AI in Education, The AI Chip War, and Meta Superintelligence Lab Restructuring | Aug 18, 2025

0:13Welcome, everyone, to the Informations TI TV. My name is Akash Basriche. It is Monday, August 18th, and we have got loads of exciting news to get to today. We have got the CEO of Grammarly coming on the show to talk about the company's new AI agents. We've also got Chris Miller, author of Chip War, coming on the show to talk about, well, the chip war and where he sees this playing out. We're also going to get to Meta's fourth restructuring that its AI team is going through. That was a story that we published a few days ago, and we are going to get the latest on that from our meta reporter. But I want to start with a story that we just published this morning about Walmart.

0:51Walmart has been quickly growing its e-commerce business as it tries to compete with Amazon. And along the way, it has started to encounter some of the exact same problems that Amazon has long been grappling with. One of those problems has been balancing the number of small merchants and big brands on the platform. And my colleague Anne Guillen published a story this morning about how Walmart is dealing with all of that. I want to bring on Anne to tell us all about what she found. Anne, happy Monday. It's great to see you. Happy Monday, Akash. Good to see you. So let's talk about Walmart and e-commerce platforms, broadly speaking.

1:28I mean, when I go on Amazon or Walmart, I know that there are these big brands, and then there are also some of these smaller merchants. and what you started to tell us about in the story this morning is how these companies balance these two groups. Tell us about that. Yeah. So I think more and more consumers are starting to understand this in a bit more detail. And I'll just use Amazon as the example because they've been around longer. They've been dealing with this longer. But a lot of the products on any online marketplace, whether it's Amazon, Walmart, they're not directly sold by Amazon or even by the brand whose product it is.

2:12So, you know, Olay moisturizer, Coppertone sunscreen, those products aren't necessarily being sold by the brand directly. Oftentimes, a lot of those products are being sourced and listed and sold by what are known in the industry as third-party sellers, but they're typically small, independent sellers. A lot of times people do it as like a side hustle or as their full-time gig, just sourcing and selling different products. But that's where, you know, a bulk of the products on Amazon's marketplace come from these days are these independent sellers. And they're like resellers, essentially? Is that the idea?

2:55essentially yeah they're new they're new products um but they it's not being sold directly by the brand they're either the resellers are either purchasing the products from the brand or they're getting them from a wholesale distributor like beauty supply company sometimes they get them you know at off price or liquidation stores there are all kinds of ways that these resellers can get their hands on these products, which is part of the problem that Amazon and Walmart both want to have more information about where these products are coming from. Okay, so you've got these big brands and then you've got the resellers.

3:34These are the third-party vendors, I guess. Your reporting in the story this morning was about how Walmart is dealing with some of these smaller resellers. Tell us about what you found. Yeah, so in my story this morning, I reported that over the past couple of weeks, a lot of small resellers on Walmart's marketplace have seen their products be suddenly taken down, delisted. And what Walmart has told them so far is that they are changing a lot of their policies around who is allowed to sell certain products, particularly in the health and beauty and personal care categories. So any cosmetics, skincare products, sunscreen, lotion, anything like that, Walmart is taking another look at who is allowed to sell those products.

4:26So a lot of these third-party sellers, these resellers, they're kind of stuck in limbo right now because Walmart has taken their products down from sale, so they're not making any new sales. Walmart is asking them for more information about where these products are coming from. And Walmart has told them that, you know, that the new rule going forward is that only certain people will be allowed to sell these products, primarily the brands and then people who have explicit authorization to be a distributor or a seller of these products. And so Walmart is taking this perspective that they want less of these sort of smaller merchants, it sounds like, more of these bigger brands.

5:13Which one is better from the consumer's angle? You know, for me, is it better to have the smaller resellers or the bigger brands? It's a really interesting question. And it's what you mentioned, Amazon, at the beginning. And it's a question that Amazon and other online marketplaces have been trying to sort out basically since the beginning. You know, if you have these smaller sellers, You get more product selection, more pricing competition, a wider variety of stuff that you can promote to shoppers. But kind of the downside of that is that you don't necessarily have a clear window into where these products are coming from.

