# Agent Inference at the "Speed of Light" — How NVIDIA moves like a $4.3 Trillion Startup Latent Space · 2026-03-08 NVIDIA's Nader Khalil and Kyle Kranen join Swyx and Vibhu to explain how the company moves like a $4.3 trillion startup through speed-of-light (SOL) first-principles thinking, agent security boundaries, and the Dynamo inference engine. They argue agents should only do two of three things (files, internet, code) to prevent vulnerabilities, and detail Brev's acquisition to improve developer UX with one-click GPU access and DGX Spark integration. Kyle describes Dynamo as a data center scale inference engine that optimizes serving by scaling out, using prefill/decode disaggregation, Kubernetes-based scheduling, and model-hardware co-design to improve cost, latency, and quality. The episode covers SOL's role in creating urgency, long-context limits and potential 'unhobblers' like multi-head latent attention, and the shift toward CLI-first agent workflows for enterprise tools. ## Questions this episode answers ### What is NVIDIA Dynamo and why was it built for large-scale inference? Kyle Kranen explains Dynamo is a data center scale inference engine that sits on top of frameworks like vLLM and SGLang. It was created because serving at scale cannot just scale up single replicas without performance problems; instead, Dynamo enables scaling out across many nodes. It uses techniques like prefill/decode disaggregation to separate compute-bound and memory-bound phases, and Kubernetes-based orchestration to dynamically adjust resource ratios, improving cost, latency, and quality tradeoffs. [27:22](https://addtry.com/d7e50155-1fb0-453a-9823-f29a73675287?t=1642000) ### What is the 'SOL' principle used at NVIDIA? Nader Khalil describes SOL as 'speed of light'—a first-principles urgency concept where you first determine the physics-based theoretical limit of how fast something can be done, then layer real-world constraints back in. Jensen Huang uses it to push teams to question assumed timelines and break through organizational noise, originating from hardware design as the maximum accelerator speed. It helps create compelling events and reveals true bottlenecks. [14:12](https://addtry.com/d7e50155-1fb0-453a-9823-f29a73675287?t=852000) ### What security limit should AI agents have according to NVIDIA? Nader Khalil says agents can access files, the internet, or write and execute custom code, but you should only allow them to do two of those three things at once. For example, if an agent has file access and code execution, internet access should be blocked to prevent malware injection. This limits the attack surface while still enabling powerful agentic workflows, reflecting NVIDIA's approach to safely adopting agent technology. [0:00](https://addtry.com/d7e50155-1fb0-453a-9823-f29a73675287?t=0) ## Key moments - **[0:00] Agent Security** - [0:00] Agents can access your files, the internet, and write and execute custom code. You should really only let an agent do two of those three things. - **[1:26] Surfboard Booth** - [1:26] Brev's surfboard booth with palm trees at GTC caught NVIDIA's attention, leading to acquisition, says Nader Khalil. - [3:00] Nader Khalil's wife applied Cricut stickers at 1 AM before GTC and said, "If you pull this off, you son of a bitch." - [9:05] NVIDIA's security team recommends running OpenClaw on Brev VMs to isolate agent risks off the corporate network. - **[14:12] SOL Principle** - [14:12] Jensen's "SOL" principle: Start from physics to find theoretical limits before layering in practicality, says Nader Khalil. - [15:26] NVIDIA's "SOL" culture is used by frontline engineers to create urgency and break through noise, says Kyle Kranen. - **[19:12] Kyle's Journey** - [21:42] "The mission is the boss" — NVIDIA allows employees to choose their own projects, says Nader Khalil. - **[26:55] Dynamo Basics** - [26:55] Amazon Ads uses Dynamo for generative recommendation, replacing earlier recommendation models, says Kyle Kranen. - [28:04] Dynamo is a data-center-scale inference engine that optimizes serving by scaling out across frameworks, says Kyle Kranen. - [29:44] Q: Why can't you just scale up inference? A: Hardware bounds like NVLink domain limit scaling up to 8 GPUs, says Kyle Kranen. - [30:14] NVLink provides ~500 GB/s while InfiniBand gives ~50 GB/s, an order-of-magnitude difference in GPU communication. - [34:14] Inference decisions balance three axes: quality (accuracy), cost (budget), and latency (SLA), says Kyle Kranen. - [36:00] Re-running a failed prompt with previous failure context doubles GPU performance, per a recent Google paper, says Vibhu. - **[39:25] Disaggregation** - [40:00] Disaggregated inference separates prefill (compute-bound) from decode (memory-bound) for efficiency gains. - [41:14] Decode is memory-bound because it does a linear pass over all weights per token, while prefill is compute-bound with quadratic attention. - [42:15] NVIDIA's upcoming Reuben CPX accelerator is purpose-built for prefill, says Kyle Kranen. - **[44:36] Context Limits** - [45:25] Kimi K2 reduces attention heads while increasing total experts to manage long-context attention scaling, says Kyle Kranen. - [50:25] "Unhobblers" like multi-head latent attention could enable 10-100 million token contexts, predicts Kyle Kranen. - [51:33] DeepSeek's KV cache for 128k context fits in just 8 GB, vs 40-80 GB for similar-sized models, says Kyle Kranen. - **[54:33] GTC & Agents** - [57:10] Jensen said "Let's make this work across the company" as Codex rolled out to tens of thousands at NVIDIA, says Nader Khalil. - [58:07] Nader Khalil's Outlook CLI agent summarized his inbox, highlighted escalations, and archived 500+ emails automatically. - **[1:03:43] Hackathons** - [1:06:11] NVIDIA's "zero-minute hackathon" at GTC lets agents compete autonomously on undisclosed challenges. - [1:08:04] Coding agents outperform general agents because terminal access enables compile-and-test loop, says Nader Khalil. - [1:09:39] Nader Khalil: LLMs prefer command-line interfaces over APIs because they are deterministic and ubiquitous. - [1:10:22] Pre-training data has far more command-line examples than API docs, making CLIs more effective for agents, says Kyle Kranen. - [1:15:26] Nader Khalil predicts 2025 as the year of the sub-agent, where main agents delegate to specialist sub-agents. - **[1:15:32] Closing** - [1:16:50] Kyle Kranen predicts 2025 as "system as model" where multi-agent systems emulate a single black-box model. - [1:18:48] Kyle Kranen predicts an agent running autonomously for 24+ hours with self-consistency by end of 2025. - [1:20:14] Meter data shows human 5-hour tasks done by agents in 30 minutes; Anthropic Cloud Code averages 20-45 minutes autonomy. ## Speakers - **Swyx** (host) - **Vibhu** (host) - **Kyle Kranen** (guest) - **Nader Khalil** (guest) ## Topics Inference, Agent Infrastructure, Compute ## Mentioned NVIDIA (company), Brev (product), CUDA (product), Claude Code (product), Cloud Code (product), Codex (product), DGX Spark (product), DeepSeek (product), Dynamo (product), Grace Blackwell (product), Kimi (product), Kubernetes (product), Llama (product), NIM (product), Nemotron (product), OpenClaw (product), Reuben CPX (product), SGLang (product), TensorRT-LLM (product), vLLM (product) ## Transcript ### Agent Security **Kyle Kranen** [0:00] Agents can do three things. They can access your files, they can access the internet, and then now they can write custom code and execute it. You should really only let an agent do two of those three things. If you can access your files and you can write custom code, you don't want internet access because that's one, a safety vulnerability, right? If you have access to internet and your file system, you should know the full scope of what that agent's capable of doing, otherwise malware can get injected or s- something bad can happen. And so that's a lot of what we've been thinking about is like, you know, how do we both enable this because it's clearly the future, but then also, you know, what, what are these enforcement points that we can start to, like, protect? **Swyx** [0:39] All right. Welcome to the Latency podcast in the Chroma Studio. Welcome, um, to