Raise. Raise. Paris. Paris is beautiful, but it's hot as hell. In the middle of the summer, middle of the heat. People are ignoring it. They're showing up.
Okay, we have a very special guest. I know I've said that before, but we really do this time. Coming off the heels of the Jamon interview at CO2, we have Max Cook of COTU. 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 you've got an old cement wall, which you have many of in Europe, so it might 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, we have to get into this. I think you have a couple of big insights you want to share with us today.
I hope they're insights, but yeah, I mean, I have a couple things I've been thinking about.
Like, what's your hottest take?
Hottest take might be that the keyboard and mouse are slowly dying. I think code too. You've heard Jamon speak about it. You've heard Philippe and Thomas speak about it. We talk about big tech waves. You went from mainframe to PCs to networking the PCs to desktop internet, which is where you get a mouse and keyboard, to mobile internet, which is where the iPhone supremacy began. And with each of these big waves, we end up interacting with compute, interacting with computers in a different way. And it seems pretty clear at this point. I think there's a good stat that like the average person has 60 to 80 apps on their phone.
Wow. They use 11 apps daily on average and 30 apps in any given month. Like it seems quite clear that we're going from a world in which you have 60 to 80 apps to an agentic AI world where you've got a lot fewer than that. Two, three, four apps where everything else plugs in as an API or an MCP. Um, and in that world, it would only make sense that we reinvent how we interact with compute. It's not the keyboard and the mouse. It's probably natural language. Like if we're talking, we speak in natural language. If you're working with a coworker, you ask them what to do, you figure it out together, and you go do it. So, I think there's been a number of good discussions about this. Uh, Mikuel from Replet was talking about how uh 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 tweet about we've always conformed ourselves to computers like with D by inputting specific data with a keyboard and a mouse and now we're finally entering the era where computers have to conform to us. And so this microphone is reminding me it's we're already seeing it at CO2 like in the famous CO2 bullpen that you got to see. Uh we've got a number of of investors that are whispering into their collar half half the day because that's how to give models better context, better prompting and ultimately like produces much better work output. So that's my that's my take. The keyboard and mouse you should throw them away and we're going to natural language interacting with agentic models.
Sounds like bullish on humans.
It's bullish humans. Yeah. No, I I agree. Yeah. I think sure we're building for like an autonomous agentic future, but also there's going to be a lot of that which is just supercharging humans. Like that's what we see at K2 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 fluidly interact with these models. Like I think I think the Thinking Machines interaction model demo that came out in May kind of went uh underloved. Uh wasn't what there wasn't enough hype around that like 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 it's you're going to interact with compute at all times it's got to be have zero low latency back and forth it can has to be able to input images understand what you're saying in real time respond to you and so I watched that and said of course this is like going to sit on top of every intelligence layer every model every app that's how we're going to interact with compute and I think people were excited about it at the time but haven't necessarily put the dots together between speech to text interaction models and thinking about the historical context which is that every single time we've moved from one era of tech to another the device has changed and how we interact with that device and with compute has changed so I think it's only logical that that's happening again now.
Damn. Also sounds like you need Assembly AI speech models.
Definitely, yeah. I know Assembly is working on it. There's a lot of lot of exciting companies that are working in this space and I think we'll see what the device ends up looking like. But I think there will be a new format, a new way to interact with AI versus how we used to input data into a structured output via graphical user interfaces, GUIs, apps, 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. Like we we've come up with a better way to interact with intelligence.
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, it's a hell of a lot easier to see that like the keyboard and mouse are dying a slow death. I can't tell you. It's for smarter people than me, builders, operators, founders to figure out what the new format looks like. But I I think that has to be the direction we're going where you've got OpenAI coming out with a super app for the first time that's going to introduce their almost a billion weekly active users to agents for the first time. It's going to change how people interact with their phone and change how people want to interact with their phone. And we're going to need something new for that.
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 in private too. Snowflake, Databricks, uh Datadog, all of these businesses that are tied to growth in AI app production, AI app development are doing great and is going to be the exact same type of story.
You have one. Just wait. There's more. What What is the second hot take?
One more just because I've been spending a lot of time on it. Um, this one I don't have an answer for. So I'll just leave you with a with a question that I don't think people have thought enough about. There's a lot of discussion of open source models versus closed the frontier OpenAI and Anthropic right now. Um a lot of the discussion I see is sort of there's either one is a winner and one is a loser or the other is a winner and the other is a loser. I think that's totally the wrong way to think about this. I also think like there's been some great posts Jesse Zhang from Decagon posted recently on X about how I don't want to get his numbers wrong but I think 90% of Decagon's workloads run on open source models. And that is because they are at kind of production scale. They need low latency. They need to be able to fine-tune a model and use a small model for that low latency. That's not a product that Anthropic or OpenAI are offering today. You you can't do that with whatever their frontier model is. Um, so they're on 90% open source, yet at the same time enterprise LLM spend was 19% open source last year and it's 11% this year.
