GPT-5 Analysis, Brex Growth, Maven's Fertility Tech & NVIDIA Family Dynasty | Aug 8, 2025

0:13Welcome, everyone, to the Informations TI TV. My name is Akash Basritcha. It is Friday, August 8th, and we are taking a full tour of the tech sector today, folks. We have got some earnings to debrief in the grocery delivery and social media sectors. We are also hearing about some early reviews about OpenAI's GPT-5. We're then going to get into fintech and talk with the chief business officer of Brex. And we're also going to get into health tech with the CEO of Maven. And finally, we've got a feature story about Jensen Thong's kids. That is from our weekend magazine, and that promises to be a fun one.

0:47Before we get started, I want to alert you about another meta acquisition that the information was first to report last night. Meta has bought Waveforms AI, a startup that uses AI not only to understand emotion from voices, but it can also mimic emotions using AI-generated audio clips. Now, you might recall, Meta just bought another voice AI company called Play AI. We actually had one of Play AI's investors, Christine Tsai from 500 Global on the show. It seems like voice is going to be an increasingly important part of Mark Zuckerberg's personal super intelligence ambitions. And so I'm quite curious to see what other deals he gets done.

1:25We, of course, know that OpenAI debuted some new voice features itself last night, yesterday at the big GPT-5 event. And so without further ado, let's get into that. OpenAI made quite the splash yesterday with its launch of GPT-5. There was a big fancy keynote. Sam Altman clearly took some notes from his new friend, Joni Ive, about how to launch a product kind of in a similar way, the way Apple does with its iPhones. But AI models aren't as easy to evaluate and rank as iPhones are. And so today, I want to bring on a panel of people who are very familiar with GPT-5 to help us make sense of yesterday's launch and understand whether OpenAI lived up to its own hype.

2:06Stephanie Palazzolo is our AI reporter at The Information. Matt Schumer is the CEO of AI personal assistant company OtherSide AI. And Siki Chen is the CEO of AI financial analysis company Runway. And a quick note before we get started, Siki is a small investor in OpenAI rival Anthropic through an SPV. Just wanted to get that out of the way. It's so great to have you all here. I want to get right into it. Stephanie, Let's start with you. Why was GPT-5 such a big deal yesterday? This was huge. Definitely. I mean, I think GPT-5 has been a long time in the making. People have been looking forward to this model for quite a while.

2:43The last time that OpenAI really released a new flagship model was probably last May with GPT-4.0. And then earlier this year in February, they released GPT 4.5, which actually didn't go over quite well. And so I think this is just a situation where hype has been building for a long time. Another reason why it was so important is because this is the first time that OpenAI has actually combined its traditional GPT LLMs with its reasoning O-series models. And so this has been something that Sam Altman has been talking about, you know, wanting to do for quite a while in order to make the, you know, experience of using ChatGPT much easier for consumers so they're not having to decide between so many different models.

3:30So, yes, I think a lot of excitement and hype for sure leading up to this launch. Matt, you had early access to GPT-5? Correct. And so you've been playing it for a couple of weeks. What do you think of it? I do think it's an amazing model, but there are caveats, right? So the first thing is, and you have to think about this when you think about this release, it's not like previous Open AI releases when we're talking about going from GPT-3 to GPT-4, right? When that happened, in that era, Open AI was the only game in town. And what that meant was that when you looked at the difference between those two models and those two releases, the difference was dramatic.

4:09But for the first time now with two major model releases from OpenAI, right, GPT-4 to GPT-5, we've seen tons of models in between from OpenAI and from others. We've seen GPT-4.1, we've seen GPT-4.0, GPT-4 Turbo, Claude Sonnet, right, Gemini. We've got to see sort of the incremental improvements. So I think a lot of people are feeling like this isn't a crazy jump from what they're seeing because they've seen the incremental jumps. but look at the actual models from gpt4 to now gpt5 it's night and day you can't go back so it's a definitely it's an interesting release i think it lived up to the hype but i think the hype was a bit weird for this one because you have to take it out um i do think it's an amazing model i had an incredible experience with it but there are points where it's weak for example opening i talked a lot about its creative writing abilities i did not find that to be a strong point at all in fact I found it to be quite weak.

