OpenAI's GPT-5 In 2025?, Big Tech’s Big AI Spend, The Polymarket Election

0:00OpenAI's Sam Altman says GPT5 isn't coming this year. We discuss why. Plus, big tech is spending big on AI. Alexa can't get its head straight. Meta says yes to AI slop. And is this the polymarket election? All that and more is coming up right after this.

0:19You're used to hearing my voice on the world bringing you interviews from around the globe. And you hear me reporting environment and climate news. I'm Carolyn Buehler. And I'm Marco Werman. We're now with you hosting The World Together. More global journalism with a fresh new sound. Listen to the world on your local public radio station and wherever you find your podcasts.

0:46Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced format. What a show for you today. We have so many forces colliding, big moments in AI combined with big moments in big tech earnings. And of course, we have an election that's just a few days away here in the United States. So there's so much to talk about, and we're going to do it as always with Ranjan Roy of Margins. Ranjan, good to see you. Welcome to the show. It's 75 degrees here in New York on November 1st, so I'm not complaining right now. And absolutely nothing going on in the world, right?

1:19Absolutely. Nothing to talk about, Alex. Well, let's bring the heat, okay? Because this has been a very interesting week. You know, it seems like you can't go a week without OpenAI making an interesting headline. And they made two this week. So let's talk a little bit about it. So to me, the most puzzling thing we've seen from OpenAI this week is the fact that Sam Altman went on Reddit, did an AMA, and told all the Redditors, the Reddit hopeful, the Reddit faithful, shall we say, that there's not going to be any GPT-5 this year. Remember, OpenAI months ago talked about how it started training GPT-5, but no GPT-5 at the moment.

1:56This is what he says. He says, all these models have gotten quite complex and we still can't ship as many things in parallel as we'd like to. He says the company has limitations and hard decisions when it comes to allocating compute resources toward many great ideas. He says OpenAI is going to have some very good releases coming later this year, but nothing that we're going to call GPT-5. I have some thoughts on why we haven't seen GPT-5, but I'm curious what your perspective is. Yeah, I think the bullish case about this would be that they are recognizing that just investing in some massive new model rather than actually productizing the models that they do have.

2:41And 4.0, 4.0 Mini, these are all pretty good. So I think if they're actually moving that direction, it'll look better for their financials. It'll look better for their actual user adoption. It'll look better for any future fundraising. so the bullish take is that sam altman is finally recognizing talking about the here and now rather than some kind of far off pipe dreams around what our agi could look like that would be my bullish take the the bearish take is just that they don't have anything and they're just stalling right now so i have two perspectives on this the first is just like the simplest one which is that expectations have been built up to the point that releasing something GPT-5, and if it's not AGI, it's going to look bad for open AI.

3:29And we just said Aidan Gomez, one of the authors of the original Attention is All You Need paper that kicked off the generative AI revolution on the show. And he talked about how, look, you can train the models with more compute, they're going to get better. But there is some sort of saturation point where the gains don't become as exponential as they've been. That's his perspective. Here's a slightly different take if we put that aside. So I think that they did begin training GPT-5. And since the time they've been training GPT-5, the amount of spending that's taken place within their competitors on competing models has been beyond what they expected.

4:07I'm talking about the 100 ,000 GPUs that Elon Musk has put together at XAI. And this week, we just got news that Meta has put together a cluster of more than 100 ,000 NVIDIA H100s for its Llama 4 model. Zuckerberg told investors that they're training Llama 4 models on a cluster that's bigger than 100 ,000 H100s, bigger than anything I've seen reported for what others are doing. And we do know that as you add more compute into the model, you do get better results. I think OpenAI thought, and this is just speculation, but I think they thought maybe we can get GPT-5, let's say 60 ,000 H100s or 70 ,000 H100s.

4:54And then they've like watched Musk and Zuckerberg pass them with this massive cluster. And they've probably said to themselves, we're going to need a bigger cluster here. So they raised 6.6 billion. They're going to lose$3 billion on training costs this year. I think what they're doing is basically saying we need to match these competitors on size. Otherwise, we might lose our lead. Because if GPT-5 comes in worse than Llama 4 or worse than whatever XAI is doing, it's not curtains for OpenAI, but it's certainly not good. What do you think? I disagree completely. I think they showed with the last model release, calling it 4.0 and 4.0 mini and just kind of adding some additional letters and going into some kind of iPhone naming convention is good enough that people will be fine with it.