Yet you've got businesses like Decagon telling you that 90% of their workloads are done via open source. Jesse's explanation I think is the correct one which is that open source is good for production grade like production scale workflows yet we're so early in this AI journey that there's not that many production grade workflows and so if you are trying to discover a new use case you want the smartest model you can possibly get you use the frontier um and I think the question that I don't think is being asked in this open source closed frontier debate is Clearly, people want to do model routing. People clearly want to fine-tune and and customize smaller models. I don't think anyone's asking the question of what happens when Anthropic and OpenAI decide we can offer that as a product. I don't know that they're going to do that, but it seems plausible that they would see this opportunity to take their older, cheaper, more efficient, smaller models and sell that as a product. I wonder if OpenAI will open source.
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. Uh, so that's like the question I don't think is being asked enough when we consider what is the endstate compute ecosystem look like, X% open, X% closed. I don't think people are thinking about the idea that like a Microsoft who's saying you can they can help you build your own model and fine-tune that that Anthropic and OpenAI could just as feasibly offer that product at some point.
Damn. Well, we just had Matan from Factory on and he said within his enterprise customers he saw in the beginning of the year it was less than 1% 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.
Open source. Yeah. I mean, you're seeing it on X, which is obviously one of the better places to see some of the the people on the forefront talking about what they're doing. I don't think that's a crazy end state of the world. End state of AI compute, AI inference would be, I don't know, 10% open source. Like the Decagon Jesse's article made a ton of sense. If you know all of the parameters of the work workflow you're trying to um execute, you've thought about all the edge cases, you've used a frontier model to to hone that, then it makes sense that maybe you're actually seeing better capability with an open source model that's fine-tuned or the combination of a frontier model for your hardest tasks and an open source or or previous version of a model for your kind of production grade workflows. Like customer service is a great example of that, but to me that's we're really just so early on in terms of what will be the kind of adopted um production grade workflows that are built on top of these AI models that the frontier is I mean every this is everything's growing and so I think that this kind of zero sum discussion of open versus sourced is missing the point and there are a lot of questions that I don't think people are fully thinking about yet but it's It's It's interesting. Like, it's fun to be sitting in a world that we don't know what the end state looks like, but it's our job to at least try to predict what it could be.
Amazing. Two hot takes. And I think where we are, which is not a very large palace in a very random European country is a good kind of it's a good symbolism of how far AI has gone. We can now buy palaces.
True. True. Yeah. I mean this is just the early days of the AI.
This is just the early days. This is literally just the early days where we are. So room to go. Speaking to that and Rays, what are you most excited about with Rays this year?
Yeah, I mean I think the lineup of speakers is amazing. That cuts across like the compute layer, infrastructure layer, app layer, there's investors, there's venture investors, uh public equities investors. And so to me to be have all these people in one place to be able to talk about these questions that I'm asking around open source frontier models uh sovereign AI um the power constraints grid bottlenecks like you I that's what I focus on all day every day with my colleagues at CO2 but it's really really damn fun to get to talk about it with with founders operators entrepreneurs and and hear what they're thinking on these same topics. So having all these people in the same place is what I'm most excited for.
Amazing. Great place to end it. Thank you so much, Max.
The sun came out for Henry. Henry, thank you so much for having us here in this very secret location and I think it's we're in Germany. Is that correct?
Exactly. We are in south of Germany. Thanks for being here, Molly. It's going to be a fun night. Thanks for being here. And uh welcome.
So, tell us about Ray this year. It's huge and very fancy.
Yeah, it's it's fun. Like the the idea was to bring the German suburer. So, we're not in Germany. We're in in France. We can't really disclose where we are. That's part of the of the night and the whole deal. But uh no, it's kind of bringing like a very different iconic experience to a very tech event. So we are all in our tech bubble uh token maxing on a daily basis, but it's good to have a bit of a history in our north our side, right? So thanks for being here.
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?
So my hot take is Germany is actually extremely beautiful and we're super extremely fortunate to be here.
Perfect. Thank you, Henri.
Thank you, Molly. Have a good night.
You too. We're here with Pim of General Intuition. Pim, welcome to Sorcery.
Okay, so for those who don't know what General Intuition is, can you give us a little brief overview?
