AMD, Starcloud, Coatue..10 Hot Takes From The Biggest Names in AI

[00:00] Raised. Raised. Raised. Raised. Paris. Paris. Paris. Paris. Paris is beautiful, but it's hot as... Paris in the middle of the summer, middle of the heat. People are ignoring it. They're showing up. [00:16] Okay, we have a very special guest. I know I've said that before, but we really do this time. Coming off the heels... [00:22] of the Jamin interview at KOTU, we have Max Cook. [00:27] of CO2. [00:28] Max, welcome. Thank you. Excited to be here. It's pretty nice. It is very nice in this very secret location in Europe, which we cannot talk about.

Not disclose the official location, but... [00:40] You've got an old cement wall, which you have many of in Europe, so it could be anywhere. It could be anywhere. It could be a country that Philippe and Thomas know very well. True. That is true. Cannot confirm, though. No. Okay, so... [00:51] We have to get into this. I think you have a couple of... [00:54] big insights you want to share with us today? - I hope they're insights, but yeah. I mean, I have a couple of things I've been thinking about. - What's your hottest take?

[01:03] Hada's take might be [01:06] that the keyboard and mouse-- [01:08] are slowly dying. [01:09] I think. Damn. [01:11] I think. [01:12] CO2, you've heard Jamin speak about it, you've heard Philippe and Thomas speak about it. [01:15] We talk about [01:17] big tech waves, you went from mainframe to PCs, to networking, the PCs, to [01:23] desktop internet which is where you get a mouse and keyboard to [01:25] mobile internet, which is where the iPhone supremacy began, [01:29] And with each of these big waves, [01:32] we end up interacting with computers in a different way.

[01:36] And it seems pretty clear at this point [01:38] I think... [01:39] there's a good stat that like [01:41] the average person has 60 to 80 apps on their phone. [01:44] Wow. They use... [01:46] 11 apps daily on average and 30 apps in any given month. [01:50] It seems quite clear that we're going from a world in which you have 60 to 80 apps, [01:54] to an agentic AI world. [01:56] where [01:57] you've got a lot fewer than that. Two, three, four apps where everything else plugs in as an API or an MCP.

[02:02] And in that world, [02:04] it would only make sense that we reinvent how we interact with compute. [02:08] It's not the keyboard and the mouse. [02:10] natural language, like [02:12] If we're talking, we speak in natural language. If you're working with a coworker, [02:16] You ask them what to do, you figure it out together, and you go do it. [02:19] So... [02:20] I think there's been a number of good [02:23] discussions about this. [02:24] uh [02:25] Michele from Replit was talking about how [02:28] Natural language is how people communicate.

That's clearly the direction we're going in. There's a Greg Brockman quote that I always think about. I'll butcher it, but he had a [02:38] We've always conformed ourselves to computers, like, by inputting specific data with a keyboard and a mouse. [02:44] and now we're finally entering the era [02:46] where computers have to conform to us. And so [02:49] this microphone is reminding me. We're already seeing it at Code 2, like in the... [02:54] famous KOTU bullpen that you got to see. We've got a number of investors that [02:59] are whispering into their collar half the day.

Because that's, [03:03] how to give models better context, better prompting, and ultimately produces much better work output. [03:10] That's my take. The keyboard and mouse, you should throw them away, and we're going to natural language interacting with agentic models. [03:16] Sounds like bullish on humans. Bullish humans, yeah. No, I agree. Yeah, I think, sure, we're building for like an autonomous agentic future, but also there's going to be a [03:28] a lot of that which is just super charging humans. That's what we see at KOTU is all of our analysts, all the investors that are whispering into their collar are producing better and more work output because they're able to more [03:39] fluidly interact with these models.

[03:42] Like, I think... [03:43] I think the Thinking Machines interaction model demo that came out in May kind of went [03:48] uh, [03:49] under loved. There wasn't enough hype around that. Really? Why? Say more. Yeah. I mean, to me, that Thinking Machines Interaction Model demo was like a bit of a peek into the future. You're going to interact with compute at all times. It's got to have zero low latency back and forth. It has to be able to input images. [04:09] understand what you're saying in real time, respond to you.

[04:12] And so I watched that and said, of course, this is going to sit on top of every... [04:16] intelligence layer, every model, every app, [04:19] that's how we're going to interact with compute. [04:21] I think people were excited about it at the time, but... [04:23] haven't necessarily put the dots together between [04:26] speech to text, interaction models, and thinking about the historical context, which is that every single time we've moved from one [04:34] era of tech to another. [04:35] The device has changed. [04:36] and how we interact with them.

[04:38] that [04:38] device and with compute has changed. So I think it's only logical that that's happening again now. [04:43] Damn. Also sounds like you need Assembly AI's speech models. Definitely. Yeah, I know. There's Assembly's working on it. There's a lot of exciting companies that are working in this space and [04:55] I think we'll see what the device ends up looking like, but I think there will be [04:59] a new format, a new way to interact with AI versus how we used to input [05:04] data into a structured output via [05:07] graphical user interfaces, GUIs, apps, [05:10] keyboards, mice, like to me that we're going to look back in five years and say, of course, we're not using keyboards and mice anymore.

