THIS WEEK IN AI: Grok Bot (and 4.6), AI Watermarks, Deepseek Strategy, Selling the Lakers

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Josh: There's a new Frontier model in town, and it comes from a not-so-suspecting company in SpaceX AI. (00:00):
undefined

Josh: They just released Grok 4.6. This is now possibly the best model that there (00:05):
undefined

Josh: is out there, depending on what you care about. As we look at this benchmark (00:10):
undefined

Josh: card, it shows that Grok 4.6 High is scoring higher than even Fable 5 on some tasks. (00:13):
undefined

Josh: And this is shocking because, I mean, just last week, if you would have asked (00:19):
undefined

Josh: us where Grok stood in the AI race, we would say, well, they're like a distant (00:21):
undefined

Josh: third competing probably with Meta and Gemini. (00:25):
undefined

Josh: The reality is that with this new model, they are now right at the front. (00:28):
undefined

Josh: And not only that, but the new model, Grok 4.7, is expected to come out in the (00:31):
undefined

Josh: next three to four weeks and be even better than this one. (00:35):
undefined

Josh: This comes on the back of Grok releasing a new product called GrokBot, (00:39):
undefined

Josh: which we'll get into, as well as a lot of other interesting news from this week. (00:42):
undefined

Josh: This is the roundup, the weekly roundup every Friday that we publish, (00:46):
undefined

Josh: where we cover all things about news and AI, things like covering earnings from (00:49):
undefined

Josh: NeoClouds all the way over to the new acquisition of your favorite or least (00:52):
undefined

Josh: favorite sports team. But we have to start here with Grok 4.6. (00:56):
undefined

Josh: They're back. They're right here at the frontier. This is an amazing model. (00:59):
undefined

Ejaaz: So back. Yeah, no. So for the longest time, SpaceX had kind of lost the frontier race. (01:02):
undefined

Ejaaz: Elon's engineering team had made a few wrong moves and they kind of fell behind (01:08):
undefined

Ejaaz: the likes of Anthropic OpenAI and Google, which is crazy to say because Google's (01:12):
undefined

Ejaaz: fallen behind a lot since then. (01:16):
undefined

Ejaaz: They're officially back. Grok 4.6 comes, I think, three weeks after Grok 4.5 (01:19):
undefined

Ejaaz: got released, and it is a bigger and better model. (01:25):
undefined

Ejaaz: Now, specifically, if we look at a few different benchmark comparisons, (01:28):
undefined

Ejaaz: there's this thing called the AA Intelligence Index. It's a really good way (01:33):
undefined

Ejaaz: of testing just the general level of intelligence across different models. (01:37):
undefined

Ejaaz: And on your screen here, you'll see it compared directly to GPT 5.6 Sol Max. (01:41):
undefined

Ejaaz: So that's like the highest, hardest version from OpenAI, as well as Fable 5 Max. (01:47):
undefined

Ejaaz: And you'll see it is the equivalent to GBD 5.6 Sol and just a point of Fable (01:52):
undefined

Ejaaz: 5. So you could argue that its general intelligence level is right up there. (01:58):
undefined

Ejaaz: But the best part is it comes at a fraction of the cost. So it's 5x cheaper (02:02):
undefined

Ejaaz: output than Sol, 8.5x cheaper output than Fable. (02:07):
undefined

Ejaaz: Overall, you're kind of ranging between 3 to 5x cheaper than these two models, (02:12):
undefined

Ejaaz: which is a big thing if you're like a company that's using these frontier models (02:15):
undefined

Ejaaz: and want to kind of figure out (02:19):
undefined

Ejaaz: how to cut your cost down from tens to hundreds of millions of dollars. (02:21):
undefined

Ejaaz: Now, the main way that they're able to achieve this frontier intelligence, (02:24):
undefined

Ejaaz: well, there's a few ways, but the main one is they spin up a bunch of agents. (02:28):
undefined

Ejaaz: So Grok 4.6 uses multiple agents to kind of like solve your query. (02:32):
undefined

Ejaaz: Now, this isn't something that's relatively new. (02:35):
undefined

Ejaaz: This is something that Meta is doing in their smaller models. (02:38):
undefined

Ejaaz: It's something that a lot of Chinese open source models are also doing. So that's one part. (02:41):
undefined

Ejaaz: Two, and this is like, I have to emphasize, this is SpaceX's like biggest moat. (02:45):
undefined

Ejaaz: They have a lot of data, a lot of useful data that makes these models insanely (02:50):
undefined

Ejaaz: smart. They can just bung it into it when they're training it. (02:55):
undefined

Ejaaz: And also a hell of a lot of compute, which Elon has been very strategically (02:57):
undefined

Ejaaz: amassing over the last year. He has like the most compute out of anyone in the game. (03:03):
undefined

Ejaaz: And we're starting to see this, like you mentioned, like 4.7 is coming out in (03:07):
undefined

Ejaaz: a few weeks time, and it's going to be even bigger and better than this. (03:10):
undefined

Ejaaz: I think it's going to be like a three to five trillion parameter model. (03:13):
undefined

Ejaaz: So you can only imagine that, like, you know, with the cost benefits and with (03:16):
undefined

Ejaaz: the intelligence that's increasing, like SpaceX is in a new world right now. It's very impressive. (03:20):
undefined

Josh: Yeah, this is where things are going to start to get really interesting because it was. (03:26):
undefined

Josh: Mostly a matter of if not when and now it feels like it's a matter of when when (03:29):
undefined

Josh: they actually get to their frontier and eclipse it (03:34):
undefined

Josh: for a few reasons the first of which is that i mean as we mention all the time (03:37):
undefined

Josh: the spacex team is the best team in the world at building physical things at (03:41):
undefined

Josh: moving atoms around at compiling data centers quickly (03:45):
undefined

Josh: it is nothing compared to building a rocket ship to just put a couple servers (03:49):
undefined

Josh: on a rack on the ground and plug them into electricity so they have been doing (03:52):
undefined

Josh: this quickly and efficiently in the episode yesterday we talked about micro (03:56):
undefined

Josh: hard which is the small version of macro hard (03:59):
undefined

Josh: yet it is the same exact footprint in terms of gpus in a cluster so basically (04:02):
undefined

Josh: they've able to they've been able to figure out how to get 200 000 coherent (04:07):
undefined