all the guests here. Uh, we're back with our guest host, Vibhu. Welcome. Good to have you back. Uh, and our friends, uh, Nader f- and, uh, Kyle from NVIDIA. Welcome. **Nader Khalil** [0:51] Yeah, thanks for having us. **Swyx** [0:52] Yeah, thank you. Actually, I don't even know your titles. Uh- I know you're, like, architect something of Dynamo. **Kyle Kranen** [0:59] Uh, yeah. I, I'm one of the engineering leaders f- and ar- uh, architects of Dynamo. **Swyx** [1:03] And you're director of something developers- **Nader Khalil** [1:05] Of developer tech. Yeah **Swyx** [1:06] ... You're the developers' developers' developers guy- ... at NVIDIA. **Nader Khalil** [1:09] Yeah. Open source, agent marketing, Brev- **Swyx** [1:11] Yeah, yeah **Nader Khalil** [1:11] ... and, like, dev tools and stuff- **Swyx** [1:12] Yeah **Nader Khalil** [1:13] ... kind of in the focus. **Swyx** [1:13] And we're, we're kind of recording this ahead of NVIDIA GTC, which is coming to town, uh, again, uh- **Nader Khalil** [1:18] Yeah **Swyx** [1:18] ... or, uh, taking over town uh, which, uh, which we'll all be at. Um, and we'll talk a little bit about your sessions and stuff. **Nader Khalil** [1:25] Yeah. **Kyle Kranen** [1:25] Yeah. **Swyx** [1:25] Yeah. **Kyle Kranen** [1:25] We're super excited for it. ### Surfboard Booth **Swyx** [1:26] One of my favorite memories, uh, for Nader, like, you always do, like, marketing stunts. And, like, while you were at Brev, you, like, had this surfboard that you, like, went down to GTC with. **Kyle Kranen** [1:38] Yeah. **Swyx** [1:38] And, like, Na- uh, NVIDIA apparently liked it so much that they bought you. Like, what, what, what was that like? What was that- **Kyle Kranen** [1:43] Yeah. Um, yeah, we, we, um... Our logo was a shaka. We, we, uh, we were always just kind of, like, trying to keep true to who we were. I think, you know, so many sort of startups, you're, like, trying to pretend that you're a bigger, more mature company than you are, and it was actually Evan Conrad, uh, from SF Compute, who was just like, "You guys are-" **Swyx** [1:59] Yeah, previous guest. Yeah. **Kyle Kranen** [1:59] Yeah. **Swyx** [1:59] Amazing. **Kyle Kranen** [1:59] Oh, really? **Swyx** [2:00] Yeah, yeah. **Kyle Kranen** [2:00] Amazing. Yeah. Um, he was just like, "Guys, you're two dudes in a room. Why are you pretending that you're not?" Uh, and so then we were like, "Okay. Let's make the logo a shaka." We brought surfboards to our booth to GTC, and the energy was great. **Swyx** [2:10] Yeah. **Kyle Kranen** [2:11] Um, some palm trees, too. **Nader Khalil** [2:13] They, they actually poked out over, like, the, the walls, so you could, you could see the Brev booth- **Kyle Kranen** [2:18] Oh, that's so funny **Nader Khalil** [2:18] ... and no one else- **Kyle Kranen** [2:19] Okay **Nader Khalil** [2:19] ... just from very far away. **Kyle Kranen** [2:20] That was- **Swyx** [2:20] Oh, so you remember it back then. **Nader Khalil** [2:21] Yeah, I, I remember it. Pre-acquisition, I was like, "Oh, those guys are cool." **Kyle Kranen** [2:25] Dude, that makes sense 'cause, uh, we-- So we signed up really last-minute, and so we had the last booth. It was all the way in the corner. **Nader Khalil** [2:31] Yeah. **Kyle Kranen** [2:31] And so, like, I was, I was worried that no one was gonna come, so that's why we had, like, the palm trees. We really came in with the surfboards. We even had one of our investors, uh, bring her dog- **Nader Khalil** [2:39] Oh **Kyle Kranen** [2:39] ... and then she was just, like, walking the dog around to try to, like, bring energy towards our booth. **Swyx** [2:44] Yeah, Steph. **Kyle Kranen** [2:45] But yeah, yeah. She's the best. **Swyx** [2:46] You know, as a conference organizer, I love that, right? Like, it's like e- everyone who sponsors a conference comes, does their booth. They're like, "We are changing the future of AI," or something, some generic bullshit. And, like, no. Like, actually try to stand out, make it fun, right? They-- And people still remember it after three years. **Kyle Kranen** [3:00] Yeah. **Nader Khalil** [3:00] Yeah. **Kyle Kranen** [3:00] Like you know what's so funny? I'll, I'll send- I'll give you this clip if you wanna, if you wanna add it in, but, uh, my wife was, uh, at the time fiancée. She was in medical school, and she came to help us 'cause it was, like, a big moment for us, and so we, we bought this Cricut. It's like a vi- like a vinyl, uh, printer 'cause, like, how else are we gonna label the surfboard? So, um, we got a surfboard. Luckily was able to purchase that on the company card. We got a Cricut, and it was just, like, fine-tuning for enterprises or something like that that we put on the, on the surfboard. And it's 1:00 AM, the day before we go to GTC. She's helping me put these, like, vinyl stickers on, and she goes, "You son of a..." She's like, "If you pull this off, you son of a bitch." And so, uh, right-- pretty much after the acquisition, I stitched that with the news of the acquisition, and I sent it to our family group chat. **Nader Khalil** [3:41] Aw. **Swyx** [3:42] Aw. Yeah, yeah, yeah. No. Well, she, she made a good choice there. Uh- ... it-- Was that, like, basically the origin story for Launchables? Is that- **Kyle Kranen** [3:49] We-- It was- **Swyx** [3:49] And maybe we should explain what Brev is and- **Kyle Kranen** [3:51] Yeah. Uh, I mean, Brev is just-- It's a developer tool that makes it really easy to get a GPU. So we connect a bunch of different GPU sources. So the basics of it is, like, how quickly can we SSH you into a G- into a GPU? And, um, whenever we would talk to users, they wanted a GPU. They wanted an A100. And if you go to, like, any cloud provisioning page, usually it's like three pages of forms, or in the forms somewhere there's a dropdown, and in the dropdown there's some weird code that you know to translate to an A100. And I remember just thinking, like, every time someone says they want an A100, like, the piece of text that they're telling me that they want is, like, stuffed away in the corner. **Nader Khalil** [4:23] Yeah. **Kyle Kranen** [4:23] And so we were like, what if the biggest piece of text was what the user's asking for? And so when you go to Brev, it's just big GPU chips with the type that you want. **Swyx** [4:29] With beautiful animations that you worked on. Pre-- Like, pre you can do-- Like, now you can just prompt it. **Kyle Kranen** [4:35] Yeah. **Swyx** [4:35] But back in the day- **Kyle Kranen** [4:36] Yeah. Yeah. Those were handcrafted **Swyx** [4:37] ... handcrafted, artisanal code. **Kyle Kranen** [4:38] Yeah. Yeah. I was actually really proud of that because, uh, it was in-- I, I made it in Figma. **Swyx** [4:42] Yeah. **Kyle Kranen** [4:42] And then I fou- I was, like, really struggling to figure out how to turn it from, like, Figma to React. So what it actually is is just an SVG, and I, and I, I have all the styles. And so when you change the chip, whether it's, like, active or not, um, it changes the SVG code, and that somehow, like, renders-- like, looks like it's animating, but it, it-- We just had the transitions slow. But it's just, like, the, a JavaScript function to change the, like, underlying SVG. **Swyx** [5:04] Yeah. **Kyle Kranen** [5:04] And that was how I ended up, like, figuring out how to move it from, uh, from Figma. But yeah, that's art-artisan. **Nader Khalil** [5:10] S-speaking of marketing stunts, though, he actually used those SVGs, or kind of used those SVGs to make these cards. **Kyle Kranen** [5:18] Oh, yeah. **Nader Khalil** [5:19] Like, a GPU gift card- **Kyle Kranen** [5:20] Yes **Nader Khalil** [5:21] ... that he handed out everywhere. That was actually my first impression of Nader. **Kyle Kranen** [5:24] Yeah, yeah, yeah. **Swyx** [5:24] Yeah, I think I still have one of them. **Kyle Kranen** [5:25] Yeah. **Nader Khalil** [5:26] Yeah. They look great. **Kyle Kranen** [5:27] Yeah. I have a ton of them still actually in our garage, but just they don't have labels. We should honestly, like, bring, bring them back. But, um, I found this old printing press here actually just around the corner on Van Ness, and, um, it's a third-generation San Francisco