We've come up with a better way to interact with intelligence. [05:20] All right, so we'll upgrade your wires to Neuralink chips. Yeah, maybe that's it. I mean, obviously, wearables is a place people are spending a lot of time. Neuralink is a place people are spending time. I don't know what the answer is, but, like, to me... [05:34] It's a hell of a lot easier to see that the keyboard and mouse are dying a slow death. [05:38] I can't tell you [05:39] It's for smarter people than me, builders, operators, founders, to figure out [05:42] what the new [05:44] format looks like, but...

[05:45] I think that has to be the direction we're going, where... [05:48] You've got OpenAI coming out with [05:50] a super app for the first time that's going to introduce [05:53] almost a billion weekly active users to agents for the first time. [05:58] It's going to change how people [05:59] interact with their phone and change how people want to interact with their phone and we're going to need something new for that. [06:04] And probably some MongoDB databases for all those apps. Yeah, definitely. I mean, you've seen it. You've seen it in some of the public software companies already, and private too, Snowflake, Databricks.

[06:14] Data Dog, all of these businesses that are tied to [06:18] growth in [06:19] AI app production, AI app development, [06:22] are doing great, and Mongo is going to be the exact same type of story. [06:26] Amazing. Well, Max... [06:28] I've got one more. You have one. Just wait. There's more. What is the second hot take? I have one more just because I've been spending a lot of time on it. Yeah. Yeah. [06:37] Um, [06:38] This one I don't have an answer for. [06:40] So I'll just leave you with a question that I don't think people have thought enough about.

- Okay. - There's a lot of discussion. [06:46] open source models versus closed, the frontier, open AI and Anthropic right now. [06:51] Um, [06:53] A lot of the discussion I see is sort of there's either [06:56] One is a winner and one is a loser, or the other is a winner and the other is a loser. [06:59] I think that's, [07:00] Totally the wrong way to think about this. [07:02] I also think there's been some great posts. Jesse Zhang from Decagon posted recently on X about [07:07] How... [07:08] I don't want to get his numbers wrong, but I think 90% of...

[07:10] Decagon's workloads run on [07:12] open source models. [07:14] Damn. [07:15] And that is because... [07:16] they are at [07:17] production scale, they need low latency, they need to be able to fine tune a model, [07:22] and use a small model for that low latency. [07:25] That's not a product than Anthropik or of an AI. [07:28] are offering today. [07:29] you can't do that with [07:31] whatever their frontier motto is. [07:32] and I think that's a good thing. [07:33] So [07:35] They're on 90% open source, yet at the same time, [07:38] enterprise LOM spend.

[07:40] was 19% open source last year, and it's 11% this year. [07:44] So it's dropped. Yet you've got businesses like Decagon telling you that 90% of their workloads are [07:49] are [07:50] done [07:51] via open source. [07:52] Jesse's explanation, I think, is the correct one, which is, [07:55] that [07:56] Open source is good for production grade, like production scale workflows. [08:02] yet we're so early in this AI journey that [08:05] there's not that many production grade workflows, and so [08:07] If you are trying to discover a new use case, you want the smartest model you can possibly get.

You use the Frontier. [08:13] um [08:14] And I think. [08:16] The question that I don't think is being asked in this open source closed frontier debate is, [08:21] clearly people want to [08:23] do model routing. People clearly want to [08:26] fine tune and customize smaller models. [08:29] I don't think anyone's asking the question of what happens when Anthropic and OpenAI decide. [08:33] we can offer that as a product. [08:36] I don't know that they're going to do that, but it seems-- [08:39] plausible that [08:40] they would see this opportunity to [08:42] take their [08:43] older, cheaper, more efficient, smaller models.

[08:46] and sell that as a product. [08:48] I wonder if OpenAI will open source. [08:51] Thank you. [08:52] Yeah, I don't know. I mean, as of now, they won't, but model routing, I think, will be in their best interest, maybe fine-tuning of some kind. [08:59] Uh, [09:00] So that's like the question I don't think is being asked enough. [09:02] when we consider what is the [09:04] end state [09:05] compute ecosystem look like. [09:06] X percent open, X percent closed. [09:09] I don't think people are thinking about the idea that [09:11] like a Microsoft who's saying they can help you build your own model, [09:15] and fine tune that.

[09:17] that Anthropocon Open AI could [09:19] just as feasibly offer that product at some point. [09:21] Damn. Well, we just had Matan from Factory on, and he said... [09:26] Within his enterprise customers, he saw in the beginning of the year it was less than 1%. [09:31] got to 1% in January. Then I think it was this month or something like that. I could be quoting this completely wrong, but it got to 10% open source. Yeah. [09:41] I mean, you're seeing it on X, which is obviously one of the better places to see some of the [09:46] people in the forefront talking about what they're doing.