5:44They'll still be able to raise$6.6 billion and everything will be okay. So, and especially because we had talked a lot about OpenAI's financials and like consumer facing subscriptions are still going to be the bulk of what they're actually projecting. So they need to, and we're going to talk about it, make search GPT work. They need to release Sora into the wild and allow users to create videos. They need to make chat GPT better and better. Clawed artifacts is amazing. And they're supposedly working on, I think it's called a chat GPT canvas, which is going to be an equivalent product. So, and it's still only in beta.

6:23So I think it's good. I think it's like if they really focus on those kinds of things and let the massive arms race for pure compute play out, but actually focus on the products, I think Sam Altman could have a chance. Well, I don't think these things are mutually exclusive. And I ask you this, if you were running that company, are you okay with seeding the leadership of having the best foundational model out there to Elon Musk or to Mark Zuckerberg? I think, I mean, the ego involved alone means that will never happen. And OpenAI is almost like built itself on the kind of mythical nature of its foundation model.

7:04So if I am Sam Altman, there's no way that's happening. If I am me and I'm looking at the Ron John Roy of margins and I'm looking at the future of the company in the business, actually, sure. I don't, because I genuinely don't believe, like productizing generative AI is hard. Like what's happening with Alexa, there was a long piece in Bloomberg this week, making search GPT as good as perplexity and work well and making the UI nice. Like this is where I genuinely believe the actual battleground is going to be, not the next frontier model. Even though none of them say that and they all are talking about clusters and future models.

7:48And I still just think the companies that have the best products are going to be the one that actually win. Well, we've been talking about on the show about how the application is where the real value of AI is this entire time. And now you're telling me that OpenAI is going to where the real value of AI is, which is to making products out of the technology. And I'm like being skeptical. So, okay, I totally understand you. I understand where you're going. but I do think the leadership on the foundational model component is super important except if there's one thing that's happening that I think the general public hasn't taken into account and that is that the delta between GPT-40 today and whatever meta builds on it's a 100 ,000 GPU plus model isn't going to be that big like obviously the next foundational models will be better.

8:39But unless you can tangibly feel the step change, and I'm not sure you will, you know, then if you can't feel that step change, then it makes sense to go product only, right? Like, basically, if you could have GPT, this next GPT-5 or this meta model, that's like really capable of like getting deep into some PhD type territory. Well, it's only going to be useful if you're like PhD trying to discuss that thesis, right? Like there are limits to how smart you can make this thing which is kind of weird to say out loud i don't know i'm spitballing here but what do you think about that take no no i i think like okay do you remember what the difference between uh four and four oh no the voice thing maybe but that's a productize that's the product side of it yeah exactly and like uh o was omni so i think that was related to the voice like omni-channel delivery of large language models but but again it's and still the company's moving they're releasing products they're trying to get users they're getting subscriptions so overall i think the company is moving into where they see the battleground and remember they just closed the largest fundraising round in vc history they are going to have to deliver some revenue like uh they have to deliver something and they're probably going to from all speculation have to raise again at some point.

10:04Largest VC round in history. It's going to last them 11 months. Exactly. So show some revenue numbers and sales numbers and traditional financial metrics. And maybe in the background, they are actually developing something amazing, but it would still make more sense to not build hype around it right now and not raise expectations to an obscene level unless they're delivering anything very, very soon. One note about Sora, because they did address that in this AMA, Kevin Wheel, the chief product officer, former chief products officer at Instagram and Twitter, he's now running product at OpenAI. He's also been on the show.

10:43He talked about Sora. He said it's delayed due to the amount of compute time required for inference due to safety. He said they need to perfect the model and get the safety impersonation other things right and scale the compute so i just read that and we're going to get in search gpt but on sora quickly which is the video generation model i think we could go all of 2025 and not see it i'm pretty sure we will like video i've played with runway ml has a video generation feature i've tried a few of them it's interesting right now it's kind of where image was a few years ago text was even before that but we are so far away from video i genuinely believe being a reliable thing because it is so incredibly difficult to do well so i think it weren't i agree we're not seeing sora this year this year they're not gonna i mean not even this year in 2025 did you just do an about face and agree with me mid-sentence i i agree with you we're not going to see Sora.