5:56You know, if you're the brand, you could get tired of all these small sellers undercutting your prices or selling things from last season. And so it's a really tough line for the marketplaces to walk. You know, do they keep the small sellers happy because they're buying advertising and they're paying for warehousing and fulfillment services? Or do they kind of forget the resellers and just go to work directly with the brands? So as a shopper, you don't necessarily see that going on behind the scenes. And what does this have to do with, you know, Walmart's goal to sort of catch up to Amazon? I mean, Walmart, obviously, it's a huge retail giant, right?

6:46It has this fast growing e-commerce business. I think you mentioned the e-commerce business right now is growing 11 times as fast as the overall business, if I have that number correctly. So it's trying to catch up to Amazon's e-commerce business. Is it doing a good job? Is it catching up? I mean, how does that competition look right now? Yeah, e-commerce is really important for Walmart right now. You know, obviously Walmart's a giant retailer, but e-commerce is still only about 20 % of their overall sales. So there's lots of room for them to grow that part of their business. And they've been trying to do that pretty aggressively over the last couple of years.

7:25And obviously Amazon is kind of their key rival in e-commerce. Um, but so the marketplace has been one way that Walmart has really been trying to grow its e-commerce sales and, you know, especially get shoppers where you might not be going to a Walmart store, uh, you know, find a new way to bring those shoppers in and get them shopping at Walmart. So Walmart has been doing a lot over the last couple of years to grow the number of products on their marketplace, get new sellers signed up. They've tried to be a little bit more friendly and welcoming to the sellers over the last couple of years compared to Amazon.

8:06But so for this move is really, I think they're trying to emphasize the quality of the products on their site. And, you know, this is something that Amazon has been dealing with for years. And so I think for Walmart, being able to say we're getting our products pretty much exclusively from brands, we know all our products that we're selling are legit, you know, I think that gives them a lot of kind of marketing firepower and potentially take a lot of shoppers from Amazon. Well, and it's a fascinating story. Thank you so much for coming on the show and telling us all about it. That is Ann Guillen, who covers e-commerce for the information.

8:45Now, we all know about how active meta has been in the AI talent war, spending hundreds of million dollars to poach star researchers. Of course, the next big challenge for the company is figuring out where to put all these people that they've hired, figuring out who will report to whom. Late last week, my colleague Callie Huang published some reporting about all the different variations that the company has been experimenting with to figure out what its AI org chart actually looks like. And I want to bring on Callie to tell us the current state of play. Callie, welcome to TITV. It's great to have you back.

9:18Thanks for having me. So tell us about what you found. What's going on in the land of org chart land for Meta's AI team? Yeah, so on Friday, I reported that Meta's new AI organization, Meta Super Intelligence Labs, is planning to split itself into four groups. The first is a lab called TBD Labs to be determined. Creative name. And then there's a group focused on products like it's MetAI Assistant. There's going to be a group focused on infrastructure. And then there is going to be FAIR, which is MetA's fundamental AI research lab. And that lab works on longer term AI research. So really, this is Meta's new AI leaders, Alexander Wang and Nat Friedman, starting to take some of their first big steps as the co-leads of Meta Superintelligence Labs.

10:22They come in after a particularly turbulent period for the company. And so Alexander Wang and Nat Friedman, they are sitting at the top of these four groups? Is that the way it is? Yeah, exactly. Okay. This is not the first time that Meta has reorganized itself in its AI team. No, it's not. It's actually the fourth restructuring in the last six months. Wow. Okay. So really, people are trying to figure out who their boss is. Yeah. I've talked to people who said, I have been here for three, four months. I've had the same number of bosses in that amount of time. And so how does this rearrangement now affect the company's overall AI strategy?