Josh: gpus into a package the size that's (04:10):
undefined

Josh: about one third of macro hard so it's this amazing iterative process that they (04:13):
undefined

Josh: have on hardware and then they have this amazing iterative process on software (04:16):
undefined

Josh: where they're planning to publish these models once a month (04:19):
undefined

Josh: that are incrementally stronger and stronger you said now you just i think we're (04:22):
undefined

Josh: at three to five i'm pretty sure there's plans for like eight to 10 trillion parameters (04:25):
undefined

Josh: in terms of parameter count. (04:29):
undefined

Ejaaz: They actually have that training run going right now for a 10 trillion parameter model. (04:31):
undefined

Josh: It's in the oven right now. So we are going to get that after it's done with (04:35):
undefined

Josh: the post-training alignment phase. (04:38):
undefined

Josh: But the idea is that it's all coming. It's all happening. (04:40):
undefined

Josh: The hard part is mostly done. They've gotten from zero to one. (04:43):
undefined

Josh: They've reached it back to the frontier. (04:47):
undefined

Josh: Now it's just a matter of seeing how far they could push it, how quickly. (04:48):
undefined

Josh: I'm going to guess like pretty quick. I don't know. when i think about other (04:52):
undefined

Josh: advantages like their data set like you mentioned having access to all the proprietary (04:55):
undefined

Josh: space rocket science data that seems valuable in terms of science and intelligence (04:59):
undefined

Josh: basically they have this open corpus of data that no model has ever been trained (05:04):
undefined

Josh: on that's proprietary and for (05:07):
undefined

Josh: those that aren't familiar like starship the giant rocket that's going to mars (05:09):
undefined

Josh: that was built 100 by humans there was no ai in the process so all of the math (05:13):
undefined

Josh: all of the science all of the hardware schematics (05:17):
undefined

Josh: that's incredibly valuable and that's what grok 4.6 is being trained So this (05:20):
undefined

Josh: model is an amazing model and I'm very excited to see where Grot goes from here. (05:23):
undefined

Ejaaz: Not to mention that Elon Musk has specified that by the end of the year, (05:28):
undefined

Ejaaz: Grok 5 will be released. I believe that will be the 10 trillion parameter model. (05:33):
undefined

Ejaaz: So we're talking about mythos grade level model. (05:37):
undefined

Ejaaz: But it'll be the smartest engineering AI model in the world. (05:40):
undefined

Ejaaz: And the reason why he says that is it'll be trained entirely on SpaceX's engineering data. (05:44):
undefined

Ejaaz: And there is no better physical engineering team in the world that isn't at (05:49):
undefined

Ejaaz: SpaceX. So not only is he building better generalized models, (05:54):
undefined

Ejaaz: he is building hyper-specific ones that are going to hopefully push the frontiers (05:59):
undefined

Ejaaz: of science, material sciences forward. (06:02):
undefined

Ejaaz: Now, for those haters out there that have, you know, validly pointed out that (06:04):
undefined

Ejaaz: Elon makes a lot of big promises but doesn't really fill his shoes, (06:08):
undefined

Ejaaz: this is one instance where he definitely is doing this. (06:12):
undefined

Ejaaz: And I don't think it was, it was until a few days ago, priced into the market. (06:14):
undefined

Ejaaz: But like right now, when you look at the stock, it is up 26% in the last five days. (06:18):
undefined

Ejaaz: I remember when we made our episode last week when the stock was down, I don't know, like 35%. (06:24):
undefined

Ejaaz: And we literally said that like, you know, like there's a lot of unlocks coming out. (06:29):
undefined

Ejaaz: We understand that there's a lot of supply constraints and issues here. (06:32):
undefined

Ejaaz: But if you are somewhat a mid to long-term holder, SpaceX so obviously has a (06:35):
undefined

Ejaaz: track record for execution, being able to pull feats like this off and people (06:41):
undefined

Ejaaz: just weren't pricing into the air completely. (06:45):
undefined

Ejaaz: Remember, when you buy SpaceX stock, You are just buying like a rocket company, right? (06:47):
undefined

Ejaaz: You're buying a satellite connectivity company. (06:51):
undefined

Ejaaz: You are buying the cheapest way to get into space and launch your own compute (06:55):
undefined

Ejaaz: if you wanted to harvest that. You're buying a rocket company as well. (06:59):
undefined

Ejaaz: And you're also buying an AI company. Oh, also the social media company of the (07:02):
undefined

Ejaaz: world that's going to distribute a bunch of these things. Include it. (07:06):
undefined

Ejaaz: This new product that has just come out, which we have been playing around with (07:09):
undefined

Ejaaz: called GrokBot. It's my understanding that this is basically their version of (07:13):
undefined

Ejaaz: OpenClaw, except it actually works. Do I have that right, Josh? (07:18):
undefined

Josh: It's a remarkable product. I am obsessed with GrokBot. It feels like the SpaceX (07:22):
undefined

Josh: AI team is jumping to endgame of what a agentic product could actually look like. (07:26):
undefined

Josh: Like you mentioned, it's very similar in sense to OpenClaw, but it feels like (07:31):
undefined

Josh: it's a much more improved and refined version of OpenClaw in the sense that it's fully agentic. (07:35):
undefined

Josh: And maybe I could explain the way it works. So basically you join this application (07:39):
undefined

Josh: and it just gives you a text box similar in style to iMessage. (07:42):
undefined

Josh: You could picture like where you send your text messages, except you could spin up an agent. (07:45):
undefined

Josh: You can give them a task. You can give them a personality. You give them a name (07:49):
undefined

Josh: and then they will be that person for you. So for example, if you like working (07:52):
undefined

Josh: out, you go to the gym, you create a little chat bot and you say, (07:56):
undefined

Josh: Hey, you're my gym workout companion buddy. You are a world-class expert at fitness. (07:59):
undefined

Josh: You are going to give me an entire workout program based on these 10 things (08:03):
undefined

Josh: that I know about my body and I'd like to work with you every single day to (08:07):
undefined

Josh: Actually get more fit and more healthy and remind me every day to go and work (08:11):
undefined

Josh: out. And if I don't check in with you, then remind me again and again until (08:15):
undefined

Josh: I actually get my ass to the gym and work out. (08:18):
undefined

Josh: And this is a very basic version of what it's capable of. But I think it's really (08:20):
undefined

