just interesting to see Schmidt saying this in the pages of the Times and Thomas Wolfe from Hucking Face saying this to the FT and Sam Altman saying there's a bubble and Mustafa Suleiman saying, you know what, let's stop calling these things conscious.

27:38Everybody together is coming together and talking about this AI, you know, hitting a wall or reaching a curve, reaching the diminishing returns point of the curve. as opposed to the exponential. And so I'm curious, Ranjan, what you think about the context of where AI is heading overall, given that this is what we're hearing. Material security is transforming how companies protect their most critical cloud assets, like Google Workspace and Microsoft 365, with modern purpose-built security that actually works the way people do. The biggest cloud threats walk through three doors, email, identity, and data.

28:14Material was built from the ground up from day one for Google Workspace and Microsoft 365, not as a retrofit, so it closes all three. It provides continuous protection before, during, and after an attack. You detect problems early, contain them fast, and recover without chaos. Lean security teams scale through intelligent automation rather than adding headcount. Material blocks evasive phishing and impersonation. It protects sensitive content using built-in rules for personal health and financial information, or rules you define. It also spots risky applications, unsafe settings, and sketchy verification or password reset attempts, all without slowing people down.

28:57So get the overview at material.security. That's material.security. 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. Every 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.

29:35Check 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. This content is not investment advice and trading crypto involves risk. The Gemini credit card may not be used to make gambling-related purchases. I mean, I could not be happier.

30:11Listeners can't see how much I'm smiling right now that Eric Schmidt's on team product, team build. It's not the model. Everyone is coming around right now saying, let's take the technology we have now and actually try to figure out how to build with it. I think, again, his whole point that in China, people are focused on actually integrating AI into hospitals and farming and other areas of life and medical diagnostics. That's what we need to be doing. That's what we have been saying for a long time. And the kind of over-focus on this one model to solve all of our problems. If it kind of feels like that's going away, and I don't know, that I'm excited by this.