[09:48] I don't think that's a crazy end state of the world. Let's, [09:51] end state of AI compute, AI inference would be, I don't know, 10% open source? Like, [09:56] The Decagon is [09:57] Jesse's article. [09:59] made a ton of sense. [10:00] Know all of the parameters of the workflow you're trying to [10:04] execute, you've thought about all the edge cases, you've used a frontier model to hone that, [10:10] then it makes sense that [10:11] maybe you're actually seeing better capability with [10:14] an open source model that's fine-tuned or [10:16] the combination of [10:17] a frontier model for your hardest tasks and [10:20] an open source or [10:21] previous version of a model for your kind of [10:24] production grade workflows.

[10:26] like customer service is a great example of that. [10:29] But [10:29] To me... [10:30] That's... [10:31] were really just so early on in terms of what will be the [10:35] kind of adopted... [10:36] Um... [10:37] production grade workflows that are built on top of these AI models that [10:41] the frontier is [10:42] I mean, everything's growing. And so I think that [10:46] this kind of zero sum discussion of open versus sourced is missing the point. [10:50] and [10:52] There are a lot of questions that I don't think people are fully thinking about yet.

[10:55] But it's interesting. Like, it's fun to be sitting in a... [10:58] world that we don't know what the end state looks like, but [11:01] It's our job to [11:01] at least try to predict what it could be. [11:03] amazing two hot takes. And I think where we are, which is not a very large palace in the very random European country is a good kind of, it's a good symbolism of [11:19] how far AI has gone, we can now buy palaces. [11:22] True, true. Yeah, I mean, [11:25] This is just the early days of the AI industry.

This is just the early days. This is literally just the early days. Look where we are. So, okay. A lot of room to go. [11:32] Speaking to that and Rays, what are you most excited about with Rays this year? [11:38] Yeah, I mean, I think the lineup of speakers is amazing. That cuts across like the compute layer, infrastructure layer, app layer. [11:45] There's investors, there's venture investors, [11:48] public equities investors. [11:50] And so to me, to have all these people in one place, to be able to talk about these questions that I'm asking around [11:54] open source.

[11:56] Frontier models. [11:57] Sovereign AI, [11:59] I'm not sure. [11:59] power constraints, grid bottlenecks, like, you... [12:02] That's what I focus on all day, every day, with my colleagues at Cotu. [12:06] It's really, really damn fun to get to talk about it with founders, operators, entrepreneurs, and hear what they're thinking on these same topics. [12:13] Having all these people in the same place is what I'm most excited for. [12:16] Amazing. Great place to end it. Thank you so much, Max. Yeah, thanks, Molly. The sun came out for Henri. Henri, thank you so much for having us here in this very secret location.

And I think we're in Germany. Is that correct? Exactly. We are in South Germany. Thanks for being here, Molly. It's going to be a fun night. Thanks for being here. And welcome. [12:35] So tell us about RAISE this year. It's huge and very fancy. Yeah, it's fun. The idea was to bring the German savoir-faire. So we're not in Germany, we're in France. We can't really disclose where we are. That's part of the... [12:48] of the night and the end of old Dale. But it's kind of bringing like a very different, iconic experience to very tech events.

So we are all in our tech bubble. [12:57] token maxing on a daily basis, but it's good to have a bit of a history on our site, right? [13:02] Thanks for being here, everybody. Of course. I know you're the most popular girl at the party right now, so we're going to keep this really, really short. What is your hottest take right now? [13:10] So my oldest stake is Germany is actually extremely beautiful, and we're extremely fortunate to be here. Perfect. Thank you, Henri. Thank you, Moni. Have a good night. We're here with Pim of General Intuition.

Pim, welcome to Sorcery. Thank you for having me. [13:26] Okay, so for those who don't know what general intuition is, can you give us a little brief overview? [13:31] As humans, we can talk or choose to output texts as one of the actions we can take, whereas [13:39] and LLM is kind of the only thing they can do. So we train models that can predict [13:42] Lots of different actions where text is just a subset, which is a much more general approach in LLMS. [13:47] And what is your hottest take today?

Most of AI is just glorified data science. [13:53] Damn. [13:54] Say more. [13:55] I'm [13:56] Good models mostly are downstream from good theta and simple things. [14:00] So what are you most bullish on in that respect? [14:04] Really good data sets for the most part [14:07] Where do you store that data? [14:09] I cannot disclose that. Well, if you need to store your data, put it in MongoDB databases. No, no, no, no. [14:19] Okay. [14:20] All right. Any other hot takes? Yeah. [14:22] Jan LeCun is underrated. [14:24] You think Jan Lacoon is underrated?