shop. And so I come in, an excited startup founder trying to, like... And they just have this crazy old machinery, and I'm in awe 'cause the, like, the whole building is so physical. Like, you're seeing these machines. **Nader Khalil** [5:48] Yeah. **Kyle Kranen** [5:48] They have, like, pedals to, like, move these saws and whatever. I don't know what this machinery is. But I saw all three generations. Like, there's, like, the grandpa, the father, and the son. The son was, like, around my age. **Swyx** [5:58] Oh, it's like a holy, holy trinity or something like that? **Kyle Kranen** [6:00] Yeah, yeah. And it's funny 'cause we, we-- So I just took the same SVG, and we just, like, printed it, and it's foil printing, so they make a, a, a mold that's, like, an inverse of, of, like, the A100. And then they put the foil on it, and then they press it into the paper. And I remember once we got them, he was like, "Hey, don't forget about us." Um, you know, I guess, like, early Apple and Cisco's first business cards were all made there. And so he was like, "Yeah, we, we get, like, the startup businesses, uh, but then as they mature, they kind of go somewhere else." And so I actually-- I think we were talking with marketing about, like, using them for some of our cool- **Nader Khalil** [6:28] We should go back and make some cards. **Kyle Kranen** [6:29] Yeah, 100%. **Swyx** [6:29] Yeah, yeah. **Kyle Kranen** [6:30] But- **Swyx** [6:30] Yeah, uh, you know, I, I remember, you know, as, as a very, very small Brev investor, I was like- ... "What-- Why are we spending time, like, you know, doing these, like, stunts for GPUs?" Like, you know, I think, like- ... as a, you know, cl- typical, like, cloud har-hardware person, you go into an AWS, you pick, like, T5 XXL whatever- **Kyle Kranen** [6:46] Yeah **Swyx** [6:46] ... and then you just, like, confirm a list, and you look at the specs. Like, why animate this GP? And, and I, I do think, like, it just shows the level of care that goes throughout Brev and- **Kyle Kranen** [6:55] Yeah **Swyx** [6:55] ... and now-- and also Dynamo. **Kyle Kranen** [6:56] And, and NVIDIA. I think that's what the, the thing that struck me most when we first came in was, like, the amount of passion that everyone has. Like, I think, um, you know, you talk to, you talk to Kyle, you talk to, like... Every, every VP that I've met at NVIDIA goes so close to the metal. Like, I remember it was almost a year ago, and, like, my VP asked me, he's like, "Hey, what's Cursor," and, like, "Are you using it, and if so, why?" And I'm just, like, surprised at this. And he downloaded Cursor, and he was asking me to help him, like, use it. And I thought that was-- Or, or, like, just show him what he-- you know, why we were using it. And so, um, the amount of care that I think everyone has and the, uh- **Nader Khalil** [7:26] Passion **Kyle Kranen** [7:26] ... appreci- passion and appreciation for the moment, right? This is a very unique time. Um, so it's really cool to see everyone, uh, really, like, uh, appreciate that. **Swyx** [7:34] Yeah. One thing I wanted to do before we move over to sort of, like, research topics and, uh, the, the stuff that Kyle's working on is just tell the story of the acquisition, right? Like, not many people have been- ... been through an acquisition with NVIDIA. What's it like? Uh, what-- Yeah, just an-anything you'd like to say. **Kyle Kranen** [7:49] I mean, it's, it's a crazy experience. I think, uh, you know, we were- **Nader Khalil** [7:54] The, the thing that was the most exciting for us was our goal was just to make it easier for developers. We wanted to find access to GPUs, make it easier to do that, and then all... Oh, a- actually, your question about launchables. So launchables was just make one-click expe- like, one-click deploys for any software on top of the GPU. Mm-hmm. And so what we really liked about NVIDIA was that, um, it felt like we just got a lot more resources to do all of that. I think, uh, you know, NVIDIA's goal is to make things as easy for developers as possible, so there was a really nice, like, synergy there. I think that, you know, when it comes to, like, an acquisition, I think the amount that, uh, the SOL of the products align, I think is gonna be- is gonna speak to the success of the acquisition. Yeah. And so it in many ways feels like we're home. This is a really great outcome for us. Like, we, um... You know, I love brev.nvidia.com. Like, you should, you should use it. It's the front page for GPUs. Yeah. If you want GPUs- **Swyx** [8:41] Yeah **Nader Khalil** [8:41] ... you go there, you get 'em. **Swyx** [8:42] And it's like internally it's growing very quickly. I, I- **Nader Khalil** [8:44] Yeah **Swyx** [8:44] ... don't remember, you said some stats there. **Nader Khalil** [8:46] Yeah. Yeah, yeah. It's, uh... I, I wish I had the exact numbers, but, like, internally, externally, it's been growing really quickly. We've been working with a bunch of partners, with a bunch of different customers and ISVs. If you have a solution that you want someone... that runs on a GPU and you want people to use it quickly, we can bundle it up, uh, in a launchable and make it a one-click run. If you're doing things and you want just, like, a sandbox or something to run on, right? Like, OpenClaw, huge moment, super exciting. Our, uh... And we'll, we'll talk into it more, but, you know, internally, people wanna run this, and y- you know, we have to be really careful from the security implications. Do we let this run on the corporate network? Security's guidance was, "Hey, run this on Brev. It's in, you know, it's, it's, it a, it's a VM, it's sitting in the cloud, it's off the corporate network, um, it's isolated." And so that's been our stance internally and externally about how to even run something like OpenClaw while we figure out how to run these things securely. But- Yeah ... um, yeah. **Swyx** [9:33] I think there's also... Like, you almost, like, were the right team at the right time when NVIDIA is starting to invest a lot more in developer experience or what- whatever you call- ... uh, UX or I, I don't know what you call it, like software. **Nader Khalil** [9:47] Yeah. **Swyx** [9:47] Like, obviously NVIDIA's always investing in software, but, like, there's like... this is, like, a different audience. **Nader Khalil** [9:51] Yeah. It's a wider developer base. Yeah. **Swyx** [9:53] Yeah. **Nader Khalil** [9:53] Right? **Swyx** [9:53] Yeah. **Nader Khalil** [9:54] Yeah. And you know, it's funny, it's like it's not, uh- **Swyx** [9:55] Sorry, what, what is it called internally? What, what is this that people should be aware that is going on there? **Nader Khalil** [9:59] Uh, what, like developer experience or- **Swyx** [10:01] Yeah, yeah. Is this, it's called just- **Nader Khalil** [10:02] Yeah **Swyx** [10:02] ... developer experience or is there, like, a broader- **Nader Khalil** [10:03] Yeah, I think- **Swyx** [10:04] ... strategy here? **Nader Khalil** [10:05] In- NVIDIA, um, NVIDIA always wants to make a good developer experience. The thing is, uh, and, you know, a lot of the technology's just really complicated. Like, it's not, it's, uh... You know, I, I think, um, the thing that's been really growing or the, the... AI's growing and it is having a huge moment, not because, like, let's say data scientists in 2018 were quiet then and are much louder now. The pie is com- right? Yeah. There's a whole bunch of new audiences. My mom's wondering what she's doing. My sister's learned, like, taught herself how to code. Like, the, um, uh... You know, I, I actually think just generally AI's a big equalizer and you're seeing a more, like, technologically literate society, I guess. Like, everyone's- ... everyone's learning how to code. Uh, there isn't really an excuse for that. And so building a good UX means that you really understand who your end user is. And when your end user becomes such a wide, uh, variety of people, then you have to almost, like, reinvent the practice, right? Yeah. You have to- And actually build more developer UX, right? Because the, there are tiers of developer base that were added. You know, the, the hackers that are building on top of OpenClaw, right, for