11:07How does it actually, sorry, backtrack. How did it sort of represent the new AI strategy? Yeah, so I mean, the company has had a pretty challenging last couple of months with AI. And in particular with its work on its flagship model, LAMA. And so Meta CEO Mark Zuckerberg has gone to great lengths to kind of turn that ship around. He, Meta agreed to spend$14.3 billion in June to invest in Scale.ai, which Alexander Wang previously led, and then struck a similar deal with Nat Friedman and Daniel Gross, the former CEO of Safe Superintelligence, to partially buy out their venture fund and bring them on board.

11:58um so this is really you know those new leaders kind of starting to figure out what they want um previously they brought in all the ai efforts under alexander wang right um and now they are sort of starting to sort out um where things might be duplicated etc so one of the questions i want to ask you is look you mentioned this is the fourth reorg and who knows it might reorg again two, three weeks from now? We don't know. However many variations it goes through. What I want to know from you is how employees are feeling on the ground on this team. Because in my mind, if you don't know who you're reporting to or you don't know what the structure looks like, how do you get any work done in the AI team?

12:45Yeah, it's really tumultuous. People say that it feels chaotic. I think the idea is that this reorg will hopefully be the last one for a while. You know, Meta is known for doing a lot of reorgs, but typically that is once every six months, not four in six months. They did one in February, where they moved out two engineering leaders and brought in a new product leader. They did another one in May. where they split up the group. And then they did one in June after Wang and Friedman came in. So that's a lot of change. It's super distracting. It's hard to focus. And so I think the idea is that hopefully this time around, things will settle down a little bit.

13:37And last question for you before we let you go. That new team, TBD Labs, what is to be determined? What is that team going to do? What's to be determined is its name. Oh, okay. Okay. What does the group do? The group is going to be working on Meta's new model, the newest generation of its model llama. So it strikes me as like, that's like the most important team. Yeah. I mean, this is where a lot of the new AI researchers who are coming in are working. And they've brought in people from around the company. Got it. Any bets on the name? I've heard some things, but probably to save for another story.

14:18Okay. All right. Well, when you figure out the name and when you figure out what else the team is, you know, what they're actually coming out with, then come back in the show and tell us all about it. That is Callie Huang, who covers Meta for the information. Well, it is officially getting close to back-to-school season, and Grammarly made news this morning, launching new AI agent tools for students. Part of the new product suite includes tools to make sure students are being responsible with their AI use, and Grammarly has also been trying to expand itself beyond just grammar-checking software with a series of acquisitions lately.

14:51Here to talk about all of that is CEO Shishir Mehrotra. Shishir, welcome to TITV. It's great to have you.

15:01Haha, that's a rookie mistake. Great to be here. We'll go with it. We'll go with it. Shashir, it's great to have you here. So look, it's back to school season. Tell us about the products that you launched today. Yeah, I think this is a great step forward in what we think is the agentification of Grammarly. We view Grammarly as the original, the OG AI agent, works everywhere you work, assists you in everything you do. But up till now, we've been using it only for grammar check for, I like to call it, having your high school grammar teacher available all the time. And we're working on taking that platform and broadening it much, much broader to cover many use cases beyond grammar.

15:42So we launched today, actually last Friday, a set of agents that are expanded in areas that are pretty close to what we were doing, but really stretch how you might think about Grammarly and especially focused on students that are writing. So there's an AI grader that can take whatever you're writing and give you a grade for it. There's a citation finder that can go and make sure that you're citing the right work. There's an AI detector that can help predetermine what the AI detectors at the schools are going to say about whether there's AI used in the writing or not. A paraphraser, so on. So it's a set of different agents that broaden.

16:22And then you'll see us go much, much further with other agents coming very soon. So what exactly is the agent part of this? Because I think I understand ground release. It kind of checks things as you work the way that I understand the product, at least. So what is the new agent part of this? Yeah, I mean, I think this is a little bit, as I think about the definition of agents, I think agents, people have classified them in a few different ways. I think of there as being sort of three classes of agents out there. There's agents that are really consultative. There's the agents you ask. Those are, I think, what we're used to with products like ChatGPT.