5:05I still much prefer GPT 4.5. But on things that involve a lot of math, a lot of math, a lot of code, this model is a huge step up. It's allowed me to do things that other models just can't touch. Siki, you've been using it since yesterday, since the launch. What have you been using it for just in your early tech? Yeah. So two things. So I forgot to mention, I'm actually the creator of the number one education GPT in the ChatGPT store called Universal Primer. so i've been testing it on that um and i think uh for that purpose it feels like a small downgrade um and the reason is it is much more easy to read it feels a lot more approachable but amount of detail it goes into out of the box for the prompt that exists that i have without additional fine tuning uh it's substantially less um and that's somewhat surprising to me what do you What do you mean by the amount of detail it goes into?

6:03What do you mean by that? What Universal Primer does is it tries to explain any difficult topic, like you want to learn about gauge theory or quantum mechanics or how a transformer works. And it tries to do so with really easy-to-understand analogies. And I prompt LLM to go into as much technical detail as possible. The amount of technical detail it goes into by default and the understandability of these analogies seems like a meaningful step down from the answers i got before right i can probably get it to uh do something better if i retune it so i haven't adjusted a prompt right right you know from what matt and i oh matt mentioned to me is that uh it's not going to always be better out in box but it does respond to fine tuning um or fine tuning to prompt what i will this morning is that I got a note from our team and said their experience with GBD5 and Cursor is a meaningful step up.

7:02Matt, I want to talk about this idea of benchmarking. I mean, you've got all these websites. Ella Marina is probably the most popular one, right, where everyone is just so into this idea. We need to compare which one is better. We need to benchmark things. I mean, how do you actually benchmark these models? How do you quantitatively compare which one is better? For the most part, I think we're past benchmarks, or at least we need way better benchmarks. I tend not to look at them for new model releases. You'll notice in my commentary on the announcement yesterday, I didn't mention benchmarks at all, which is different from previous models.

7:38But we're really at this point where it's so hard to tell from benchmarks. You have to use the models. You have to try them for your use cases. What I find works for me, for my use cases, might be different for somebody else, right? It might be so great for the things I want to do with AI, and it might be terrible for the things somebody else does. So I think we're entering this era where you just have to use it. I mean, that's always been the case, but you can kind of get a sense of the progress with these benchmarks. We're past that. So the leaderboards are kind of, I mean, they're just really, we're down to subjective use now.

8:09That's the main opinion that matters. Yes, right. The model is a step up, for example, in pretty much every coding benchmark I've seen, if I'm not mistaken. But it doesn't seem like that big of a step up. But when I actually use it in practice for coding, this model is a dramatic leap even over things like Claude Opus 4 or 4.1. It's been able to touch tasks that previously I wouldn't be able to do with the help of Opus over the course of, let's say, a month that it's been able to do them in like a day. So yeah, it's really hard to see the numbers. You have to use it. But I can't say that my experience is going to be the same one that everybody else has.

8:44You have to try it because everyone's use cases and everyone's style of prompting are different. And this model takes a lot of fun to it on the prompting side. Stephanie, I mean, coding tools have been all the rage lately. You know, it's what's raising a ton of money in the AI space. How much of a focus was the coding capabilities of this model for OpenAI? I mean, it seems from our, you know, reporting that we've done that it was a huge focus. I think up until this point, that's been an area that they've maybe fallen a little behind on in terms of Anthropic has really had the models that are being used in all the coding assistants.

9:23And so we know that in the process of developing GPT-5, that was a huge focus for them. And I do think that, you know, there is a focus on not only being able to do more practical everyday coding tasks, but even as we saw in the demo yesterday, there is a focus on GPT-5 kind of understanding the aesthetics and design of apps and websites better so that even if you give it a more simple prompt, it'll know how to lay out a website in a certain way to make it aesthetically pleasing or what colors or button designs to use. And I think in that way, it's also making it more approachable and easier to use for non-technical users who might want to use, you know, ChatGPT to help design like a personal website or like an app to, you know, make flashcards to help study with.

10:14Siki, we just heard from Matt about how the leaderboards, you know, might be coming less relevant. I mean, from my view, it feels like the reason that it's less relevant to sort of focus on how good any one model is, is because the reality is six weeks from now, there'll be another model from another company that inevitably does something better. I mean, are we past the point of really caring what any one model can do it doesn't really even matter that these new releases are happening uh i think it matters um i mean the the evaluations are getting saturated um but as we saturate them new uh eval sets are coming out and so i don't think it's quite the case that these rankings don't matter anymore.