example, like, have never used GPU. They don't know what CUDA is. They, they, they just wanna run something. Yeah. Right? You need new UX that is not just, "Hey, you know, how do you program something in CUDA and run it?" And then, and then we built, you know, like, when deep learning was getting big, we built, we built Torch and, and... But so re- recently the amount of, like, layers that are added to that developer stack has just exploded because AI's become ubiquitous. Everyone's using it in different ways. Yeah. It's moving fast in every direction, vertical, horizontally . Yeah. **Swyx** [11:29] Yeah. **Nader Khalil** [11:29] Yeah. **Swyx** [11:29] No, you guys, you even take it down to hardware, like the DGX Spark, you know, it's- **Nader Khalil** [11:33] Yeah **Swyx** [11:33] ... it's basically the same system as just throwing it up on big GPU clusters. **Nader Khalil** [11:36] Yeah, yeah. Yeah. It's a Grace Blackwell. Yeah. Yeah. Uh, we saw the preview at the last year's GTC, and that was one of the better performing, uh, videos of our NVIDIA coverage so far. **Swyx** [11:45] Awesome. **Nader Khalil** [11:45] This'll, this'll beat it. Um That was actually fun. **Swyx** [11:48] We hope, fingers crossed. **Nader Khalil** [11:49] Yeah. Even when Grace Blackwell or when, um, uh, DGX Spark was first coming out, uh, getting to be involved in that from the beginning of the developer experience, and it just comes back to like- **Swyx** [11:57] You were involved? **Nader Khalil** [11:58] Yeah. Yeah, yeah. **Swyx** [11:59] Okay. **Nader Khalil** [11:59] Very directly. **Swyx** [12:00] Oh, yeah. Say more, say more. Yeah. **Nader Khalil** [12:01] Yeah. I mean, from... It was just like I, I got an email. We just got thrown into the loop. Um, and suddenly, uh... Yeah, I, it was actually really funny 'cause I'm still pretty fresh from the acquisition, and I'm, I'm getting an email from a bunch of the engineering VPs about, like, the new hardware GPU chip, like, we're... or not chip, but just GPU system that we're putting out. And I'm like, "Okay, cool. Nader's now involved with this for the UX." I'm like, "What am I gonna do here?" So I remember the first meeting, I was just, like, kinda quiet as I was hearing engineering VPs talk about what this box could be, what it could do, how we should use it. And I remember, uh, one of the first ideas that people were ideating was like, "Oh, the first thing that..." It was like, I think a quote was like, "The first thing someone's gonna wanna do with this is get two of them and run a Kubernetes cluster on top of them." And I was like, "Oh, I think I know why I'm here." I was like, "The first thing we're doing is easy SSH into the machine." And then, and, you know, just kind of like scoping it down of, like, once you can do that, every- you... Like, the person who wants to run a Kubernetes cluster on two Sparks has a higher propensity for pain **Swyx** [12:55] Mm. **Nader Khalil** [12:55] ... than, than, you know, someone who buys it and wants to run OpenClaw right now, right? If you can make sure that that's as effortless as possible, then the rest becomes easy. So there's a tool called NVIDIA Sync that just makes the SSH connection really simple. So, you know, if you think about it, like, if you have a Mac, uh, or a PC or whatever, if you have a laptop and you buy this GPU and you wanna use it, you should be able to use it like it's a, a GPU in the cloud, right? Um, but there's all this friction of, like, how do you actually get into that? That's part of Brev's value proposition is just, you know, there's a CLI that wraps SSH and makes it simple. And so our goal is just get you into that machine really easily. And one thing we just launched at CES, it's in, it's still in, like, early access. We're ironing out some kinks, but it should be ready by GTC. Um- You can register your Spark on Brev. And so now if you- **Swyx** [13:40] Like remote managed, uh- **Nader Khalil** [13:41] Yeah **Swyx** [13:41] ... local hardware. **Nader Khalil** [13:42] Single pane of glass. **Swyx** [13:43] Yeah, yeah. Because Brev can already m- manage other clouds anyway, right? **Nader Khalil** [13:47] Yeah, yeah. And you use the Spark on Brev as well, right? Yeah. Yeah. But yeah, exactly. So, so you, so you, you set it up at home. Um, you can run the command on it, and then it gets, uh... It's essentially, it'll appear in your Brev account. And then you can take your laptop to a Starbucks or to a cafe, and you'll continue to use your s- you can continue to use your Spark just like any other cloud node on Brev. **Swyx** [14:04] Yeah, yeah. **Nader Khalil** [14:04] And it's just like a pre-provisioned GP that doesn't cost you anything. **Swyx** [14:07] So your little data center in your home. **Nader Khalil** [14:07] Yeah, exactly. Yeah. **Swyx** [14:08] Yeah, yeah. **Nader Khalil** [14:09] Tiny little data center. Ti- tiny little data center. It's like the size of your phone. **Swyx** [14:12] Yeah, yeah. One more thing before we move on to Kyle. Just have so many Jensen stories- ... and I just love, love mining Jensen stories. Uh, my favorite so far is SOL. Uh- ### SOL Principle **Nader Khalil** [14:21] Oh, yeah **Swyx** [14:21] ... what is, what is SOL? **Nader Khalil** [14:22] SOL is actually... I, I, I think of all the lessons I've learned, that one's definitely my favorite. Um- It'll always stick with you. Yeah, yeah. I, you know, when you're a startup, everything's existential, right? Like, we've, we've run out of money. We were, like, on the risk of r- of losing payroll. We've had to contract our team because we l- ran out of money. And so, like, um, because of that, you, you're really always forcing yourself to i- to, like, understand the root cause of everything. If you get a date, if you get a timeline, you know exactly why that date or timeline is there. You're, you're pushing every boundary, and, like, you're not just sa- you're not just accepting, like, a, a no just because. And so as you start to introduce more layers, as you start to become a much larger organization, SOL is es- is essentially like, what is the physics, right? The speed of light moves at a certain speed, so if light's moving some- slower, then you know something's in the way. So before trying to, like, layer reality back in of, like, why can't this be delivered at some date, let's just understand the physics. What is the theoretical limit to, like, uh, how fast this can go? And then start to tell me why. 'Cause otherwise people will start telling you why something can't be done. But actually, I think any great leader's goal is just to create urgency. There's an infi- Yeah. Create compelling events- Yeah ... right? Yeah. Uh, uh, SOL is a term NVIDIA's used to sort of, like, instigate a compelling event. You say, "This is done. How do we get there? What is the minimum, as much as necessary, as little as possible thing that it takes for us to get exactly here?" And it helps you just break through a bunch of noise. **Swyx** [15:45] Yeah. **Nader Khalil** [15:45] Instantly. **Swyx** [15:45] One thing I'm unclear about is can only Jensen use the SOL card? Like- **Nader Khalil** [15:49] Oh, no, no, no **Swyx** [15:50] ... you know, like- **Nader Khalil** [15:50] No, everyone **Swyx** [15:50] ... get the bullshit out. **Nader Khalil** [15:51] Yeah. **Swyx** [15:51] Because obviously it's Jensen, but, like, can someone else be like- **Nader Khalil** [15:54] Yeah **Swyx** [15:54] ... "No," like- **Nader Khalil** [15:55] Frontline engineers use it. **Swyx** [15:56] Okay. **Nader Khalil** [15:57] Every- yeah, every... I think it, it's not so much about, like, get the bullshit out. It's like, it's like give me the root understanding, right? Like, if you tell me something takes three weeks, it's like, well, what's- First principles. Yeah, the first principles. It's like, what's, what's the, what, like, why is it three weeks? What is the actual, uh, like, you know, yeah. What's the, what's the actual limit of why this is gonna take three weeks? Um, if you're gonna... If you, if let's say you wanted to buy a new computer and someone told you it's gonna be here in five days, what's the SOL? Well, like, the SOL is, like, I could walk