16:56They're agents that are primarily focused on taking action on your behalf. They're sort of asynchronous action performers. And then we think of there's a class of agents we think of as assistive agents. And their job is to work right next to where you work. And the least interrupted way possible, help you perform your task. And so from that perspective, we view Grammarly as one of those original agents. It works in about 500 ,000 different applications. Every web app, desktop app, mobile app you can think of, we sit right next to you. We read what you're doing. We annotate it in ways that don't disrupt you or the application.

17:31and then we make changes on your behalf. I'm just trying to understand the back end here. I mean, as somebody writes something, as it's checking, you know, assessing its plagiarism score, or I guess assessing ways in which you can improve it, is that agent then checking things in the background? Is it going and communicating with other data sets in the background? Yeah, that's right. I mean, we're one of the biggest users of LLMs out there. We make over a billion LLM calls a day. And so every time you're typing, as you're typing on every key press, we're going and running a set of agents that look at what you've done and try to come up with the best possible suggestions, mux them together into a common set of recommendations.

18:15Everything from did you get the grammar right to did you cite the right work? Did you, you know, how are people going to react to it? What's the tone? And so on. soon we'll add a set of agents that go much beyond that, connect all your different data sources, be able to do things like check it against facts in your CRM system or in your past emails or past writing, and then agents that can take that further and use that as an initiation point to take action on your behalf as well. And which models are you using to do all this? We use everything. We have a set of in-house models that we've built over time.

18:50We use a lot of every level of the LLAMA models. We use all of the publicly available models from OpenAI and Anthropic. But one thing to note about us is we have every use case of LLM. So we have cases that need to run in incredibly fast speed and very low latency. So as you type, we have others where we can take our time and we can come back with a grade for your writing or for a reaction for what people might think. And so I think we end up being kind of at the front of every, every LM you can find. I want to talk to you sort of about Grammarly's broader strategy in just a second. But last question on this point here, you know, the thing about combining AI with education for me, at least the way I think about it, if you take a step back is some of this is the work that the teacher would have probably done, you know, in terms of providing feedback for the student.

19:43And then on the other hand, some of this is the work that the student would have done insofar as doing some of the thinking with respect to what goes into the writing. And so I guess if you have a software like Grammarly that is doing some of the work of the teacher, some of the work of the student, where do you think education is going here three, four, five years from now? How is it going to change? Aren't you just doing their jobs? Yeah, I mean, I think maybe two different points there. So when we talk about the teacher side and the student side separately, On the teacher side, I do think the role of teachers has changed dramatically, not just in the last couple of years, but over the last decade.

20:19I mean, I was at Google running the YouTube team when the advent of taking all your lecture materials online became a real thing. And, you know, nowadays it's very typical that many of the classes at colleges are actually broadcast online for free. And it completely changed what, you know, what do you pay for in college? Are you paying for going to the lecture? Are you paying for something else? And I think that was a massive step forward for education, but it had to really change the definition of what a teacher does. The teacher's job is not the lecturer. The teacher's job is the assistance.

20:48It's the work with the student. I think in this case, we're seeing yet another step of that. And I think what we're going to see is professors act a little bit more like the term I use is digital twins. And we work with a lot of different universities. Most of the large ones are customers or partners of ours. And I think what we hear from them regularly is that while in some senses what's happening with the I is scary, it's also a huge opportunity to bring a teacher directly into your experience. So today you can think of Grammarly as your high school grammar teacher sitting right next to you.

21:25Right. And very effective. It really not only does it help you do your work, it helps you learn. You learn all the patterns as you go. Now, imagine instead of, you know, today you have a class, you get an assignment, three weeks later you submit the assignment, you get back feedback. What if that happened every minute? As you were working, it wasn't just your grammar teacher, your history teacher, your chemistry teacher, so on, working right alongside you. I think on the other side on students, I think that, you know, since technology started, even with calculators, we've always had a dynamic of there are tools that help students in a way that could be used to circumvent learning.