As humans uh we can uh talk uh or choose to sort of output text as one of the actions we can take whereas an LLM is kind of the only thing they can do. So we train models that can predict lots of different actions where text is just a subset um which is a much more general approach in LLMs.
And what is your hottest take today?
Uh most of AI is just glorified data science.
Uh good models mostly are downstream from good data and simple things.
Uh so what are you most bullish on in that respect?
Uh really good data sets for the most parts.
Where do you store that data?
Uh I cannot disclose that.
Well, if you need to store your data, put it in MongoDB databases.
Okay. All right. Any other hot takes?
Yeah. Uh uh Yan Lun is underrated.
You think Yan Lun is underrated? Why?
Uh the French just told me to say that. It's like I pulled into a room and they were like, "You have to say this. I'm saying this."
No more comments. He's got to walk away now. Out of here. I may not. Damn. Okay. She have him. Oh my god.
Hi. Okay. We're here with Mark, CTO of AMD. Mark, welcome to Sorcery.
Thank you, Molly. Glad to be here.
Amazing. We're in a very secret location in a European loca. I just we can't you don't know where we are but it's very secret. Okay. So I have to ask you we're at the raise summit. AI is booming. What is your hottest take right now?
Hottest take is people don't even understand how transformational Agentic AI is. It's just transforming how work gets done. You know just entire end to end workflows are being reimagined and I think it's starting to come out when you hear some of the stories this week at Ray.
And what is your biggest lesson that you learned in the last maybe two or three years of this boom at AMD?
Well, AMD, we've we got ahead of it. We've worked getting our CPUs and GPUs ready for this. And now with these agent workflows, you actually need both. In fact, the ratio of CPU to GPU is becoming like one to one. Um, but none of us anticipated how fast these agentic workflows would just take off. Last six months have just been amazing and it's just a harbinger of what's yet to come. We're just at the early stages.
How many agents are you using today?
I every day I start with agents. They they run my day, look at what's coming up. But what's really cool is how we do our chip our chip design. We have thousands of sub agents just doing incredibly complex chip design, speeding how fast we can get the next generation to market.
Amazing. Okay, as the drums roll, what are you most excited about with Ray this year? raise.
Uh, well, it starts with the president of France and it ends with Yan Lun. I think we're going to we got it bounded pretty well here. No, it's great. It's really great speakers. I think we're going to really hear some um insights across every facet of the AI ecosystem.
Amazing. Well, thank you so much, Mark. Really appreciate it.
Okay, we're here with Robin of Black Forest Labs again at a secret location in Europe. Robin, welcome to Sorcery.
Thank you. Hi. Nice to meet you.
Nice to see you. I I saw you a couple months ago.
In the London at the Legends Gala.
That's right. Yeah. Yeah.
Tough night. Anyways, okay. So, for people who don't know, what is Black Forest Labs?
Uh, we are building multimodal visual models uh for content creation and for um 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 versus closed, open source versus closed?
I mean, you can approach this from like so many different angles, but I think like the most important one is um that open innovation is good for the world. And I think fear-mongering around AI models is only going to lead to them being more close and that's ultimately going to slow leading to slow down all AI progress and I think ultimately it's super important that uh AI is actually an accessible technology. It's also going to make it more um safe in the end. I think that's like something that is a pretty fundamental principle. If you then think about like business models that you build around open versus closed and all of that then you can like start to be like more like elaborate but I think fundamentally open innovation is good for the world. Is it open weights? Is it open source? It doesn't really matter in the end but I think having accessible AI models is is key.
So you're based in Black Forest Germany. We're here in Europe and there's a nice convergence of European tech and American and Silicon Valley, everything that's converging over here. What should people know that aren't from Europe about what's actually going on in European tech right now?
Um, I think it's getting better. It's there's not enough going on yet, I would say. Um, even though I don't know, there's like a fancy event like this. Uh like it looks really nice, right? But I think like what really matters in the end is that Europe is not known for having like I don't know nice castles or stuff like that, but actually nice AI or nice technology. And I think that's something that yes, it's changing and I think like events like that where a lot of people come here. I think that's a good start, but ultimately there needs to be like a fundamental mindset to really push AI innovation from here as well.
I know you're super excited about the Ray Summit. So, what makes you so excited?
I'm not going to answer this question.
Seeing all your friends, maybe.
No, it's great. Yeah, I mean, there's like so many good people here. I think it's uh it's fantastic.
Perfect place to end it. Thank you so much, Robin. Great to see you.
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 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. Learn more at brex.com/sorcery. That's brex.com/sorcery. Bye.
We have a very special guest here today. We have Matan of Factory. Matan, how are the droids doing?