into a Best Buy and pick it up for you, right? So then anything that's, like, be- beyond that is... And is that practical? Is that how we're gonna s- you know, let's say s- give everyone in the company a laptop? Like, obviously not. So then, like, that's the SOL, and then it's like, okay, well, if we have to get more than 10, suddenly there might be some, right? And so now we can kind of piece the reality back. **Swyx** [16:36] So, so this is the Paul Graham, "Do things that don't scale." **Nader Khalil** [16:39] Yeah. **Swyx** [16:40] And this is also the, what people would now call be high agency, right? **Nader Khalil** [16:43] Yeah. Yes. Yeah. **Swyx** [16:44] Just like- **Nader Khalil** [16:44] It's actually really interesting because there's a, there's a second hardware angle to SOL that, like, doesn't come up for, like, all of the org. So SOL is used, like, culturally at NVIDIA for everything. **Swyx** [16:53] Yeah. But I, I'm also mining for that. I think that can be annoying sometimes when, like, someone keeps going, "SOL, SOL, SOL" at you and you're, and you're like, "Guys, like, we have to be stable. We have to- ... we have to fucking plan." **Nader Khalil** [17:03] Yeah. It's an interesting balance. Yeah. I encounter that with, like, actually just with, with Alec, right? 'Cause we, we have a, we have a new conference, uh, uh, so we need to launch... We, we have, we have goals of what we wanna launch by, uh, by the conference. And like, yeah, at the end of the day- **Swyx** [17:14] Wait, is this GTC something? **Nader Khalil** [17:15] Um, well, this is like, so we... I mean, we did it for CES. **Swyx** [17:17] I see, okay. **Nader Khalil** [17:17] We did it for GTC DC before that. We're doing it for GTC San Jose. So I mean, like, every, you know, we have a, we have a new moment, um, and we want to launch something. **Swyx** [17:25] Yeah. **Nader Khalil** [17:25] And we want to do so at SOL, and that does mean that some- there's some level of prioritization that needs to happen. And, uh, so it, it is difficult, right? I think, um, you have to be careful with what you're pushing. Y- you know, stability is important, and that should be factored into SOL. SOL isn't just, like, build everything and let it break, you know. That, that's part of the conversation. So as you're laying, layering in all the details, one of them might be, "Hey, we could build this, but then it's not gonna be stable for X, Y, Z reasons." **Swyx** [17:50] Yeah. **Nader Khalil** [17:50] And so that was, like, one of our conversations for CES, was, you know, "Hey, like, we, we can get this into early access, uh, registering your Spark with Brev, but, um, there are a lot of things that we need to do in order to feel really comfortable from a security perspective," right? There's a lot of networking involved before we deliver that to users. So it's like, okay, let's get this to a point where we can at least let people experiment with it. We had it in a booth. We had it in Jensen's keynote. And then let's go iron out all the networking kinks, and that's not easy. And so, uh, that can come later. And so that was the way that we layered that back in. Yeah. But- It, it's not really about, like, saying, like, you don't have to do the, the maintenance or operational work. It's more about saying, you know... It, it's kinda like highlights how progress is incremental, right? Like, what is the minimum thing that we can get to? And then there's SOL for, like, every component after that, but there's the SOL to get you, get you to the, the starting line, right? And that, that's usually how it's asked. Yeah. Um, on the other side, you know, like, SOL came out of, like, hardware at NVIDIA, right? So SOL is, like, literally if we ran the accelerator or the GPU with, like... at basically full speed with, like, no other constraints, like, how fast we'd be able to make a program go. **Swyx** [19:00] Yeah, yeah. **Nader Khalil** [19:00] Right? Uh, so- **Swyx** [19:02] In, in training that, like, you know, then you work back to, like, some percentage of, like, MFU, for example. **Nader Khalil** [19:07] Yeah. That's, that's a great example. **Swyx** [19:08] Yeah. **Nader Khalil** [19:08] So, like, there's an, there's an SOL MFU, and then there's, like, you know, what's practically achievable. ### Kyle's Journey **Nader Khalil** [19:12] Yeah. **Swyx** [19:12] Cool. Uh, shall we move on to sort of, uh, Kyle's side? Uh, Kyle, you're sort of coming more from the data science world. Um, and- **Nader Khalil** [19:19] Yeah **Swyx** [19:19] ... uh, I, I mean, I always, whenever, whenever I meet someone who's done work in tabular stuff, graph neural networks, time series, these are basically... When I go to NeurIPS, I go to ICML, I walk the back halls, there's always, like, a small group of graph people- **Nader Khalil** [19:33] Yes **Swyx** [19:33] ... small group of tabular people. **Nader Khalil** [19:34] Absolutely. **Swyx** [19:35] And, like, there's no one, no one there. And, like, it's very... Like, you know what I mean. Like- **Nader Khalil** [19:39] Yeah, no **Swyx** [19:39] ... like, it, it's, it's important, interesting work if you care about solving the problems that they solve. **Nader Khalil** [19:46] Yeah. **Swyx** [19:46] But everyone else is just LMS all the time. **Nader Khalil** [19:49] Yeah. I mean, it's like, it's like the black hole, right? Has the event horizon reached this yet in NeurIPS? Um- **Swyx** [19:54] But, like, you know- ... tho- those are, those are transformers too. **Nader Khalil** [19:57] Yeah. **Swyx** [19:57] And, and those are also, like, interesting things. Anyway, uh, I just wanted to spend a little bit of time on, on those, that background before we go into Dynamo, uh, proper. **Nader Khalil** [20:05] Yeah, sure. I took a different path to NVIDIA than Nader. I joined six years ago, seven if you count when I was an intern. So I joined NVIDIA, like, right out of college, and the first thing I jumped into was not what I'd done in during internship, which was like, you know, like some stuff for autonomous vehicles, like heavyweight object detection. I jumped into like, you know, something I'm like, recommenders, this is popular. Um, and- **Swyx** [20:27] Oh, yeah. You did RecSys as well. **Nader Khalil** [20:28] Yeah, RecSys. **Swyx** [20:28] Yeah. **Nader Khalil** [20:28] Yeah. I mean, that, that was the tabular data at the time, right? You have tables of like, you know, audience qualities and item qualities, and you're trying to figure out like which member of the audience matches which item or, or more practically which item matches which member of the audience. And at the time, really, it was like we were trying to enable, uh, recommenders, which had historically been like a little bit of a CPU-based workflow into something that like ran really well on GPUs. And it's since been done, like there are a bunch of libraries for RecSys that run on GPUs. Uh, the common models like deep learning recommendation model, which came out of Meta, and the wide and deep model, which was used, uh, or was re-released by Google, were very accelerated by GPUs using, you know, the fast HBM on the chips, especially to do, you know, vector lookups. But, uh, you know, it was very interesting at the time and super, super relevant because like we were starting to get like this explosion of feeds and things that required rec- recommenders to just actively be on all the time and sort of transition that a little bit towards graphical networks when I discovered them because I was like, "Okay, you can actually use graphical neu- neural networks to represent like relationships between people, items, concepts." And that, that interested me, so I jumped into that at NVIDIA and, and got really involved for like two-ish years. **Swyx** [21:41] Yeah. **Nader Khalil** [21:41] Uh- **Swyx** [21:42] And something I learned from Bryan Catanzaro- **Nader Khalil** [21:44] Yeah **Swyx** [21:44] ... is that you can just kind of choose your own path in NVIDIA. **Nader Khalil** [21:47] Oh my God. Yeah. **Swyx** [21:48] Which is not a normal big corp thing. **Nader Khalil** [21:50] Yeah. **Swyx** [21:50] Like you, you have a lane, you stay in your lane. **Nader