Josh: interesting in a way that OpenClaw wasn't because OpenClaw required a bit of (08:24):
undefined

Josh: technical expertise. You need to host your own server. (08:28):
undefined

Josh: You need to download the software on the server. You need to run maintenance. (08:30):
undefined

Josh: You need to manage the memory system. You need to manage all the nuance and complexities. (08:33):
undefined

Josh: With this program all you have to do is just log in and you just start talking (08:37):
undefined

Josh: to it it hosts it on their servers the application is beautiful they've done (08:41):
undefined

Josh: this really cool thing that i love which is the design of the actual (08:45):
undefined

Josh: agents they have this very cute animation it kind of disarms the feeling that (08:48):
undefined

Josh: you're working with this like (08:52):
undefined

Josh: kind of scary spooky ai and just turns it into a really fun experience (08:54):
undefined

Josh: and it's available on mobile it's available on desktop and i like cannot encourage (08:57):
undefined

Josh: trying it out enough and you don't even realize that you're talking to a system (09:02):
undefined

Josh: similar to open claw because it's so easy to set up it's all handed on the background (09:05):
undefined

Josh: i think amazing product from grok (09:09):
undefined

Ejaaz: If you're listening to this and you're wondering okay is this a new ai model (09:11):
undefined

Ejaaz: should i be using this i tried that open call thing out and it was terrible (09:15):
undefined

Ejaaz: it stole all my money from my bank account (09:18):
undefined

Ejaaz: Let me try and explain like who this product might be pitched at, (09:21):
undefined

Ejaaz: right? So it is not the ideal product to use if you are a hardcore software (09:24):
undefined

Ejaaz: engineer, if you want to code every day. (09:30):
undefined

Ejaaz: This is pitched at sort of knowledge workers who spend a lot of time doing software (09:32):
undefined

Ejaaz: or desktop-based tasks, but want to automate some of the more. (09:37):
undefined

Ejaaz: Menial things like email admin or calendar scheduling or customer support and stuff like that. (09:42):
undefined

Ejaaz: These types of things take up a lot of time. And if you kind of like sum up (09:49):
undefined

Ejaaz: the economic value of all of it across like every single sector, it is quite a large sum. (09:53):
undefined

Ejaaz: And I think to understand why GrokBots isn't just a toy is to understand what (09:57):
undefined

Ejaaz: Elon Musk said, I think about six months ago, which is, (10:02):
undefined

Ejaaz: it is very much XAI and now SpaceX AI's mission (10:05):
undefined

Ejaaz: to replace a lot of human software-based tasks, not so that it replaces people's (10:08):
undefined

Ejaaz: jobs, but to give them the freedom to not waste their time on stuff that doesn't need to be wasted. (10:15):
undefined

Ejaaz: They can kind of like automate away all the boring manual labor that you get tired of doing at work. (10:20):
undefined

Ejaaz: And it frees you up to do more creative, impactful things in your actual job. (10:24):
undefined

Ejaaz: He references it as a human emulator. So it's basically, you can interact with (10:29):
undefined

Ejaaz: these agents and you don't even know that it's another human. (10:34):
undefined

Ejaaz: That's like the real test, the benchmark on the other end. And this is the first (10:36):
undefined

Ejaaz: step towards that. Now, is this that? (10:39):
undefined

Ejaaz: No. But is this a really refined and easy to use version like you mentioned, Josh? (10:41):
undefined

Ejaaz: Yes. You sign up. They spin up a virtual machine in the cloud and your agents (10:46):
undefined

Ejaaz: just kind of like run riot and they can like simultaneously take over your computer, (10:51):
undefined

Ejaaz: but it's done in a very safe mannered way so that they can't do anything crazy. (10:54):
undefined

Ejaaz: And the thing that pops to the top of my head when I read through and watch (10:59):
undefined

Ejaaz: all the demos of this problem and read the feedback is... (11:04):
undefined

Ejaaz: This is a direct attack against not OpenAI or Anthropik, but Apple, (11:07):
undefined

Ejaaz: who is releasing Siri AI in a few months' time now and is meant to function (11:12):
undefined

Ejaaz: and perform completely on your phone. (11:17):
undefined

Ejaaz: It's meant to interact with all your apps and all that kinds of stuff. (11:20):
undefined

Ejaaz: Now, granted, GrokBot can't do that right now, but it is accessible and available on your phone. (11:22):
undefined

Ejaaz: You can text it whenever you want. And I can only imagine that eventually this (11:26):
undefined

Ejaaz: will get access to some types of phones, maybe Android first because the Apple ecosystem is so closed. (11:31):
undefined

Ejaaz: But I see this as a direct competitor to that. And I'm just so happy to see (11:35):
undefined

Ejaaz: Elon make a functional product that isn't just a model because that's why he's best at. Yeah. (11:39):
undefined

Josh: And at the very least, you at least you get the app on your phone and the new (11:42):
undefined

Josh: Siri is actually launching in about a month from today. So it's coming very soon. (11:45):
undefined

Josh: But this feels like the magic that I felt with Open Claw, except a more refined (11:49):
undefined

Josh: and far more productive version. (11:54):
undefined

Josh: Currently, it's baked in with GPT 4.6, which we just said is a remarkable model (11:55):
undefined

Josh: for incredibly low price. (12:00):
undefined

Josh: As these models continue to improve having those baked into this agentic experience (12:01):
undefined

Josh: feels pretty incredible (12:07):
undefined

Josh: it's like you open up your phone out of your pocket and you have this series of brilliant (12:09):
undefined

Josh: supercomputer helpers that can do anything in the world for you they can work (12:13):
undefined

Josh: together they can gain access to your context it feels like a peek into the (12:16):
undefined

Josh: future this feels like open claw 2.0 so it's a really exciting product i think (12:19):
undefined

Josh: we're going to follow this one (12:23):
undefined

Josh: closely and see what the other company's reactions to a product like this (12:24):
undefined

Josh: are going to be but that is the grok update congratulations on your new Frontier slot. (12:27):
undefined

Josh: Looking forward to the next one. For the next topic, we have another, (12:31):
undefined

Josh: I mean, kind of Frontier model, except this one on the open source side. (12:35):
undefined

Josh: DeepSeek is back with another release as well. DeepSeek v4.0. (12:38):
undefined

Josh: Pro. And the price floor has moved once again. This is a pretty big deal. (12:43):
undefined