Why? The friends just told me to say that. [14:29] I pulled into a room and they were like, you have to say this. [14:32] No more comments. [14:34] No more comment. He's got a long way now. [14:39] Damn. Okay, see you, Bim. [14:44] Oh, my God. Hi. Okay, we're here with Mark, CTL of AMD. Mark, welcome to Sorcery. Thank you, Molly. Glad to be here. [14:52] Amazing. We're in a very secret location in a European location. I just we can't. [14:57] You don't know where we are, but it's very secret.

Okay, so I have to ask you. [15:01] We're at the Rays Summit. AI is booming. [15:05] What is your hottest take right now? [15:07] Hottest take is people don't even understand how transformational agentic AI is. [15:13] transforming how work gets done. [15:15] and just entire end-to-end workflows are being reimagined. [15:19] And I think it's starting to come out when you hear some of the stories this week at Ray's. [15:23] And what is your biggest lesson that you learned in the last maybe two or three years of this boom at AMD?

Yeah. [15:29] Well, AMD, we got ahead of it. We've worked to getting our CPUs and GPUs [15:34] ready for this. [15:35] And now with these Agenta workflows, you actually need both. In fact, the ratio of CPU to GPU is becoming like one to one. [15:42] But none of us anticipated how fast [15:45] these agentic workflows would just take off. Last six months [15:49] have just been amazing. [15:50] And it's just a harbinger of what's yet to come. We're just at the early stages. [15:54] How many agents are you using today?

Every day I start with agents. They run my day, look at what's coming up. [16:01] But what's really cool is how we do our chip design. [16:05] Thousands of sub-agents just doing incredibly complex chip design, speeding how fast we can get the next generation to market. [16:13] Amazing. Okay, as the drums roll, [16:17] What are you most excited about with RAISE this year? [16:20] Well, it starts with the president of France and ends with Jan LeCun. I think we're going to, we've got it bounded pretty well here. No, it's great.

It's really great speakers. I think we're going to really hear some. [16:30] insights across every facet of the AI ecosystem. [16:34] Amazing. Well, thank you so much, Mark. Really appreciate it. Thank you, Molly. Okay, we're here with Robin of Black Forest Labs, again at a secret location in Europe. Robin, welcome to Sorcery. [16:45] Yes. [16:45] Thank you. Hi. Nice to meet you. [16:48] Nice to see you, I saw you a couple months ago. [16:51] Really? Well. [16:52] In London at the Legends Gala. That's right, yeah, yeah. [16:56] Tough night. Anyways.

Okay. So. [17:00] For people who don't know, what is Black Forest Labs? [17:05] multimodal visual models for [17:08] content... [17:09] creation and for now physical AI. - Amazing. And so we were talking about a couple hot takes earlier, but I'm really curious, what is your take on open [17:20] versus closed, open source versus closed. [17:23] - [17:27] I mean, you can approach this from so many different angles, but I think the most important one is that open innovation is good for the world. [17:34] And I think fear mongering around AI models [17:39] is only going to lead to them being more close, and that's ultimately leading to slow down all AI progress.

And I think ultimately it's super important that [17:49] AI is actually an accessible technology, it's also going to make it more [17:52] safe in you. I think that's something that is [17:57] a pretty fundamental principle. [17:59] if you don't think about like business models that you build around open versus closed doors and all of that, [18:02] Then you can start to be more elaborate. But I think fundamentally open innovation is good for the world. Is it open way? Is it open source? It doesn't really matter in the end, but I think having accessible AI models is key.

[18:16] So you're based in Black Forest, Germany. [18:20] We're here in Europe. [18:22] And there's a nice convergence of European tech and American and Silicon Valley, everything that's converging over here. [18:31] What should people know that aren't from Europe about what's actually going on in European tech right now? [18:38] Thank you. [18:39] Um... [18:41] I think it's getting better. [18:43] enough going on yet I would say? [18:45] even though, I don't know, there's like a fancy event like this. Like it looks really nice, right? But I think like what really matters in the end is that [18:53] Europe is not known for having nice castles and stuff like that, but actually nice AI or nice technology.

[18:59] And I think that's something that, yes, it's changing. And I think events like that where a lot of people come here, I think that's a good start. But ultimately, there needs to be a fundamental mindset to really push AI innovation from here as well. [19:13] I know you're super excited about the RAISE Summit. [19:16] What makes you so excited? [19:19] I'm not gonna answer this question. [19:21] Seeing all your friends maybe? No, it's great. Yeah. I mean, there's like so many good people here. I think it's fantastic. [19:27] Perfect place to end it.

Thank you so much, Robin. Great to see you. Thank you. [19:32] This episode is brought to you by Brex, my favorite. You become what you spend on. And I refuse to spend my time on work that shouldn't exist. Expense reports, receipt chasing, and manual closes. The companies building what's next from Vercel, OpenAI, Anthropic, Granola, and Deepgram, [19:52] All made the same call. They all run on Brex. Brex is the intelligent finance platform that combines cards, expenses, and banking into a single stack with agentic finance built in. AI agents that handle expenses automatically, enforce policy before spend happens, and close your books in minutes.