11:44I think they're going to focus on the products that can actually win, like Search GPT. Ranjan, I'm going to let you run with this because I'm very curious to hear your perspective. This is one of your open AI products, and this is the new strategy. So this is a big moment for you in particular. But let me just read quickly what it can do. So you can search it, search the web with ChatGPT, and it can return both information and links. So this is from The Verge, the search lead Adam Fry showcased the feature by searching for Apple stock and any relevant news in return. ChatGPT displayed an interactive stock graph, upcoming earnings information, and news articles with clickable citation links to original sources.

12:24He also did a search for restaurants in San Francisco and he asked follow-up questions and got it to narrow down on certain types of places, which I think has been like one of the dream applications of this technology. So talk a little bit about Search GPT. Have you tried it yet? And do you think this is a big moment for OpenAI? I have tried it, and I think it's a big moment. And I think that now they're going head to head with perplexity. And in the whole area of search, I've been thinking about this a lot. There's kind of traditional search in the Google sense of you have a search bar and you type in some loose words to get a general direction and list of websites.

13:05Then there's a more traditional kind of generative AI chat GPT query where you ask a long question and you try to get really condensed information. And then there's something in between. And I think perplexity has been the leader in that in-between space where you want some information, but you also want some links out. And perplexity, I honestly, the related questions, the links out, that has become my search. And OpenAI playing in that space, I think, is very, very interesting. So basically, right now, there's a little globe icon at the bottom of the entry bar in ChatGPT. If you click it, it still gives you a generative AI-style chat answer, but it also has a list of links and citations.

13:54And I think that's basically perplexity. The UI, the field, the actual utility and use case is perplexity. And in fact, I asked, is the big technology podcast good? And I can tell you that Search GPT told us overall, the big technology podcast is well regarded for its informative and thought provoking discussions on technology and society. What's up with that qualifier overall? Come on, Search GPT. Overall. Unbelievable. Overall. Well, it also tells us that a reviewer on Apple podcast described it as a must listen for anyone interested in the tech world. So a reminder, not only will your five star reviews help us in terms of the platforms of Spotify and Apple podcast.

14:43As as the world moves towards generative AI search, your reviews will help us even more in terms of showing up in search GPT results. but overall i'm excited about it because perplexity has replaced google for me i would say majority of my information querying which i don't even just say is search is done on perplexity and they already on day one search gpt is pretty good it's very interesting because you have like these new uh search engines whether it's search gpt perplexity did you see this week also the information reported that meta is building its own AI search engine. It says to lessen its reliance on Google and Microsoft.

15:26And basically like people search meta for news sports stocks, and it will have to use Google and Microsoft to answer those queries. And if it builds its own AI search engine, there's a chance it won't need to rely on them. So that's interesting. But I think it's interesting that there are all these companies developing search engines, and yet Google, they They had great earnings results this week. And they've remained dominant. Like they showed that they were like making money from search and that they're growing. And I saw an interesting back and forth between Josh Miller, who runs the ARC browser, and Arv and Srinivas, who runs Perplexity.

16:04They were both talking to each other on X this week. And basically, Josh was saying that as long as Google has the controls, the browser and the control T, right, which means new tab, it will control search, bottom line. And Srinivas said he agreed. And I'm just kind of curious. I mean, yes, a lot of people are gravitating towards these new search engines, but also just the structural advantage. The built-in platform advantage is pretty big for Google. Are these two like this? It's like the unstoppable force and the immovable object. Well, first, I would like to say regarding Josh Miller and the Arc browser, I don't know if you had seen this as a avid fan of the Arc browser.

16:46They just announced they're going to be moving away from it onto a new project. So if Josh Miller is listening, please keep working on the Arc browser. It's a wonderful way to browse the web and actually has integrated generative AI features. but I think beyond that the platform question it's true because the arc browser was going after chrome the android operating system obviously is like a massive platform that allows the space to be dominant in generative ai like google has the platform they have chrome the web browser but I think they're going to be they're losing they're going to be losing more and more of that utter dominance in terms of platform.