Ejaaz: Yeah. So DeepSeek, if we kind of harken back to, what is it, (12:48):
undefined

Ejaaz: like a year and a half now, which is not that long ago at all, (12:52):
undefined

Ejaaz: caused a massive disruption in the AI field because back then, compute was king. (12:56):
undefined

Ejaaz: Sam Altman was on top. And if you could buy as much compute as you can, (13:02):
undefined

Ejaaz: you could train the smartest models. (13:05):
undefined

Ejaaz: DeepSeek came along, which is, I believe, birthed out of a hedge fund, Liang Wangfeng. (13:06):
undefined

Ejaaz: Yeah, he's a hedge fund manager in China and basically said, (13:11):
undefined

Ejaaz: hey, yeah, I ran a bunch of quants and asked them to make a model that would (13:14):
undefined

Ejaaz: help it with financial trading and ended up creating a groundbreaking model (13:18):
undefined

Ejaaz: that didn't require as much compute. And it kind of like tore the market apart. (13:21):
undefined

Ejaaz: Well, since then, they've released, I think, one more model. (13:24):
undefined

Ejaaz: And now they're back with like their big boy model. Now, this isn't an official (13:27):
undefined

Ejaaz: release. This has been rumored to be released soon, but we had an official leak (13:31):
undefined

Ejaaz: about the benchmarks for this model, DeepSeek V4 Pro. (13:35):
undefined

Ejaaz: Now, yes, it is re-rated the cost sheet as always. It is always the cheapest (13:39):
undefined

Ejaaz: cost per unit intelligence for an AI lab out there. (13:43):
undefined

Ejaaz: But what struck me was some of these benchmarks that were leaked. (13:47):
undefined

Ejaaz: Now, if you look at Terminal Bench 2.1, which is a popular coding paradigm or (13:50):
undefined

Ejaaz: benchmark, it comes in at 87.9, right? (13:55):
undefined

Ejaaz: Now, when you compare that directly to, say, Fable 5, it is, (13:58):
undefined

Ejaaz: what is that, 0.1 of Fable 5's level of coding capability, measured at this benchmark. (14:02):
undefined

Ejaaz: If you look at Opus 4.8, Kimi K3, and GLM Fiber 2 absolutely crushes it. (14:06):
undefined

Ejaaz: But the other worrying part is this one over here. (14:11):
undefined

Ejaaz: It's the CyberGym benchmark. (14:15):
undefined

Ejaaz: It is not only rivaled, but it's slightly better than Fable 5 at cyber capabilities. (14:17):
undefined

Ejaaz: Now, the reason why this worries me is Dario had a really good post. (14:23):
undefined

Ejaaz: I think about a month ago, where he said, in about six months time, (14:27):
undefined

Ejaaz: we are going to have an open source version of Mythos' level of cyber capability. (14:31):
undefined

Ejaaz: And that's going to be an issue because it's going to have no safeguards around it. So. (14:36):
undefined

Ejaaz: The reason why Mythos hasn't been able to wreak havoc on the world is because (14:41):
undefined

Ejaaz: they have safeguards around it. (14:44):
undefined

Ejaaz: DeepSeek, if they release this in an open source fashion, is going to have, (14:46):
undefined

Ejaaz: you know, a bunch of amazing effects where you can run it at home, (14:49):
undefined

Ejaaz: it's cheaper, private, blah, blah, blah. (14:52):
undefined

Ejaaz: But also, if it was put in the hands of a malicious actor, it could end up causing (14:54):
undefined

Ejaaz: havoc. So I just wanted to point that out. Like, it isn't an official release (14:58):
undefined

Ejaaz: just yet, but it's something that I'm keeping my eye on. (15:01):
undefined

Josh: It's coming. Well, we did have an official release, and it came in the form (15:04):
undefined

Josh: of an earnings release. In fact, we had two of them. (15:06):
undefined

Josh: Uh one from core weave and the other from nebius for those not familiar these (15:09):
undefined

Josh: are neocloud providers neocloud providers they're basically (15:13):
undefined

Josh: they they work with the gpu providers to collect gpus they build their own infrastructure (15:17):
undefined

Josh: and they rent out that infrastructure to people who want to train their ai models (15:22):
undefined

Josh: or use them for inference or whatever it may be so you could think of it as (15:25):
undefined

Josh: like ai training or ai inference as a service they offer to people (15:28):
undefined

Josh: they had a pretty unbelievable quarter both of these companies when we look (15:32):
undefined

Josh: at nebius how much was it up on the day i'm not seeing it on the chart but i (15:36):
undefined

Josh: know it was up like a huge amount i think it was up (15:39):
undefined

Ejaaz: 25 on the day. (15:41):
undefined

Josh: Yeah and core weave i believe was up 31 at least this week yes both of these (15:43):
undefined

Josh: companies showed blowout numbers mostly because um (15:48):
undefined

Josh: as we mentioned in the previous episode the price per rental hour of gpus is (15:52):
undefined

Josh: just going through the roof everyone when they initially evaluated these neocloud (15:56):
undefined

Josh: providers they were baking in depreciation of these gpus over time (15:59):
undefined

Josh: because when you have a gpu you assume it's a depreciating asset, (16:03):
undefined

Josh: you mark it down year over year until it reaches a settling point. (16:06):
undefined

Josh: What's happening is the inverse. The number's going up. So therefore, (16:10):
undefined

Josh: the amount of revenue that they're making is going up. And from the Corby Verning's (16:13):
undefined

Josh: call, we have a quote here. It says, we recently signed an A100 contract. (16:16):
undefined

Josh: Again, those are the 2020 GPUs. (16:20):
undefined

Josh: That extends into 2029 GPUs. (16:22):
undefined

Josh: At an attractive price. So people are signing contracts for three years out on GPUs that are 25% (16:25):
undefined

Ejaaz: Higher than they rented out two years ago, by the way. (16:30):
undefined

Josh: That's incredible. So on the surface, I mean, that looks, that looks great. (16:33):
undefined

Josh: We're like, okay, this is a good company. But then also I do have to, (16:37):
undefined

Josh: I do have to say, I was looking into the earnings, particularly of CoreWeave, (16:40):
undefined

Josh: which I'm more familiar with. (16:43):
undefined

Josh: And I realized that they actually have a bit of an issue on their hands. (16:44):
undefined