That's why Sorcery runs on Brex, so I can spend time on building and not busy work. It's time to get Brex AF. [20:22] That's B-R-E-X dot com slash S-O-U-R-C-E-R-Y. [20:30] Bye. [20:30] We have a very special guest here today. We have Matan of Factory. Matan, how are the droids doing? The droids are doing well. The factory is growing. [20:39] Very excited on things that have been happening. [20:42] Are the droids in Europe with us right now? [20:44] The droids are in Europe. They've recently made the jump across the pond. A lot of the largest banks and enterprises...

[20:52] in Europe are now automating their software engineering with droids and with factory. We're very excited about it. So we have very [21:00] difficult question for you today. [21:03] What is your hottest take? [21:04] My hottest take is that [21:07] In the next 12 months, 90% of tokens will be going to... [21:11] open models. [21:12] Um, [21:13] Now, not all the tokens, and maybe not the most important tokens, but [21:16] I think, you know, [21:19] Big Token has been spewing some propaganda against... [21:22] Chinese models, which is, you know... [21:25] how they're labeling the open models.

[21:27] And I think in fact these open models are incredibly performant, incredibly cheap, incredibly fast. [21:33] and there is going to be a [21:34] an important use for them in kind of the portfolio of models. I think still the frontier clothes models will have a place. [21:39] But I think that place will be shrinking at least in token share. Maybe not in cost, but in token share. [21:44] Are you measuring the switches on your side? Yes, and at the beginning of the year we saw enterprises were using less than 1% [21:52] open models, like less than 1% of their tokens were going to open models.

[21:56] Then in around March, it kind of crossed that [21:58] 1% threshold. [22:00] And then in May, it crossed the 10% threshold. [22:02] It's growing pretty quickly because of the... [22:04] kind of cost optimization and also like [22:07] You know, not every task you need Opus 4.8 or like GPT 5.6 ultra high. [22:12] and many tasks day to day you do [22:15] are very simple and you know, at Factory we will actually dynamically route you [22:19] to whichever kind of open or cheaper model is better for that task. [22:23] What's your prediction for percentage of open usage by the end of the year?

[22:29] By the end of the year, I think we'll probably get to, at least in the enterprise, [22:32] we'll probably cross the 50% threshold. [22:34] by the end of this calendar year. [22:36] Oh my God. Breaking news. This is crazy. Okay. As we finish up, [22:41] Anything else you want to share? [22:43] Um... [22:44] What are you most excited about for Rays? [22:47] I'm most excited, honestly, about... [22:49] AI making its way into Europe. They've been [22:53] They haven't been as urgent about it, but I think this is kind of a sign that AI is making its way across the pond [22:59] people really care about it.

There are a lot of businesses [23:02] that can benefit. [23:03] dramatically from some of the kind of [23:04] aggressive, [23:05] sometimes too aggressive moves that people have been making in the United States. [23:09] Excited to see. [23:10] more of that urgency in Europe. [23:13] There are a lot of great companies that it's fun to work with here. [23:15] Amazing. Thank you so much, Maton. Thanks for having me. [23:17] We have Philip Johnson of StarCloud, one of the hottest... [23:22] companies in space right now. [23:24] Philip, how's it going? [23:25] It's going great.

Thanks very much for having me on the podcast. How does it feel to be in SpaceX's S1? [23:32] Amazing. I mean, to be honest, I was not expecting that. And the article they quoted me in, to be honest, [23:38] I'm very surprised they picked that one to quote me. What did they quote? [23:42] So I had to write as part of doing the World Economic Forum like Davos. [23:47] thing you have to like write an article and I basically [23:50] very rapidly threw together an article on data centers in space.

I guess because it was WF they picked that one, but yeah. [23:57] I'm surprised they picked that one. [23:58] It must have been really good. [24:01] Okay, so we have to ask you. We're here on Earth. [24:04] But what is your hottest take right now? [24:06] My hottest take right now is that the SpaceX IPO is the [24:10] going to be viewed historically as the most undervalued IPO of all time. [24:14] I think they will tear through Tentralin [24:17] within, you know, certainly within the next couple of years. [24:20] and then [24:21] have almost unlimited time for what they're building.

[24:23] So in historical terms, it will be viewed incredibly undervalued. [24:28] This is not investment advice, but why are you so bullish? [24:32] I'm not sure. [24:33] I'm not like regulated by anybody, so I can tell people just buy this SpaceX. I think it's a great stock to buy. [24:40] No, but it's mainly because... [24:42] space, not just because of Orbital data centers, which we're doing, which lots of other people [24:46] They own... [24:48] what will be by far the most cost-effective launch vehicle. [24:51] And then that opens up every industry [24:53] in space that will be possible beyond that.