11:02I think my read on this is that if you read the GPT-5 system card, the mini models that make up GPT are all successors of existing models. And most of these models really came out in the last six months. And so my usage of it, it feels like, and the way GPT-5 works, is it's routing to these mini models that are successors of the last generation of models. And it doesn't feel like fundamentally a new set of models. It feels like a very well fine-tuned set of models for coding and for approachability. And so my feeling on this is that this is designed to make ChatGPT feel better. And from a pricing and coding standpoint, it's designed to be a direct strategic competitive move against the coding business of Anthropic.

11:50And so my expectation coming in was not that GVD5 was going to be substantially better. Because I think, Sam, about six months ago, at some point they said, we're going to shift strategies and make GVD5 a merger of all these models. And I'm expecting probably for the end of the year an actual pretty large leap in capabilities from OpenAI. I wasn't expecting that from this district. Matt, what are your thoughts on that? Because we already have Elon Musk tweeting saying, look, GROG5's going to be up before the end of the year. I mean, again, every six weeks we get a new model. Should we really be making that much fanfare about the latest model to come up from one company?

12:29I mean, I think it's just natural. It's how X works. It's how the community works. Everyone gets excited. We're all nerds and we all love seeing these updates and the progress, right? And it is, I mean, objectively very exciting. On a model-by-model basis, like, look, are the improvements going to be dramatic? Probably not. but they are going to be there and over time they're going to compound and add up right when we test new models at hyper right right for example in our main writing product the models of two years ago could do the tasks we needed but if you actually look at conversion rates and retention rates every time we increase the model's capability but or the product's capability by adding a model with more capability usage goes up retention goes up churn goes down conversion rate goes up so even though right these feel like little jumps you actually see a lot of effect over time.

13:16So I think it's important that we keep seeing bits of progress here and there. Obviously, there's gonna be more fanfare than maybe there should be for each model. But that's just how things are. Stephanie, last question for you before we go. What are you watching for from here? What questions did yesterday's announcement actually raise for you going forward? Yeah, I mean, I think one question I have is, you know, as as Siki mentioned earlier, this is a different type of model where it's kind of a merge of a bunch of different models under the hood. So there's probably some element of routing between, you know, reasoning models and traditional LLMs.

13:54I think a question I have coming out of this is, you know, with each additional model, since it'll be harder for us to tell where the improvements are really coming from, if it's coming from, you know, improvements in the reasoning model from post-training, or if it's coming from better pre-training methods in the traditional LLM. I think it'll just be harder for us to kind of understand where the trajectory of AI development is going and where are the most promising places for these improvements. Like, will it be from reasoning? Will it be from pre-training? And so I think that's a question that I have moving forward.

14:31Another question that I have is just, as mentioned, the benchmarks and the evals can only show you so much, but just like how does this work in practice with actual engineers? I think one chart that really stood out to me from the safety card is this evaluation that OpenAI does where they're trying to see how much of like a typical engineer's work the model can automate. And so actually, if you look at 03 versus GBT5, the eval that they did, I think it had like 44 % versus 45%. So I think that really stood out because you see all these really impressive results on coding evals, but then you see this very practical evaluation of like, okay, how much of just like an everyday open AI engineer's work can a model do?

15:21And it's only being better by 1%. So I think that was really interesting. And, you know, it's just one data point, but I'm curious to see how the model performs on these more like practical evals in the future. Well, look, I'll tell you what, like we said, I mean, these models are coming out so fast and so quickly. The three of you are very great at explaining it to us. And the benefit is that you actually use these models, Siki and Matt. So thank you for coming on the show. We appreciate it. We will have you on again the next time we have a model to unpack. That is Siki, Matt, and Stephanie. Before we get to our next guest, we do have some earnings to recap.

15:58Instacart reported earnings last night. And the one number that you need to know is 11%, which was the revenue growth rate. And yes, it was an acceleration from the last quarter. And yes, shares were up today. But really, the reason you need to know 11 % is because it's less than half the growth rates of DoorDash and Uber Eats. And so really, I mean, this company, from my view, has to find a way to catch up. We should say that Fiji Simo, of course, is leaving for OpenAI. And so this is very much a transition moment for the company that I'm excited to see how it plays out. Pinterest also reported last night.