Josh: They're spending a lot of money that they don't have. So one of the things that (16:49):
undefined

Josh: we look at when we look at the hyperscalers in particular, like Google and Amazon, (16:52):
undefined

Josh: is they're spending a lot of money out of their free cash flow. (16:56):
undefined

Josh: They're not going into debt. (16:58):
undefined

Josh: What's happening with Corbiv at least is they have $6.5 billion of capex, (17:00):
undefined

Josh: but they have a negative free cash flow of just under $6 billion, (17:03):
undefined

Josh: which means they're losing money. (17:07):
undefined

Josh: They have debt around $35 billion of it. And this is a company, (17:08):
undefined

Josh: when I see that, I'm like, oh, wait, that's the thing we've been looking out for. (17:13):
undefined

Josh: The hyperscalers don't have that because they're making enough revenue in order (17:16):
undefined

Josh: to offset that. But this company, (17:19):
undefined

Josh: I don't know. I just noticed that. And I was like, huh, okay. (17:21):
undefined

Josh: A little flag went off. I'm like, okay, great earnings. (17:24):
undefined

Josh: Number one up. But we're spending a little bit more than we can afford right now. (17:26):
undefined

Ejaaz: But they also have a revenue backlog of $104 billion. (17:30):
undefined

Ejaaz: And they just signed $25 billion in customer commitments, which I think about (17:34):
undefined

Ejaaz: 50 to 60% of that is paid up front, just for Q3 alone. (17:39):
undefined

Ejaaz: So you're definitely right to call out the debt backlog. I saw Martin Shkreli, (17:44):
undefined

Ejaaz: our financial analyst provocateur make the point around that they're in a lot of debt. (17:48):
undefined

Ejaaz: But I think this is just the name of the game over the next couple of years (17:53):
undefined

Ejaaz: where a lot of hyperscalers, especially these neoclouds who have to compete (17:57):
undefined

Ejaaz: with the big boys like Microsoft and Google and AWS. (18:01):
undefined

Ejaaz: Need to double down on AI CapEx if they want any shot of winning. (18:04):
undefined

Ejaaz: And if you look over at the Nebius kind of breakdown, by the way, (18:09):
undefined

Ejaaz: we should get into the revenue numbers just very briefly, like their revenue (18:13):
undefined

Ejaaz: is up 454% year upon year. (18:18):
undefined

Ejaaz: That is an insane business. And that is not too indifferent quarter to quarter (18:21):
undefined

Ejaaz: as well. I think they're compounding by like 50%, which is just absolutely insane. (18:27):
undefined

Ejaaz: But the point around Nebius is this is a more, shall I say, grown up version of CoreWeave. (18:30):
undefined

Ejaaz: And they're sitting on about $8 billion in cash. So I think of like CoreWeave (18:36):
undefined

Ejaaz: being a younger gun that's like really eager and might end up flying too close (18:41):
undefined

Ejaaz: to the sun, but they do have the backing of NVIDIA. (18:45):
undefined

Ejaaz: They do have contracts with Anthropic, I believe, which essentially prop up (18:48):
undefined

Ejaaz: that revenue backlog that I mentioned earlier on. (18:52):
undefined

Ejaaz: And I just want to explain something that separates what these neoclouds are (18:55):
undefined

Ejaaz: versus like some of these hyperscalers. When you look at Google, (18:59):
undefined

Ejaaz: when you look at AWS, when you look at Microsoft, they've been in the cloud game for a while. (19:02):
undefined

Ejaaz: So when this AI stuff came along, they basically looked at it and said. (19:06):
undefined

Ejaaz: Oh, cool. I'm squeezing 30 to 40% margins on normal cloud. (19:09):
undefined

Ejaaz: I should be able to do the same with AI, right? (19:14):
undefined

Ejaaz: So they go to Jensen and they buy all these GPUs and they realize, (19:16):
undefined

Ejaaz: oh, Jensen's taking 70% margin, so I have to take less... (19:20):
undefined

Ejaaz: Percentage on this. So they become a little more difficult to work with. (19:24):
undefined

Ejaaz: When you're a company that wants to least compute for some of these hyperscalers, (19:28):
undefined

Ejaaz: they don't do anything and everything for you. (19:30):
undefined

Ejaaz: Compare that to Nebius and CoreWeave, they will go out to build a specific cluster (19:33):
undefined

Ejaaz: that works for your specific workload. (19:37):
undefined

Ejaaz: And that's the difference between these two companies or these two types of (19:40):
undefined

Ejaaz: companies and why I think Nebius and CoreWeave are actually going to do pretty well in the future. (19:43):
undefined

Josh: Yeah, I hope so. We'll see. We also had news from another company similar in (19:46):
undefined

Josh: this space with Intel, who was saying, they're like, hey, actually, (19:50):
undefined

Josh: we kind of need some more money. Can we get $15 billion? (19:54):
undefined

Josh: And they asked for $15 billion and the market kind of hated it. (19:57):
undefined

Josh: The stock fell 5% after announcing that they are doing a $15 billion common (20:00):
undefined

Josh: stock issuance to help fund the AI build out and growth opportunities. (20:05):
undefined

Josh: Market was like, ooh, we don't like that. Again, to the CapEx thing. (20:09):
undefined

Josh: Everyone's allergic to CapEx for some reason. They think that these companies (20:12):
undefined

Josh: are not good at allocating money. (20:15):
undefined

Josh: Clearly, they're wrong because, I mean, in the last day, Intel has now bounced (20:17):
undefined

Josh: back. They're doing fairly well. (20:20):
undefined

Josh: But what is the story behind this reaction and the kind of knee-jerk reaction (20:22):
undefined

Josh: in which initially they were a little upset? But then it seems like today the (20:26):
undefined

Josh: market's kind of coming to its senses and figuring, like, wait a second, maybe we do like CapEx. (20:30):
undefined

Ejaaz: Well, people obviously do like it, but specific people, which are the investors (20:35):
undefined

Ejaaz: in Intel. So they originally asked for $15 billion, but they ended up getting (20:40):
undefined

Ejaaz: demand for $100 billion. (20:45):
undefined

Ejaaz: And they thought, okay, well, we won't take $100 billion worth of equity sell, (20:47):
undefined

Ejaaz: but we'll take an extra $5 billion. So they ended up raising $20 billion in (20:51):
undefined