So that's all of [24:56] asteroid mining, lunar resource mining, all of the comms businesses that are going to be built [25:02] Everything else is going to have to go through SpaceX. Yeah, it's like owning the railroads. [25:07] Lots of businesses will be built on top of it, like our business will be built on top of it. But the railroads are a great business to own, for sure. [25:14] So there are a handful of chip companies here this week. [25:18] We're speaking with Cerebris, Sombanova. We just spoke with the CTO of AMD.

Do you have a favorite chip? [25:25] We welcome all chips. We're actually going to be flying. We're speaking with Salmonova. We hope to fly the Salmonova chip. [25:30] with Cerebra 7, an incredible inference chip, we have to fly that chip. [25:34] Within a video, the Grokkar texture actually... [25:37] is -- [25:38] can be potentially well suited for space as well. [25:41] We flew an ARM chip on our first base graph. We flew three ARM GPUs on our first base graph. Nobody knows about that, but... [25:47] we did. [25:48] So yeah, we are.

[25:50] chip agnostic. [25:51] And so how's the factory going? How's, is it? [25:55] Are you guys outgrowing it already? What's going on? Yeah, so we are about to move into a much, much larger facility. We're building a kind of campus to set up a huge manufacturing line. [26:06] for what we're calling StarCloud 3, which is like [26:08] the vehicle which will fit on Starship. So we can potentially fit up to 10 megawatts of compute capacity per Starship launch. [26:14] And so yeah, that's [26:16] In August, we're going to move in, and you'll be the first to do a tour of it.

You heard it here first. We're going to do a tour. Okay, so as we close out, what are you most excited for for this week? [26:28] I mean, they just have some incredible speakers and incredible other attendees, so [26:33] Yeah, I'm excited. I've never met the CEO of Salmonova. The AMD CTO would be great to meet. [26:37] So yeah, excited to meet some people. We just recorded an interview with Rodrigo of Samba Nova this morning. Fantastic. Yeah. [26:45] Really recommend that interview if you haven't listened. Maybe it's out yet. I don't know if it will be.

[26:50] But he is... [26:51] Fantastic. [26:52] fantastic at explaining everything. Yeah. Okay, Philip. [26:56] Thank you so much. Thank you so much. Oh, my God. You guys have to see what's happening at this undisclosed location. All right. We're here with Chris of Good Future Media. Chris, welcome to Sorcery. [27:06] Merci beaucoup, Molly. Very nice to see you here at the show. [27:12] We're going to bleep that one out because we don't know where we are right now and we're in a secret location but... Secret location. [27:19] of a [27:19] Well, this is a nice return because I used to live in France for four years.

So AI, France together, I had to come to this event. [27:26] - Really, what were you doing? - I was an English teacher before, actually. And so for 10 years, I was teaching in the classroom [27:33] And then around the pandemic, I started listening to podcasts. [27:36] the All In Podcast, to be specific. [27:38] I was learning so much and I knew that I had to teach the internet about what I was learning on All In Podcasts. [27:44] and I started clipping up the All In pod [27:46] And that's kind of how I shifted into this new business of [27:49] clipping and so now I have a huge agency [27:52] We get out thousands of clips a month [27:54] And yeah, it's, [27:56] It's amazing what clipping has done to the social media environment, and it really is the game nowadays.

[28:02] Clipping is the new ad. [28:04] clipping is a new ad and it's it's a way to get [28:07] yourself out there, whatever the long form is, whatever message you have, whatever idea or thing you're trying to sell, [28:12] You need to get into short. [28:14] amount of time to be able to grab Sony's attention and get them to maybe check out your longer-form content. [28:20] or what it is you're trying to sell or pitch. [28:22] Okay. [28:23] I have to ask you, for my own benefit, what is the secret to clipping?

[28:27] It's really the first three seconds. [28:29] The three seconds need to be grabbed the viewer in. If the spoken line isn't catching you, you need to have a visual hook on the screen [28:37] that will tell the viewer what is the payoff for them, what's the value that the viewer's going to get [28:42] from watching this clip. [28:43] If you don't have that in the first line of the speaker or a hook on the screen, [28:48] think. [28:49] you know, restart, think of something else because it's not going to go viral, it's not going to do well.

[28:53] What is the most well-performing clip? [28:57] you've ever made. [28:58] We have Timoth [29:01] talking about [29:02] his newest [29:04] startup, [29:05] going, I think we got four million views for him. David Friedberg as well, warning about the famine in Ukraine at the start of the Ukraine war. [29:14] That one went wild on TikTok, 5 million views. [29:18] And, you know, really the all-in besties have been an inspiration for us. [29:22] We're getting lots of good content out there. [29:25] I worked with Harry Stebbings at 20VC for two years.