Khalil** [21:52] I think probably the reason why I enjoy being in a, a big company- The mission is the boss ... coming from a startup guy, yeah. **Swyx** [21:58] The mission is the boss. **Nader Khalil** [21:59] Yeah. Uh, it feels like a big game of pickup basketball. Like, you know, if you play one, if you want to play basketball, you just go up to the court and you're like, "Hey, look, we're gonna play this game and we need three." Yeah. And you just like p- find your three. Um, that's honestly for every new initiative, uh, that's what it feels like. Yeah. **Swyx** [22:12] Yeah. **Vibhu** [22:12] Yeah. It also like shows, right? Like NVIDIA is just releasing state-of-the-art stuff in every domain. **Nader Khalil** [22:17] Yeah. **Vibhu** [22:17] Like, okay, you expect foundation models with Nemotron. Voice just randomly like- **Nader Khalil** [22:22] Parakeet **Vibhu** [22:22] ... popped here. Parakeet just comes out. **Nader Khalil** [22:24] Oh. **Vibhu** [22:24] Another one, uh, voice- **Nader Khalil** [22:25] The NVIDIA voice team has always been producing incredible stuff **Vibhu** [22:27] Yeah. There's always just every other domain of paper that comes out- ... dataset that comes out and it's like... I mean, it also stems back to what NVIDIA has to do, right? You have to make chips years before they're actually produced, right? So you need to know, you need to really forecast out. **Nader Khalil** [22:39] The design process starts like- **Vibhu** [22:40] Exactly **Nader Khalil** [22:41] ... three to five years before the chip gets to the market. **Vibhu** [22:44] Yeah. I, I'm curious more about what that's like, right? So like you have specialist teams. Is it just like, you know, people find an interest, you go in, you go deep on whatever, and that kind of feeds back into, you know, okay, we, we expect predictions. Like the internals at NVIDIA must be crazy, right? **Nader Khalil** [23:00] Yeah, yeah. **Vibhu** [23:00] You know, you know, you, you must... Not even without selling the people, you have your own predictions of where things are going. **Nader Khalil** [23:06] Yeah. **Vibhu** [23:06] And they're very based, very grounded, right? **Nader Khalil** [23:09] Yeah. It, it's really interesting. Um, so there's like two things I think that NVIDIA does which are quite interesting. Uh, one is like we really index into passion. There is, there is a big sort of organizational top-down push to like ensure that people are working on the things that they're passionate about. So if someone proposes something that's interesting, many times they can just email someone like way up the chain that they would find this relevant and say like, "Hey, can I go work on this?" That's actually like I worked at a, a big company for a couple of years before, uh- Yeah ... starting on my startup journey and like it felt very weird if you were to like email out of chain, if that makes sense. Yeah. The emails at NVIDIA are like mosh pits. **Swyx** [23:44] Shoot. **Nader Khalil** [23:45] It's just like 60 people just whatever and like there, there- **Swyx** [23:48] Doesn't it get messy, like reply all, you know? **Nader Khalil** [23:49] Oh, it gets in... It's insane. It's insane. But- Agents help, you know, manage the context now. But, but, but that's actually like, um, I, I've actually... So this is a weird thing where I used to be like, "Why would we send emails? We have Slack." I am the entire... I'm the exact opposite. I feel so bad for anyone who's like messaging me on Slack because I'm so unresponsive. **Swyx** [24:05] But you're email maxing. **Nader Khalil** [24:06] I'm, I'm email maxing now. Email max. Email is a different purpose. Email is perfect- **Swyx** [24:09] Oh, man **Nader Khalil** [24:09] ... because, because important threads- **Swyx** [24:10] We can't work together on Slack then. **Nader Khalil** [24:12] Email is great because important threads get bumped back up, right? **Swyx** [24:15] Yeah, yeah. **Nader Khalil** [24:15] Um, and so Slack doesn't do that, so I just have like this casino going off on the right or on the left and like I don't know which thread was from where or what, but like the threads get... A- and then also just like the subject, so you can have like working threads. I think what's difficult is like when you're, when you're small, if it's not 40,000 people, um, I think Slack will work fine, but there's... I don't know what the inflection point is. There is gonna be a point where that becomes really messy, and you'll actually prefer having email because you can have working threads, you can CC more than nine people in a thread. You can fork stuff. You can fork stuff, which is super nice, and just like y- yeah. And so, um, but that is part of where you can propose a plan. You can also just like start... Honestly, momentum is the only authority, right? So like if you can just start, start to make a little bit of progress and show someone something, and then they can try it, that's I think what's been, you know, I think the most effective way to push anything for- forward and that's both at NVIDIA and I think just generally. **Swyx** [25:02] Yeah. **Nader Khalil** [25:02] There's, there's the other concept that like is explored a lot at NVIDIA, which is this idea of a zero billion dollar business. Like, like market creation is a big thing at NVIDIA. Like- **Swyx** [25:11] Oh, you want to go and start a zero billion dollar business? **Nader Khalil** [25:14] Jensen, he says, "We're completely happy investing in zero billion dollar markets. We don't care if this creates revenue. It's important for us to know about this market. We think it will be important in the future. It can be zero billion dollars for a while." I'm probably mangling his words here. But like, you know, like I'll, I'll give an example. NVIDIA's been working on autonomous driving for a, a long time. **Swyx** [25:34] Like an NVIDIA car? **Nader Khalil** [25:36] No. Like, like- **Vibhu** [25:36] They've used the Mercedes, right? **Nader Khalil** [25:37] Yeah. **Vibhu** [25:37] They're around the HQ. **Nader Khalil** [25:39] Yeah. **Swyx** [25:39] Yes. **Vibhu** [25:39] And I think it finally just got licensed out. Now they're starting to be used quite a bit. **Nader Khalil** [25:42] Yeah. Yeah. Yeah. **Vibhu** [25:43] But for 10 years you've been seeing- **Nader Khalil** [25:45] What? **Vibhu** [25:45] ... Mercedes with NVIDIA logos driving around. **Nader Khalil** [25:47] Oh, yeah, yeah. If you're, if you're, if you're in like the South Bay near Sanic- San Clara, it's, it's actually pretty common. **Swyx** [25:52] You're from South Bay. This track is right. **Nader Khalil** [25:53] Oh, nice. Yeah. Yeah. So, um- You know, zero billion dollar markets are, are a thing. Like, you know, Jensen- I mean, okay, look, cars are not a zero billion dollar market, but yeah . That's a bad example . I think, I think he's, he's me- messaging, uh- Zero today, but- Or even, like, internally, right? Like, like it's like y- uh, an org doesn't have to ruthlessly find revenue very quickly to justify their existence, right? Like, the, a lot of the important research, a lot of the important technology being developed, um, that- that's kind of where- Research is very ideolo- ideologically free at NVIDIA. Yeah. Like, they can pursue things that they- W- were you research officially? I was never in research officially. Okay. I was always in engineering. Yeah. We're in m- I'm in an org called Deep Learning Algorithms, which is basically just how do we make things that are relevant to deep learning go fast. That sounds freaking cool. **Vibhu** [26:35] A- and I think a lot of that is underappreciated, right? Like time series. This week Google put out TimeFX. **Nader Khalil** [26:40] New paper, yeah. **Vibhu** [26:40] A new time series paper. **Nader Khalil** [26:41] Yeah. **Vibhu** [26:41] RexxAR. Semantic ID started applying transformers, LLMs to RexxAR. **Nader Khalil** [26:46] Yes. **Vibhu** [26:46] And when you think the scale of companies deploying these, right? Amazon recommendations, Google web search. Like, it's, it's huge scale and- **Nader Khalil** [26:53] Yeah **Vibhu** [26:53] ... you want fast. **Nader Khalil** [26:54] Yeah, yeah, yeah. Yeah. Actually