Ejaaz: under 24 hours, might I add. (20:55):
undefined

Ejaaz: Now, you might be wondering, why on earth do they need more money? (20:58):
undefined

Ejaaz: Why are they selling equity? Why are they diluting shareholders? (21:01):
undefined

Ejaaz: Well, number one, the dilution isn't actually that much, who cares? (21:04):
undefined

Ejaaz: But if you listen to their last earnings call, Li Bhutan, who is the CEO, (21:07):
undefined

Ejaaz: who has been changing things around a lot for Intel, as well as the CFO said (21:12):
undefined

Ejaaz: something very specific, and I'll quote it. (21:16):
undefined

Ejaaz: He goes, things are going really well, and we have almost $20 to $30 billion. (21:18):
undefined

Ejaaz: Cash balance that we can spend on anything and everything. (21:24):
undefined

Ejaaz: However, if things are going really well, we may raise external capital. (21:27):
undefined

Ejaaz: And what he was referring to there is if they have a lot of signed contracts (21:33):
undefined

Ejaaz: and upcoming customers, big customers that want to use their foundry. (21:38):
undefined

Ejaaz: There's something called Intel Foundry where Intel is basically building a chip (21:43):
undefined

Ejaaz: fab that is going to go up against NVIDIA and the likes of AMD as well. (21:46):
undefined

Ejaaz: Then they will raise external capital. This is the clearest signal that they have those customers. (21:50):
undefined

Ejaaz: So I have a feeling over the next couple of months, we're going to see an announcement (21:55):
undefined

Ejaaz: from Intel saying that they've signed a major client like, I don't know, (21:58):
undefined

Ejaaz: an Apple or an Anthropic or OpenAI, basically saying, yeah, we're going to be (22:02):
undefined

Ejaaz: designing custom chips for them, which I think is a big deal. (22:06):
undefined

Ejaaz: So I'm happy to see Intel winning and I'm a pretty big fan. (22:09):
undefined

Josh: Yeah, well, part of the reason why they're needing this build out is because, (22:12):
undefined

Josh: I mean, again, we talk about the increased amount of inference that has been happening. (22:14):
undefined

Josh: There's just so many inference tokens required because of agents and if you're (22:18):
undefined

Josh: building with agents we have a message for you from our friends and sponsors (22:22):
undefined

Josh: over at ledger because oftentimes if you're building with agents you're probably (22:25):
undefined

Josh: worried about security we've seen all these breakouts that have happened recently (22:29):
undefined

Josh: in the space it's kind of scary ledger has a solution to this it comes in the (22:32):
undefined

Josh: form of three parts where the agents propose the humans approve and then the ledger signers enforce (22:35):
undefined

Josh: this is the three-step process to ensure that you are safe when you're working (22:40):
undefined

Josh: with agents ledger has this thing called an agent stack it's totally open source (22:43):
undefined

Josh: it comes equipped with tools to help you manage the way that you work with your agents. (22:46):
undefined

Josh: It works with cloud code, codex, cursor, anywhere you really interface with (22:50):
undefined

Josh: AI, and it's totally open source. (22:54):
undefined

Josh: So you can check it out in the link in the description below. (22:55):
undefined

Josh: Thank you so much to Ledger for sponsoring this episode. Now we have to move (22:58):
undefined

Josh: over to watermarking, which is apparently a thing that's been happening now. (23:01):
undefined

Josh: What's the watermarking story here? (23:04):
undefined

Ejaaz: I'm still trying to figure it out myself, but basically our friends across the (23:07):
undefined

Ejaaz: pond in the EU, they like regulating stuff, especially tech and especially AI. (23:11):
undefined

Ejaaz: And their latest law has basically required the top AI labs to watermark the (23:17):
undefined

Ejaaz: text output or the response that your AI chatbot gives you when you write to it. (23:24):
undefined

Ejaaz: So let's say you ask Anthropik or Claude specifically, hey, like, (23:30):
undefined

Ejaaz: can you teach me how to do A, B, and C? Or let's say you ask it to do a bit (23:35):
undefined

Ejaaz: of work at your job, right? Like help you generate a structure for a document. (23:39):
undefined

Ejaaz: That text that it generates will have an embedded watermark that can be traced (23:43):
undefined

Ejaaz: back to being Claude. So let's say you copy that text and you paste it into your work document. (23:49):
undefined

Ejaaz: And eventually your manager is like, oh, I'm going to run this through a detection (23:55):
undefined

Ejaaz: system just to see if this was written by Claude. (23:59):
undefined

Ejaaz: By the way, even if you edit it, it'll still work. So he runs it through Claude (24:01):
undefined

Ejaaz: or runs it through Anthropics New Thing. (24:05):
undefined

Ejaaz: And apparently it'll become flagged as written by Claude or AI specifically. (24:06):
undefined

Ejaaz: And the EU's justification behind this is essentially, we want to make sure (24:11):
undefined

Ejaaz: that we protect users from AI generated content and be able to distinguish between (24:15):
undefined

Ejaaz: human generated stuff and AI generated stuff. (24:20):
undefined

Ejaaz: I have a bunch of questions around this being number one, is this official? (24:23):
undefined

Ejaaz: Number two, why is an EU act being applied to the entire world? (24:28):
undefined

Ejaaz: By the way, this will be applied to not just EU countries, but models anywhere (24:31):
undefined

Ejaaz: and everywhere in the world. I don't really know what to think of this. (24:35):
undefined

Josh: Yeah, the EU has this really funny thing where like sometimes they ruin the (24:37):
undefined

Josh: world sometimes they help the world and there's a funny example that i love (24:41):
undefined

Josh: to reference here which is like the iphone (24:43):
undefined

Josh: um the iphone had this proprietary cable which was lightning and it was kind (24:46):
undefined

Josh: of annoying because it was the only device that used a lightning adapter it (24:49):
undefined

Josh: was proprietary it only worked for apple (24:52):
undefined

Josh: The EU was being being annoying as they are. They said, no, no, (24:54):
undefined

Josh: no, no, no. You must use a generic port. You must use a port that everybody (24:56):
undefined

Josh: else is using. That's not proprietary. (25:00):
undefined

Josh: So Apple had to accelerate the switch over to USB-C. And it turns out that worked (25:02):
undefined

Josh: out well for everyone. We enjoyed that. (25:07):
undefined

Josh: This seems like a different case. I'm not really sure. I think we could kind (25:09):
undefined