[29:28] And now we're working with venture capitalists and CEOs and investors across the Silicon Valley world [29:33] landscape. [29:35] What is the price? What is the standard price of clipping today? Yeah, I think I get fleece. [29:43] Quite often. [29:44] It is more expensive than you'd think when you're used to [29:48] AI clips like there's a lot of AI companies out there that will like take your [29:53] Podcast, look at the transcript, like pop out five of the best clips. [29:56] and put captions on there and call it a day.

[29:59] But you really need the human thought, the human psychology going into the clip. [30:03] and adding other visuals to keep people on their toes while watching [30:07] And that takes. [30:08] human time and skilled human time [30:10] So, you know, [30:11] 200 bucks a clip is [30:13] pretty standard in our agency, you know, minimum, and then [30:16] You want to go Cliffs Daily. [30:18] So a clip a day at least. [30:20] If you can do two or three a day, that's even better. [30:22] So that kind of starts to get you [30:24] some of the pricing for the [30:26] a better company like ours.

[30:28] Okay, if we were gonna make this clip go viral, [30:31] What should the hook be? [30:33] "Oh my God, you guys have to see what's happening at this undisclosed location." You won't believe who's here. [30:40] Underscored names. [30:42] And Molly O'Shea Shannon. And I messed up the host name. Oh my God. And I can't wait until you see what's next. [30:52] something like that. All right, we'll lead with that. Chris, thank you so much. Thank you. Thank you. It's very nice to meet you in person. I'm a big fan of your show.

Thank you. I appreciate it. And I'd appreciate any clips that you guys just happened to make. Hey, I think we've made some of your clips, actually. The Techno Optimist Prime is our Instagram account. [31:11] And it's a niche channel where you just clip up our favorite people talking about tech and AI. [31:16] And you're on there too. - Amazing. [31:19] Yeah. [31:19] Awesome. Well, thank you so much, Chris. [31:21] Thank you. [31:22] If you're building what's next in AI, you need to know MongoDB, the database platform developers love and built for the agents you're running.

MongoDB stores searches and reasons over your data in real time with vector search and embeddings from Voyage AI all in the same system. No separate pipelines, no stitching together 10 different tools. It's why 75% of the Fortune 100 and [31:52] billions of vectors. Go to com slash ai to learn more. That's com slash ai to learn more. [32:03] Bye. [32:03] We are here with Ramin of Liquid AI. Ramin, welcome to Sorcery. Thanks for having me. [32:09] So we're in a very secret location. [32:12] that has no AC, but we can't say anything more than that.

What are you so excited about, or why are you so excited about RAISE this year? And tell us about Liquid AI. - Definitely, so I feel like this year, they have exponentially kind of improved the quality of like their guests, you know, that's what I can say. And then the structure is much more mature. [32:33] like [32:34] as I've seen before, there's multiple conferences happening at the same time. There's Makina, and then there's Raze, and everything is related to the stuff that we do. We are bringing foundation models to the physical world.

[32:47] So physical AI is extremely important for us, and that space is [32:52] becoming more and more interesting. We're building foundation models that are so cheap that you can bring them on Raspberry Pis. [33:00] for example, so you can host them on any kind of device that is on the planet. - So you're saying it's really hard to token max with liquid AI. - Correct, yes. I mean, tokens are basically like we're bringing the cost of tokens like to zero, you know, and we are just bringing the value of the AI for enterprises basically.

So like imagine if there's not, [33:19] tokens that are actually important is just the, let's say the outcome that actually matters. And this has been like now, I think the discussions in enterprise AI, which is like one of the odd takes, I guess, you know. [33:30] Karp has something to say about that. I don't know if you saw him. Yeah, I saw him. That was incredible. Any notes? Yeah, I mean, I fully agree with him. You know, like 85% of everything that is happening right now in AI goes into research and development.

[33:44] of code, right? And then we don't know what this code is going to do, you know? And I like that he said, the jig is up, you know? Because enterprises like now, [33:52] In fact, actually in Europe, they are actually waking up right now, and they're getting into the place where they really see, they have to take their time. [34:02] similar to how they have [34:04] been treating like any other technology to really [34:07] understand the value of a tech and then deploy basically those solutions like into their products.

You know, for example, one of the things that we did, we are powering like in-car intelligence for Mercedes-Benz. [34:18] So foundation models getting inside the car, and making your car basically smarter, like this is what we do, [34:25] and then ask, [34:26] This is [34:27] done by a traditional OEM. You can imagine like Mercedes-Benz, like one of the flagship kind of brands in Europe. [34:34] But at the same time, they opened up to a technology that can actually truly unlock value. And I feel like we have to, in enterprise AI, we have to start thinking as foundation model companies [34:44] We have to start thinking about that value chain a lot more, which I think he was on the right path.

So we have to ask you, what is your hottest take right now? Well... [34:57] all foundation model companies from [34:59] 2019 when [35:01] let's say pre-training started like [35:03] panning out, like we saw that we have to kind of scale these AI systems. They started working on one axis. [35:10] The axis was... [35:12] maximizing intelligence at all costs. [35:15] Now we are at a place where we see that [35:18] Look, efficiency is not an afterthought. [35:21] Energy is not abundant. You know, chips are, you know, like there's not that many chips like available in the world right now.