it, it... I... There was a fun moment that brought me, like, full circle. Like, uh, uh, Amazon Ads recently gave a talk where they talked about using Dynamo for generative recommendation, which was, like, super, like, weirdly cathartic for me. ### Dynamo Basics **Nader Khalil** [27:09] I'm like- **Vibhu** [27:09] Hmm **Nader Khalil** [27:09] ... "Oh my God, I've, I've supplanted what I was working on." Like, I you're using LLMs now to do what I was doing five years ago. Yeah, yeah, yeah. Um- Amazing. And let's go right into Dynamo. Uh, maybe introduce- Yeah, sure ... it sort of top-down and yeah. I think at this point a lot of people are familiar with the term of inference. Uh, like, funnily enough, like, I, I went from, you know, inference being like a really niche topic to being something that's, like, discussed on, like, normal people's Twitter feeds. It's on billboards here now . Yeah. Very, very strange. Drive- driving, seeing just an inference ad on 101. Inference at scale is becoming a lot more important. Uh, we have these moments like, you know, OpenClaw where you have these agents that take lots and lots of tokens but produce, you know, incredible results. There are many different aspects of test time scaling so that, you know, you can use more inference to generate a better result than if you were to use, like, a short amount of inference. There's reasoning, there's re-querying, there's, you know, adding agency to the model, allowing it to call tools and use skills. Dynamo sort of came about at NVIDIA because myself and a couple others were, were sort of talking about the, these concepts that like, you know, you have inference engines like vLLM, SGLang, TensorRT-LLM, um, and they have, like, one single copy. They, they, they sort of think about, like, things as, like, one single copy, like, one replica, right? Like, one version of the model. But when you're actually serving things at scale, you can't just scale up that replica because you end up with, like, performance problems. Like, there's a scaling limit to scaling up replicas. So you actually have to scale out to use a, you know, maybe some Ku- Kubernetes type terminology. We kind of realized that there was, like, a lot of potential optimization that we could do in scaling out and building systems for data center scale inference. So Dynamo is this data center scale inference engine that sits on top of the frameworks like vLLM, SGLang, and TensorRT-LLM, and just makes things go faster because you can leverage the economy of scale, the fact that you have KV cache, which we can define a little bit later, uh, in all these machines that is, like, unique and you wanna figure out, like, the ways to maximize your cache hits. Or you want to employ new techniques in inference like disaggregation, which Dynamo, um, you know, ha- introduced to the world in, in, in March. Not introduced, it was a academic talk beforehand, but we're, you know, one of the first frameworks to start, you know, supporting it. And we wanna, like, sort of combine all these techniques into sort of a modular framework that allows you to accelerate your inference at scale. By the way, Kyle and I became friends on my first day at NVIDIA, and I always love 'cause, like, he always teaches me new things. Um- Yeah. By the way, this is why I wanted to put two of you together. I was like, "Yeah, this is g- this is gonna be good." It's very d- it's very different, you know. Like, we've, we've, we've, we've talked to each other a bunch. Actually, you know, uh, you asked, like, why, why can't we scale up. Yeah. Um- Model, you said mo- model replicas. Yeah. So you, so scale up means assigning more- Heavier ... yeah, heavier. Like making things heavier. Yeah. Adding more GPUs, adding more CPUs. Uh, scale out is just, like, having a barrier saying, "I'm gonna duplicate my, my representation of the model or a representation of this s- microservice or something, and I'm gonna, like, replicate it many times to handle the load." And the reason that you can't scale, scale up, uh, past some points is, like, you know, there, there, there are sort of hardware bounds and algorithmic bounds on, on that type of scaling. So I'll give you a good example that's, like, very trivial. Let's say you're on an H100. Uh, the maximum NVLink domain, domain for H100 for most DGX H100s is, uh, eight GPUs. Hmm. Right? So if you scaled up past that, you're gonna have to figure out ways to handle the fact that now for the GPUs to communicate, you have to do it over InfiniBand, which is still very fast, but is not as fast as NVLink. Is it, like, one order of magnitude, like hundreds or- It's about an order of magnitude. Yeah. Okay. Um, so- Not terrible. Yeah. I, I need to, I need to remember the, the, the data sheet here. Like, I think it's, like, about 500 gigabytes, uh, a second unidirectional for NVLink and about 50 gigabytes a second unidirectional for InfiniBand. Um, uh, I, it, it depends on the, the generation. I just, I just wanna set this up for people who are not familiar with these kinds of, like, l- layers and- Ah, of course, yeah ... the transfer speeds and all that. **Vibhu** [31:18] Also, maybe even just going, like, a few steps back before that. Like, most people are very familiar with, you see, uh, you know, you can use on your laptop whatever these SGLang, vLLM, uh, you know. You can just run inference. There's OLM and there's all the- **Nader Khalil** [31:31] You can scale, you can run it on a laptop. **Vibhu** [31:32] You can run on laptop. Then you get to, okay, uh, models got pretty big, right? GLM-5, they doubled the size. So, uh, what do you do when you have to go from, okay, I can get 128 gigs of memory. I can run it on a Spark. Then you have to go multi-GPU. **Nader Khalil** [31:46] Yeah. **Vibhu** [31:46] Okay, multi-GPU, there's some support there. Now, if I'm a company and I don't have, like- I'm not hiring the best researchers for this, right? But I need to go multi-node, right? I have a lot of servers. Well, okay, now there's efficiency problems, right? You can have multiple eight H100 nodes, but, you know, is that... Is it... Like, how do you do that efficiently? **Nader Khalil** [32:05] Yeah. How do you, like, represent them? How do you choose how to represent the model? **Vibhu** [32:08] Yeah, exactly. **Nader Khalil** [32:08] Right? That's like... That's like a hard question everyone asks. Like, how do you size? Like, oh, I wanna run GLM-5, which just came out, new model. There have been like four of them in the past week, by the way. Like a bunch of new models. **Swyx** [32:19] You know why, right? DeepSeek. **Nader Khalil** [32:21] No comment. **Vibhu** [32:23] Oh. **Nader Khalil** [32:23] Um, yeah, but G- G- GLM-5, right? We, we have this, you know, new model. It's, it's of, like, a large size, and you have to figure out how to s- both scale up and scale out, right? Because you have to find the right representation that you care about. I mean, everyone does this differently. Let's be very clear. Everyone figures this out in their own path. I feel like a lot of AI or ML even is like, is like this. I think people think, you know... I, I was... You know, there was some tweet a few months ago that was like, "Why hasn't fine-tuning as a service taken off?" Mm. And, you, you know, and like- **Vibhu** [32:48] That might be me. **Nader Khalil** [32:50] It might have been you. Yeah. But people want it to be so- such an easy recipe to follow, but even, like, if you look at an MOE model- It's specific to you. Yeah, yeah And, and the model and the situation. And there's so much, there's so much tinkering. Like, like when you see a model that has however many experts in the MOE model, it's like, why that many experts? I don't know. They, you know, they tried a bunch of things and that one seemed to do better. And I think when it comes to how you're serving inference, you know, you have a bunch of decisions to make and there... You can always argue that you can take something and make it more optimal, but I think it's, it's this internal calibration and appetite for continued calibration. **Vibhu** [33:20] Yeah. And that doesn't mean, like, you know, people aren't taking a shot at this, like Tinker