Josh: of stay tuned for more information on how this is going to roll out and what (25:11):
undefined

Josh: this will actually look like. (25:14):
undefined

Josh: But an interesting development nonetheless. There's another interesting development (25:15):
undefined

Josh: that I actually I added to this agenda intentionally because I needed to talk about it today. (25:18):
undefined

Josh: Not because i care about the la lakers but we care i care about the people who (25:22):
undefined

Josh: bought the la lakers so for those who aren't familiar (25:26):
undefined

Josh: the la lakers was sold last year for about 10 billion dollars i believe it was (25:28):
undefined

Josh: flipped in the matter of 12 months to a new owner who just purchased the stake (25:33):
undefined

Josh: for 12.5 billion dollars the most expensive sale in american history (25:37):
undefined

Josh: there are two people behind this purchase the first one (25:42):
undefined

Josh: His name is Bob Iger. You might know him as the dude who acquired Marvel, (25:44):
undefined

Josh: Lucasfilms, 21st Century Fox. (25:48):
undefined

Josh: That's like Star Wars, Iron Man, Avengers. This is the movie guy. (25:51):
undefined

Josh: This is the CEO of Disney. (25:54):
undefined

Josh: He partnered up with this other guy named Joshua Kushner, who you may also have heard of as the (25:55):
undefined

Josh: Perhaps the single largest, if not one of the largest investors in OpenAI and (26:02):
undefined

Josh: one of the most active investors in the company to which Sam Altman frequently (26:06):
undefined

Josh: references him as the flagship person who you want to invest on a cap table. (26:10):
undefined

Josh: I find this to be really interesting because this very much feels like the barbell (26:14):
undefined

Josh: strategy in which you have a fund, Thrive Capital being the name of it, (26:18):
undefined

Josh: which is fully invested in the frontier. (26:21):
undefined

Josh: They're investing in AI, they're investing in all this robotics infrastructure, (26:24):
undefined

Josh: all these things that are coming down the pipeline. (26:27):
undefined

Josh: And then they're also investing in what you could imagine is a post-EGI world. (26:28):
undefined

Josh: What things actually matter, what things go up in value when AI can automate (26:32):
undefined

Josh: everything. Well, I mean, people aren't going to stop watching basketball, right? (26:36):
undefined

Josh: And that's uniquely human. We're not going to get humanoids involved in that. (26:39):
undefined

Josh: So they're looking to purchase assets that are durable post-AGI and I just found (26:42):
undefined

Josh: this to be a really interesting story of if anything a thought experiment when (26:47):
undefined

Josh: you think about the world's after we have artificial super intelligence we have (26:50):
undefined

Josh: humanoid robots walking around everywhere we have (26:54):
undefined

Josh: robotics at scale what are the things that still matter to us as humans that (26:56):
undefined

Josh: we're going to continue to spend capital on it's probably something like these (27:00):
undefined

Josh: human experiences and the LA Lakers are a pretty sweet brand so they dropped (27:02):
undefined

Josh: a bag on securing this and I just thought it was really interesting and worth mentioning. (27:06):
undefined

Josh: And just like an interesting thought experiment on like, hey, (27:10):
undefined

Josh: where's the world going to go once we figure all these hard problems out? (27:12):
undefined

Ejaaz: Yeah. Largest transaction for a sporting franchise ever. That is a large amount of money. (27:15):
undefined

Ejaaz: I'm big fans of Thrive and I like the way that Josh thinks of a lot of his investments. (27:23):
undefined

Ejaaz: But yeah, to your point, I think this is a very intentional bet on humanity, (27:28):
undefined

Ejaaz: which is very refreshing to see from one of the largest AI investors in the world. (27:34):
undefined

Ejaaz: And I think that, you know, as time progresses, human created or human authenticated (27:39):
undefined

Ejaaz: IP is going to become incredibly valuable. And sporting is a. (27:44):
undefined

Ejaaz: Just happens to be one of those franchises where, you know, that condenses pretty (27:48):
undefined

Ejaaz: strongly. And I think, you know, it's just also pretty dope to just buy your (27:52):
undefined

Ejaaz: favorite team. I think that's a freaking epic, epic baller trade for them. (27:55):
undefined

Ejaaz: And yeah, so to round off the docket, speaking of OpenAI, they're churning through, (28:01):
undefined

Ejaaz: these employees, man. But just to round this off, Brad Lightcap, (28:06):
undefined

Ejaaz: who joined, I believe, as their head of growth and then ended up becoming the COO, has left, (28:10):
undefined

Ejaaz: Brad Lightcap was pretty definitive to a lot of OpenAI's growth stage, (28:16):
undefined

Ejaaz: specifically from like GPT-3 onwards. (28:21):
undefined

Ejaaz: He's been toiling away, but now he's left to go do bigger and better things. (28:23):
undefined

Ejaaz: We don't actually know what he's doing. But the point is, OpenAI has seen quite (28:26):
undefined

Ejaaz: a lot of churn over the last couple of months. (28:30):
undefined

Ejaaz: It happens to come around a time where they're rumored to basically be IPO-ing. (28:33):
undefined

Ejaaz: When I look over Anthropic, I see the opposite. (28:38):
undefined

Ejaaz: They're like a magnet attracting a bunch of people. And I do wonder what the (28:41):
undefined

Ejaaz: differences between them are actually. (28:44):
undefined

Ejaaz: But that being said, it doesn't seem to be having any change structurally. (28:46):
undefined

Ejaaz: The research team is still pounding along and creating better models. (28:49):
undefined

Ejaaz: But it's just interesting to see. And you know, Brad, if you're watching this, (28:52):
undefined

Ejaaz: dude, like, thanks for everything, dude, you've absolutely killed it. (28:55):
undefined

Josh: So yeah, pretty legendary run. Yeah. Open AI seems to have this remarkable rate (28:58):
undefined

Josh: of turn where a lot of top leadership leaves as a lot of very impressive people come in. (29:02):
undefined

Josh: And I think it's interesting and just noteworthy. I'm not sure what that implies, (29:07):
undefined

Josh: but a lot of people want to work there. A lot of people are going there. (29:11):
undefined

Josh: I guess it's testament to how intense it is to work there, but noteworthy nonetheless. (29:13):
undefined