17:28Like if antitrust, we've heard, could they have to spin off Android, there's already going to be some necessary repercussions around what Google will have to do in terms of like meeting the antitrust claims upon them. So I think they're going to be losing more of that platform. So it's going to be an actual battle around who is going to be able to deliver the actual best information experience, not even the best information, but the overall best information experience like even apple just released their sports app and i'm not googling anymore what the world series score was all week i was just it was like live in the uh that bar at the top of my iphone so everyone is going at google in different ways right now yet google's revenue jumped 15 to 88.3 billion this week this is according to the new york times reporting on earnings, a 34 % increase in quarterly profit as well beat Wall Street expectation driven by advertising and cloud.

18:30So we've been talking about this for two years, right? Two years since ChatGPT came out. Google is probably in the best position it's ever been. The search advertising market, they still own. And Perplexity was supposed to be launching ads in Q4. And I know you had talked about it with their CEO. Like I still have not seen them myself. Have you seen any? I'll be honest. I don't use it too often. I'm still old school Google search. And I feel like my normie tastes are going to be reflective of the way that the rest of the world uses this stuff. I am going to make the call one year from now. You are in that medium ground between traditional search and generative AI chatbots.

19:18I think majority of big tech listeners will be as well. I'm making the call here. I'm going to put in the calendar November 1st, 2025. Should we live that long and be lucky enough to be on a stable planet? I will see your bet and we'll find some stakes on it and we'll figure it out. Maybe we'll create a betting market contract. Exactly.

19:44And Before I just want to say before we move on, this week also featured one of the funniest scenes I think I've ever read about in tech history. And that is that Google employees were meeting with their executives, I think, for their weekly Q &A or they don't do it as much anymore. So I'm curious what the setting was. But this is according to some reporting from CNBC. So they showed up in their Halloween costumes. So Google chief scientist Jeff Dean wore a starfish costume to the meeting. The CFO, Anat Ashkenazi, wore a jersey from Indiana Pacers star Reggie Miller. CEO Sundar Pichai wore a black t-shirt that read, Error 404 costume not found with an image of a pixelated dinosaur.

20:29And they're there standing in front of the company. And some of the employees are like, so are there going to be more layoffs? That, I'm so happy about that because one of my wishes in life, I think all great human drama should take place on Halloween. So you have to be completely serious while in costume. Yeah, unbelievable. I mean, it's always, it just adds such an element to anything that's going on. So this is what Sundar said, while in his Arrow 404 costume not found shirt. It's like at the scale of our company, there could be moments where these small groups of people are impacted. Now, look, let's be fair to him.

21:05It's not like he's making a layoff announcement in his costume or whatever it is. But it is just interesting that like it just shows you what the company has been through with the layoffs over the past couple of years. That there's still some trauma within the workforce. And even in moments of like lightness where like executives are talking with employees, you're going to have this come up. And just to have it play out as they're wearing their costumes is I just I just think so ridiculous. I don't know how to explain it. It's an unbelievable scene. It's like from a movie. It's from Silicon Valley.

21:42Exactly. And HBO, the greatest show of all time. It's why it got it right. I'll be honest, I could never watch the show. Just too close to home. I was living in San Francisco, and I was just like, I'm not touching this thing. Maybe I'll give it a try another time. I have a similar industry on HBO I still have not watched, having been in the finance and trading world. I want to, but a lot of the times, any of that kind of programming, I feel it's tough. I agree when it's too close. Definitely. Okay. So we have to talk about this new report from Bloomberg about Alexa. So let me kick this off with saying I have three of them in my home, three echoes in my home.

22:24And my wife and I wake up in the bedroom with the echo alarm. It's great. You can leave the phone outside, set the alarm with your voice. and, you know, and it's supposed to play Amazon music. Guess what I've been woken up with for the past two days? What? Ads for Celsius energy drink that Amazon music played for me through my Echo. Because. That is exactly, when you wake up, that is the most soothing possible sound. I couldn't believe it. It was the. Transition into a beautiful New York City game. The beginning of the ad has like a can popping open. And it's like, you know, we're usually waking up to soothing music.