16:31The most important number from that print was 17%. That was the top line growth rate. And the company is forecasting it'll stay the same for the year. Now, the stock was down this morning. I really think it's because other advertising businesses like Meta have shown their top line is accelerating. It's growing much faster. Pinterest is kind of steady. And so this is all competition at the end of the day. And finally, Chime reported earnings for the first time as a public company. The number to know is 29%, which is at the high end of what it expects to grow revenue this year. That is basically right where the company grew its top line last year.

17:06It's a little bit lower. The stock also moved a little bit lower. Overall, it was a pretty vanilla quarter. Okay, speaking of time, let's go to another fintech company. Brex is a company we've written a lot about at the information. Last year, the corporate credit card company laid off a fifth of its workforce and made some big changes to its C-suite. This year, Brex has been on a bit of a rebound. The company also just locked in a license to operate across the entire European Union. And I want to bring on Art Levy, the company's chief business officer, to give us a bit of a status update on how that's all going.

17:38Art, welcome to TITV. It's great to have you. Hey, great to be here. So excited to chat this morning. So Art, I want to talk about where Brex is now, where it's going. I mean, look, we reported in February that Brex was on track to hit$500 million in annualized revenue at the end of the year. And we also reported that the company expected to start generating cash by the middle of the year. It's now officially past the middle of the year. And so what can you tell us? What's the update? Yeah, look, I can't comment on total revenue numbers right now. But what I can say is that Q2 was our best quarter ever.

18:14Our sales team hit 140 % to plan. And we actually finished the quarter growing 45%. And that's both top line and gross profit. So on those metrics, we're actually tracking ahead of where we thought we would be at this time. And so at the start of the year, you guys thought you would grow revenue significantly less than 45 %? No, I would say in line of that, but I think we see it, what's the exciting thing is we're seeing it accelerate and pass that 40 % marker that I know you guys had reported on previously to 45%. And we'll continue to accelerate past that. The reacceleration that we're seeing is super exciting, and we don't see a reason it won't actually continue to go even higher.

18:59So let's talk about where that growth is coming from. I mean, are these new customer segments you guys are targeting? Are they new products? I mean, what is responsible for that growth right now? Yeah, I think what we're seeing is twofold. Number one, I think the enterprise segment for us has really started to generate a ton of momentum. As I mentioned, as Pedro mentioned a few days ago, every new sales rep in enterprise in their first 90 days closed a$400 ,000 deal. So by the way, we are hiring. And I think the reason our product is resonating so much in this segment is also dovetailing nicely to the announcement we made yesterday around the global payment solution.

19:45We now have the only unified global payment product out there that has spend and card management that can be used with local cards all around the world. And so you're seeing AI leaders like Anthropic, Arm, and Cursor that use Brex, as well as just larger conglomerates, companies like Toast and Zoom and Wiz. Right. You know, I want to talk about all the different players that are in the space. It's no secret that Ramp has just raised a ton of money. They were valued at$22.5 billion this month. Brex was valued at$12.3 billion at its peak back in 2021. How do you guys think about differentiating yourself in this very competitive market?

20:28Yeah, look, I think obviously we saw the fundraise. Ramp, we have great respect for them. They're executing. We think it's a very large market here. We see our own traction. We see our growth at 45 % and accelerating. We see where we are playing and winning, where our AI, our global, and our partnerships are really resonating. And so our view is, you know, global payments is a$4 trillion market. There is a lot of room for a number of next-gen players to be large players. Candidly, we kind of see it as like a Stripe ad yen situation. Right. I mean, let's talk about, you know, that differentiation though, I mean, you know, is it branding?

21:12Is it tech? I mean, when you guys are in the boardroom, I mean, where do you think you could really stand out in this market? Yeah, look, I think it's threefold. Number one, it's our proprietary payment infrastructure. So we have built all of our own payments and we've done that over seven years. We built directly into our card partner, into our network partner, MasterCard and our banking partners, fifth, third, and emigrant bank. And so what we did there is become the only disruptor that can issue cards in 50 currencies. That is then dovetailing with our second mode, which is around our embedded partnerships.

21:49You saw this year that we announced deals with Navon and Zip to really see and embed inside of those leaders in the enterprise space. And then you couple that with what we believe to be an incredibly talented product engineering team that we have that is shipping products quickly and moving very fast. Just this month, as an example, our CTO was in a hacker house for the month with a number of engineers as we're working furiously on a very exciting AI launch to come later this year. I want to talk about the announcement you made this week that you are you've got your hands on a license now to operate across the European Union.