Josh: Thank you for your services. My dear Brad, you have made a meaningful impact (29:18):
undefined

Josh: on the world through OpenAI and I'm sure whatever is coming next. (29:21):
undefined

Josh: Sam Altman said that he looks forward to working together. So whatever this (29:25):
undefined

Josh: new thing is, I suspect they're going to be doing it as some form of a partner. We'll see TBD. (29:27):
undefined

Josh: But that is the roundup. This is a fairly long one we had a lot to cover this (29:33):
undefined

Josh: week a lot of cool and exciting news and it's funny it felt like this was a down week (29:36):
undefined

Josh: this week felt kind of like a lull there wasn't that much going on and yet still (29:40):
undefined

Josh: we are here talking about so many things that are happening every single day (29:43):
undefined

Josh: i think the highlights this week really i mean grok coming out with that new (29:47):
undefined

Josh: model was pretty big meta releasing the open source models were pretty big (29:50):
undefined

Josh: we had the full diagnosis on the exploit that happened (29:54):
undefined

Josh: with open ai and breaking through hugging face you could check out all those (29:58):
undefined

Josh: videos we filmed entire episodes about that if you haven't made it through. (30:02):
undefined

Josh: But if you have, and you've reached this point of the video, (30:05):
undefined

Josh: made it through another week, congratulations. Thanks for sticking with us. (30:07):
undefined

Josh: It was a really good one. Really exciting. You are now fully up to date. (30:10):
undefined

Josh: You can go and touch some grass. Enjoy your weekend. (30:13):
undefined

Josh: Maybe look at the sun, but not for too long. The eclipse happened yesterday (30:15):
undefined

Josh: at the time of recording. It looked amazing. I went to the last two full eclipses (30:18):
undefined

Josh: and I missed this one. And I have so much FOMO. So if anyone was there, (30:22):
undefined

Josh: I'd love to hear about your experience. (30:25):
undefined

Ejaaz: What do you mean you went to the last two eclipses? Like you traveled to a country (30:26):
undefined

Ejaaz: where you could see them? (30:30):
undefined

Josh: Yeah, the United States of America, dog. It was great. (30:31):
undefined

Josh: Tennessee. And then I forget where the second one was a couple of years ago. (30:34):
undefined

Josh: But they're amazing. They're unbelievable experiences. And another one just (30:37):
undefined

Josh: happened. It covered a lot of Spain. (30:40):
undefined

Ejaaz: Wait, Josh, I thought they were every 90 years. this wasn't the last one like 1999. (30:42):
undefined

Josh: No dude i saw the last one like a couple of years ago right right like it was (30:48):
undefined

Josh: like a four-hour drive away from you wait (30:51):
undefined

Ejaaz: Wait let's let's track this up so that sounds epic i mean it. (30:52):
undefined

Josh: Must be it's just like unbelievable experience similar i guess you could kind (30:57):
undefined

Josh: of make it a comparison to the world cup in a way where it's like this unifying experience that feels (31:00):
undefined

Josh: credibly neutral like sure the world cup everyone has like their favorites but (31:06):
undefined

Josh: it's this unifying thing where you experience this together and you're like (31:10):
undefined

Josh: holy shit like we're actually (31:13):
undefined

Josh: we're pretty small beings in this universe we don't mean a lot there's like (31:14):
undefined

Josh: really powerful things happening and it's it's something that's totally out (31:17):
undefined

Josh: of our control that is like remarkably beautiful and unique and rare (31:20):
undefined

Josh: and i could see why if people like who didn't know what this was (31:24):
undefined

Josh: had this happened to them you'd probably be pretty freaked out you'd probably (31:28):
undefined

Josh: believe in aliens or whatever crazy stuff because like it that's what it feels (31:31):
undefined

Josh: like um so next one next we'll find where the next one is we'll put on our calendars (31:34):
undefined

Josh: we'll book a flight and we'll go because it's amazing but that's why I just (31:38):
undefined

Josh: had a little bit of fun with that. (31:41):
undefined

Ejaaz: I saw some insane photography and videography about people seeing this for the first time. (31:42):
undefined

Ejaaz: And also there was this awesome video of a guy skateboarding off a half ramp (31:48):
undefined

Ejaaz: into the eclipse sun as it was setting. It was just the most insane thing ever. Everything was dark. (31:52):
undefined

Ejaaz: And I couldn't help but think, (31:59):
undefined

Ejaaz: is this AI. And so I'll leave you on this note. That was real. (32:02):
undefined

Josh: Wait, also one cool thing about, like, because it mostly happened in Spain, (32:05):
undefined

Josh: in Mallorca, the sun was going down right at the time of the eclipse happening. (32:09):
undefined

Josh: And when the sun is closest to the horizon, it's largest. So it looked like (32:14):
undefined

Josh: this huge eclipsing sun. And I think that's the video you're referencing of (32:17):
undefined

Josh: this, like with the telephoto lens, you can get some pretty unbelievable shots. (32:20):
undefined

Ejaaz: It looked like the end of the world, honestly. It was pretty insane. (32:23):
undefined

Josh: So cool. (32:27):
undefined

Ejaaz: But yeah, that is it. That is the roundup. Thank you so much for listening. (32:27):
undefined

Ejaaz: If you are listening to us, if you're watching this and you aren't subscribed (32:30):
undefined

Ejaaz: or you don't have notifications on, please do. It helps us out massively. (32:34):
undefined

Ejaaz: Leave us a comment, DM us. (32:37):
undefined

Ejaaz: A bunch of you have been saying hi in IRL as well. So we love that. (32:39):
undefined

Ejaaz: Keep approaching us. Keep saying hello. We'd love to hear from you. (32:43):
undefined

Ejaaz: And yeah, that is it. We will see you next week. Back again on the list. (32:45):
undefined

All Episodes

Episode Transcript

Popular Podcasts

Stuff You Should Know

Dateline NBC

The Breakfast Club

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}THIS WEEK IN AI: Grok Bot (and 4.6), AI Watermarks, Deepseek Strategy, Selling the Lakers

Episode Transcript

Popular Podcasts

.css-r6mb8g{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:1;overflow:hidden;}Stuff You Should Know

Dateline NBC

The Breakfast Club

All Episodes

THIS WEEK IN AI: Grok Bot (and 4.6), AI Watermarks, Deepseek Strategy, Selling the Lakers

Stuff You Should Know