23:06Can you understand how bad of a day we're going to have when we have to wake up to an ad from Celsius Energy Drink? Because the people behind Amazon Alexa can't figure out that they're playing music for an alarm and maybe start with a song first before shoving an ad down our throat. And that brings me to this week's story about Alexa from Bloomberg. Alexa's new AI brain is stuck in the lab. And so basically, Amazon has been on this process of trying to develop a new Alexa with new large language models, and it's just not working. So from the story, Andy Jassy asked it a bunch of sports questions, asking it to drill down into individual player performance, league standings, team history, and so on.

23:54Then when he asked about a recent game outcome, Alexa just made the score up. And the story says, Alexa's conversational abilities have improved since the Jassy demo, but top engineers and testers involved with the efforts say the AI enhanced assistant can still drone on with the relevant or superfluous information. And it struggles with the humdrum tasks it previously excelled at, like turning on and off the lights. And that explains a lot. And this is what the people say. Jassy has yet to convey a compelling vision for AI-powered Alexa, many of these people say the project still needs tons of fixes and that they're not bullish that the resulting product will compare favorably with the long list of AI apps already on the market.

24:41Some senior engineers said that Amazon's best hope is that they can ship the 13th or so permutation of chat GPT. And now it feels like Amazon is playing catch up. I mean, this is, to me, it's so disappointing. I mean, for anyone who saw the promise of Alexa, and I still believe that there is promise there. The way that the company has flubbed this is exceptional. And it's remarkable to me that when there's so much innovation on the AI front, the Alexa effort within Amazon is just, I mean, I don't know. It's total garbage. I got woken up with an energy drink ad. I mean, that's unacceptable. Sorry.

25:24I think it's a reminder that in artificial intelligence, sometimes basic intelligence can be lost when people are trying to kind of shove features into existing products. I agree. I'm disappointed. And I will take this moment to just bash Siri again for once. But I think they're facing the same problem. voice interfaces should be the most insanely natural places for generative AI interactions to happen. Like it's literally the natural language is how these things work best. So being able to talk to and get the right actions and results and information should be so easy. And honestly, uh, open AI chat GPT's voice mode is pretty amazing.

26:13Obviously it's only giving you back information and it's not taking specific actions like turn on the lights, set an alarm, interacting with third party apps, but it's pretty good. So Alexa should be way, way ahead on this. And I think this whole article, it kind of reminded me of at these big organizations, there's been a lot of talk, especially around the layoffs at Amazon, the return to five days a week in the office. It's like an organizational bloat and teams that aren't like functioning at their highest capacity, a lack of shared mission. It's kind of crazy when you can see it in the product.

26:52You see it in the output, you see it in the feature, and you can almost feel that lack of cohesion in some big organization in an office park in Seattle. You can actually feel that in your little device in your kitchen. And this one, I definitely do. I'm so happy you brought that up. You're right. The culture is tangible. And this is exactly what you're saying. It's a culture thing showing up in the product. And here's sort of my explanation for it, which is basically that Alexa was something we all know Jeff Bezos came up with it. Right. It became. And he pushed to get to the amazing stories of like, you need to get the response time down to some number of milliseconds.

27:35And like they said it was impossible. And he like just berated everyone until it got done. He pushed it. Exactly. And so it comes through. He obviously has a global view of the company, but it sits in retail. Okay. He leaves. And as Bezos is leaving, generative AI rises. But here's the thing. The new CEO, Andy Jassy, comes from cloud. And cloud and retail, not very associated within Amazon. So Jassy comes in, he's going to care a lot about cloud. Cloud is obviously the future of the company. He also has to clean up all the superfluous spending that's happened within retail because they built up as if COVID-type shopping was going to continue forever, and it hasn't.

28:13So they had this massive build-out, so he's got to get costs under control. What's the thing that's going to slip through the cracks? Alexa and generative AI. And it sure did, it seems like. and it's just a massive, I think, own goal because if they got this right, it could be a tremendous business success. I think the stat that I saw is the echoes in one in four households and I know other households have the Apple HomePod like you do or the Google Home or whatever it's called now, the Intelligent Nest 3.0. You never know what Google names these things. Maybe they should all get in costume and name their devices so they can do a better job.