You know, like you have to invest. And also like supply chains are getting disrupted on the memory side of things. Government can have control over your assets, you know, like and whatever you're building. And safe rollout of artificial intelligence at the largest scale, if you have maximized intelligence at all costs, that access alone is not going to get you to the place that you want to go. [35:51] popular, you know, as you're seeing in enterprise AI, especially like as these governance issues are coming in. [35:57] So-- [35:59] I always wanted to propose like this three axis problem, instead of one axis problem.

Imagine you're maximizing intelligence, but at the same time, while you're designing a foundation model, care about its cost. [36:12] So at the same time, you're not just gonna push transformers out. You're going to really work on, let's say, finding out what is the best computational graph of intelligence [36:23] that runs on a certain kind of processor that you want to run it. And then so efficiency and cost of intelligence as a [36:30] like a first class citizen and not afterthought. Because usually now AI companies right now are thinking about doubling down on efficiency of their models, distilling their models down, but this is not something that is like we took it.

So we actually took that axis of efficiency very, very seriously. That's axis number two. The first one was capability as intelligence. [36:49] The third axis, I would say substrate. [36:51] where does this intelligence system go? [36:55] So if you think about... [36:58] AI is majorly like getting hosted in data centers. [37:01] But you could also bring intelligence on phones, on laptops, on airplanes, on cars. You can actually host AI at any different substrate, right? And this is something that I think if you co-evolve it with the capabilities of the models, you can have true enterprise AI capabilities.

[37:17] Damn. [37:18] It's a hot take. [37:19] MS. Well -- and it's also very hot here. [37:21] It is very hot here. [37:24] But Europe or wherever we might be really needs to invest in AC and probably cloud seeding. [37:31] I think Rainmaker might be able to help with these droughts and the heat wave. Okay, any last thoughts? By the way, this segment of Raze is sponsored by MongoDB News. [37:45] BREX and Assembly AI. [37:47] So if you need a database, [37:49] Money. [37:50] Help. [37:51] Or text-to-speech and speech models.

[37:55] That's what you get. Anyways, okay, Ramin... [37:58] To close out, any last thoughts? [38:00] What are you most excited about? I'm actually excited tomorrow I will be on a panel with Ian Stoica, who is like a legend of the field, a co-founder of Databrace and co-founder of like Arena. Like we are going to actually kick off the conference tomorrow at 9 m. and excited to tell the world about open source and, you know, the impact of like, let's say like this three axis that I actually talked about into absolutely realizing kind of the value for enterprises, you know, like not just thinking about like foundation models as like.

[38:28] the token machines that are generating money and revenue for the FAMBISH amount of companies [38:33] that are useless tokens, right? [38:35] and getting them into the place where they can actually unlock [38:37] value for enterprises. [38:39] Amazing. Trees are good. [38:41] Thanks, Rameen. [38:42] Thank you. Rodrigo, welcome to sorcery. [38:45] Thanks for having me. We are in Paris. We have a long interview coming out soon, but before that, you have some really big news today. So what happened? Now, we're super excited. We're announcing the first close, our Series F, is a $1 billion raise at $11 billion valuation.

[39:04] And who led this round? Who are some of the investors in this one? The round was led by General Atlantic. Really excited to have this incredible investor on board. And significant investment from Seligman Ventures, from T. Rowe Price, and Capital Group. [39:17] What are you going to do with this money? We're seeing incredible interest in our product, and so we're using this capital to accelerate our supply chain and making sure we can deliver the racks to all these customers worldwide. So you have a chip with us in the room today.

Can you please show us the chip? This is the exciting new product that we're shipping later this year, SM50. It is a cloud-scale product. [39:44] RDU that allows us to actually deliver premium inference [39:48] at a significantly higher performance than you can get with GPUs at a fraction of the cost, a fraction of the power. For people who don't know what premium inference is, can you explain that a bit? Yeah, so the world of inference is running into an economics problem, and so what we're able to do with premium inference is drive the performance of the largest models up so you can actually deliver high-quality models

[40:11] these trillion parameter models at their full precision at [40:15] incredible speeds. And when you're running faster with a high accuracy model, you can charge more at a lower cost, allowing your service providers to generate more margins for their services. [40:25] All right. So last question before we go. And then you have to check out the full interview afterwards. But what is your hottest take? Well, you know, the world's moving into inference. And so we're excited to be here. We're excited to actually have a new chip to share with the world.

And I think you're going to see Samana would be part of a number of these services before the end of the year. [40:47] Amazing. Bullish Ensemble Nova. By the way, how did you guys come up with that name? [40:52] Well, you know, I grew up in Brazil. Okay. Yeah, I grew up in Brazil. And some of it was a new dance. It's a new dance. It's a new trip. And so check it out.