22:02And it's very important to take a balanced approach. Like, I think just because we have calculators doesn't mean we don't teach kids arithmetic. And some cases you'll take a test and say, you know, no calculators allowed. And other times you'll say, actually, in the real world, you have a calculator. And so it's pretty important for you to learn what it's like with a calculator, too. And I think it's the same thing we're seeing with AI for students is that there are cases where schools are adjusting to say, we want you to be able to learn and work without these tools. But also they're leaning into, you know, in the real world, you really need to be able to use these tools well.

22:34So these are cases where we want you to use them actually even more and learn how you can apply them in your work. So you kind of see this as the way that when I was going through high school and stuff, I get either open book tests, there are passage analysis tests, there are tests where you have a calculator, you don't have a calculator. So you see the future of education being, we will have a class for writing and creative writing, for example, but then we will have a class where it's, we're not actually focused on the writing, it's focused on the history, for example, or putting - I mean, even in the case of writing, I mean, learning how to write with AI is its own challenge.

23:11And I think the I think we'll have the same writing class will have a mode that is we want you to start from scratch. Please don't ask for any help. Let's see what you do. How well do you know your technique? How well can you handle grammar and so on? But the other case of you can use anything you like. Now, now it's really about the ideas. What are you really writing? And I think the, you know, the the like the calculator, it doesn't remove. it actually just up levels what you can accomplish now. And I think we're gonna see both patterns in the same class, I think. Right. Well, great, Shashir.

23:44We don't have time to get to the broader strategy, but that is all the more reason that you're gonna come back to the show and you're gonna talk to us about Grammarly's approach to really broaden the product suite because you just did the superhuman acquisition. I know you came from Coda as well. And so you're gonna come back on the show and we're gonna have a broader discussion about that because that is a very fascinating part of this story. That is Shashir Mirotra, the CEO of Grammarly. They introduced new AI agent tools for writing today. Okay, for our final guest, chips and semiconductors have been perhaps the biggest rollercoaster of the AI boom with supply chain crunches, geopolitical tensions, and now new entrants coming in to try to capture market share left, right, and center.

24:24Few people know this world better than our next guest who wrote a book on this exact subject even before the world started paying attention to any of this. Joining me now is Chris Miller, author of Chip Wars. Chris, it's great to have you here. Thanks for having me. So look, we're going to talk about chips, but very quickly, we just had the CEO of Grammarly on. He's saying the kids are going to start writing with AI agents and stuff like that. Are you for or against that? Very quickly. I'm for so long as we're not forgetting how to write in our own as well. I think the calculator analogy is a great one, and we're just beginning to apply it to writing.

24:58Right. Right. Have you used any AI tools with your own writing at all, with your book writing endeavors? I use it all the time for research, less for writing myself. Great, great. Okay. Well, let's put that aside. Let's go back to the fundamentals of the business. Let's talk about chips. One of the things that has really struck me about the news of the last couple of weeks with NVIDIA and their ability to sell chips in China or not is, look, this is It's less than, historically, it's been less than 15 % of their revenue coming from China. And yet, it seems like it's taking an outsized portion of Jensen Huang's attention right now.

25:35The chips that they've gotten the green light to sell are kind of this watered-down version of the chips. They've got this revenue share agreement with the U.S. government now. So my question for you is, I mean, why bother with any of this? Why even focus on it at all? I think it's a good question. I think for most tech companies, China is the world's second largest economy. for all hardware sellers, it is a big market. And so, you know, fair to note that most of NVIDIA sales go to North America or to U.S. companies with subsidiaries elsewhere in the world. But it's hard to ignore the world's second largest economy, especially since, as Jensen correctly often says, China is a center of AI innovation with lots of new open source models coming out of the region.

26:17So the company is afraid that if it does completely lose access to the Chinese market, it's also going to be cut off from future waves of innovation that might emerge from China. And what do you think about this idea that the threat of a company like Huawei, for example, getting more traction in China because of the absence of NVIDIA there, the threat that Huawei could then become the next big player? Do you think that's a legitimate concern? I think it's legitimate to be concerned about Huawei getting better and better at designing chips better and better with its partners at manufacturing them.