22:27Is this the first international market that Brex is operating in or have you guys operated outside of the US for a while? Yeah, that's a great question. So I wanna clarify that today, there are actually over 1 ,500 companies using Brex that have EU operations. What we announced yesterday is that now we can serve EU headquartered companies, not just US headquartered companies. And so this is super exciting because we now have the ability to bring our product to all of these EU companies that have frankly been asking us for it. It's a really big step in our view in deepening this global advantage.

23:05And we had been getting a ton of inbound from large European potential customers that want to use our product. And so is international expansion, I mean, is that really a core tenet of the company's expansion strategy moving forward? Yeah, look, what I would say is we are building the holy grail of global payments. one unified card and spend management solution that you can use anywhere in the world without FX fees, with local cards, with local issuance. So to do that, you have to build a deep payment infrastructure, which is what we have been doing for the last seven years and we will do for the next seven years.

23:45And having payment licenses is a key way to differentiate because then you can offer a better product in country. Well, look, Art, I want to let you go, but I can't let you go without mentioning a very important point here, which is that you mentioned, you put a bet out there that Happy Gilmore 2 was going to be over 90 % on Rotten Tomatoes. Do you know what it's at right now? Do you know the number? I got the number for you. My publicist says I cannot comment on this. Okay. It's 63%, Art. It's 63%, but it's a good movie. I will say I watched it a week ago and it's a great movie. So Happy Gilmore 3, we'll have to get you back on to make another prediction because it was a good one.

24:26That was Art Levy from Brex. Fertility support has been an increasingly popular arena for startups over the past few years, but few companies have been in the game longer than Maven. The company offers a number of telehealth services, not just for women, but also for entire families supporting their aspirations to raise healthy children and also for them to be healthy while they're doing that. I want to bring on the CEO of the company, Kate Ryder, to talk more about where the company is at. Kate, welcome to the show. It's great to have you. Thanks for having me. And wow, Happy Gilmore too. I was just listening to your last interview.

25:03I'll have to go look that one up. Did you watch it? I know. I mean, Happy Gilmore is like a classic in our family. So I've been living under a rock. But yeah, now, I don't know, 63 % on Rotten Tomatoes. It sounds okay, but we'll see. No, no, no. It's better. It's better than that. I don't know. I laughed all. Okay. Okay, well, look, let's talk about what Maven does. Explain it to us. Great, well, so Maven, you said we've been around for a while. We've been in the market for 10 years, and now we're the largest global virtual clinic in women's and family health. And so what that means is we have the largest virtual care network, the largest telemedicine network with women's and family health specialty providers for 30 different types.

25:44And we really cover the full journey from preconception and fertility care all the way through menopause. Part of that, women's health has a lot of gaps in care, so it's not just access to providers. So as part of our product, we cover the full fertility carve-out. So we offer fertility benefits as the payer. On the menopause side, we offer HRT. That's one of the big hot trends right now of women trying to get more access there. And then a lot of our core products is just giving women and families access to best-in-class providers to complement a lot of the in-person care pathways. And so the thing that's really interesting to me about your business is that you sell to the employers.

26:24I mean, this is like this is a benefit that employers offer their employees. Correct. We cover about 23 million lives through our employer and health plan clients. We work with over 2000 employers. And, you know, I think a lot of people sometimes don't realize this, but employers are on the hook for paying for health care. So a lot of the health care inflation that has been hitting the market over the last kind of three, four years, I mean, in general, but really in the last three, four years. employers are also bearing the brunt of a lot of that. Right. And, you know, I just want to make sure I understand how this business model works, because I think it's kind of interesting.

26:56So you approach a company, a company like Amazon, which is one of your customers, you approach Amazon, and Amazon pays you some amount for this benefit that they're going to offer their employees. Then you take some of that and you pay part of it to the provider. You keep some, like, how do you make money in this? Help me understand that. Sure. Well, on the fertility side, so we are like their health plan, right? So we have a contracted network of fertility clinics. We have, you know, relationships with the drug companies. And so when someone's looking to go through IVF or just look to start a family, maybe they don't want to go through IVF and they're trying to figure out, you know, how to optimize their natural conception pathways, then we are their benefit.