from Thinking Machines, you know? **Nader Khalil** [33:24] Yeah. **Vibhu** [33:24] Uh, RL as a service. **Nader Khalil** [33:25] Yeah. **Swyx** [33:26] Totally. **Vibhu** [33:26] It, it also gets even harder when you try to do big model training, right? We're not the best at training MOEs, uh, when they're pre-trained. Like we saw this with Llama 3, right? They're trained in such a sparse way that Meta knows there's gonna be a bunch of inference done on these, right? They'll open source it, but it's very trained for what Meta infrastructure wants, right? They wanna, they wanna inference it a lot. **Nader Khalil** [33:46] Yeah. **Vibhu** [33:46] Now, the, the question to basically think about is, okay, say you wanna serve a chat application, a coding copilot, right? You're doing a layer of RL. You're serving a model for X amount of people. It's a chat model, a coding model. So Dynamo, you know, back to that, it's like- **Nader Khalil** [34:00] Yeah, sorry. So y- we, we, we sort of like jumped off of, you know- Thanks. ... jumped off. Uh, uh, on that topic, everyone has, like, their own journey, and I, I like to think of it as defined by, like, what is the model you need? What is the accuracy you need? Actually, I, I talked to Nader about this earlier. There's, there's three axes you care about. What is the quality that you're able to produce? So, like, are you accurate enough or can you complete the task with enough, you know- **Swyx** [34:23] Performance **Nader Khalil** [34:23] ... high enough performance. **Vibhu** [34:24] Yeah. **Nader Khalil** [34:25] Yeah. Uh, there's cost. Can you serve the model or serve your workflow because it's not just the model anymore, it's the workflow, it's the multi-turn with an agent cheaply enough, and then can you serve it fast enough? And we're seeing all three of these, like, play out. Like we saw, we saw new models from OpenAI that, uh, you know, are, are faster. You have like these new fast versions of models. You can change the amount of thinking to change the amount of quality, right? Produce more tokens, but at a higher cost and a, and a higher latency. And really, like, when you start this journey of, like, trying to figure out how you wanna host a model, you, you, you think about three things. What is the model I need to serve? You know, how many times do I need to call it? What is the input sequence length? What is the, what does the workflow look like on top of it? What is the SLA? What is the latency SLA that I need to achieve? Because there's usually some... This is usually like a constant. You, you know the SLA that you need to hit, and then, like, you try and find the lowest cost version that hits all of these constraints. Usually, you know, you, you start with those things and you say you, you kinda do like a bit of experimentation across some common configurations. You change the tensor parallel size, which is a form of parallelism. **Vibhu** [35:30] I'd say it goes even deeper. First you gotta think what model. **Nader Khalil** [35:33] Yes. **Vibhu** [35:33] You know, there's- **Nader Khalil** [35:34] It's like a, it's like a multi-step design process because as you said, you can, you can choose a smaller model and then do more test time scaling, and it'll eq- equate the, the quality, uh, of a larger model because you're doing the test time scaling or you're adding a harness or something. So yes, it, it goes way deeper than that. But from the performance perspective, like once you get to the model you need, you need to host, you look at that and you say, "Hey, I have this model. I need to serve it at this speed. What is the right configuration for that?" Do you guys see the recent, uh... There was a paper I just saw like a few days ago that, uh, if you run the same prompt twice- **Vibhu** [36:05] Yes **Nader Khalil** [36:06] ... you're getting like double GPU perform- **Vibhu** [36:06] Just try it again. **Nader Khalil** [36:07] Yeah, exactly. **Vibhu** [36:07] And you get a lot- **Nader Khalil** [36:08] Yeah. **Vibhu** [36:08] But the, the key thing there is you give the context of the failed try, right? **Nader Khalil** [36:11] Yeah. **Vibhu** [36:11] So it takes a shot- **Nader Khalil** [36:13] Mm-hmm **Vibhu** [36:13] ... and this has been like, you know, basic guidance for quite a while. Just try again, 'cause you know, it tried. Just try again. Did you try again? **Nader Khalil** [36:20] All advice in life, just try again. **Vibhu** [36:21] And I think it's a, it's a paper from Google- **Nader Khalil** [36:23] Try again. Failures are- **Vibhu** [36:23] ... if I'm not mistaken, right? **Swyx** [36:25] Is it? Yeah, yeah. **Vibhu** [36:25] I think it's, it's like a seven page, little short paper. **Nader Khalil** [36:27] Yeah, yeah. **Vibhu** [36:27] The title's very cute and it's just like, "Yeah, just try again. Give it- **Nader Khalil** [36:29] Just give it- **Vibhu** [36:29] ... it has context. Let it go again." **Nader Khalil** [36:31] Multi-shot. You just like say like, "Hey, like, you know, like take, take a little bit more, take a little bit more information." Try and fail, fail- **Vibhu** [36:37] And that basic concept has gone pretty deep. There's like, um, self distillation RL where you, you do self distillation, you do RL, and you have past failure and, you know, that gives some signal. So people take try it again not strong enough. **Swyx** [36:51] Uh, for, for listeners, uh, who listen to here, uh, Vibhu actually and, and I, and we run a second YouTube channel for our paper club where- **Nader Khalil** [37:00] Oh, that's awesome **Swyx** [37:00] ... we, we would just cover this. **Nader Khalil** [37:01] Yeah. Awesome. **Swyx** [37:02] Self distillation and all that. **Nader Khalil** [37:03] That's awesome. **Swyx** [37:03] That's, that's why he's like so up to speed on it. **Nader Khalil** [37:05] I'll have to, yeah, I'll have to check it out. **Swyx** [37:05] Yeah. Yeah. It, it's just a good practice. Like everyone needs like a paper club where like you just read papers together and the social pressure just kind of forces you to read these papers. **Nader Khalil** [37:12] We, we have... There's like a big inference reading group at NVIDIA. I feel so bad every time I, I... He put it on, like, on our... He shared it the other day. **Swyx** [37:19] Yeah. One, one of your guys- **Nader Khalil** [37:20] Yeah **Swyx** [37:20] ... uh, is, is big in that. I forget. **Nader Khalil** [37:21] Yeah, Ishan. **Swyx** [37:21] Ishan. **Nader Khalil** [37:22] Ishan. **Swyx** [37:22] Yeah, yeah, yeah, Ishan. **Nader Khalil** [37:23] Ishan's on my team. **Swyx** [37:24] Ishan, yeah. **Nader Khalil** [37:24] Actually funny, there was a, there's a, there's a employee transfer between us. Ishan worked for Nader- Yeah, yeah ... at Brev and now he, he's on my team. He was, he was our head of AI and then yeah, once we got in, he and Kyle both- **Swyx** [37:32] So, 'cause I, I'm always looking for like, okay, can, can I start a- another podcast that only does that thing? **Nader Khalil** [37:37] Yeah. **Swyx** [37:37] And, uh, Ishan was... I was trying to like nudge Ishan into like, is there something here? I mean, I don't think there's, there's new inference techniques every day. So it's like, it's like- **Nader Khalil** [37:45] You would, you would actually be surprised. Um, the amount of blog posts you see, and, and if you- **Swyx** [37:50] There, there was a period where it was like Medusa, Hydra, what Eagle, like, you know, like there was- **Nader Khalil** [37:54] Now we have new forms of decode spec... Or we have new forms of speculative decoding or new- **Swyx** [37:58] What, what are you excited about then? **Vibhu** [37:58] And it, it's exciting when you guys put out something like Nemotron 'cause I remember the paper on this, Nemotron 3, uh, the amount of like post-training, the amount of tokens that the GPU rich can just train on- ... and it, it was a hybrid state space model, right? **Nader Khalil** [38:12] Yeah. It's co-designed for the hardware. **Vibhu** [38:13] Yeah, go design for hardware. And one of the things was always, you know, the state space models don't scale as well when you do a conversion or whatever, the performan- And you guys are like, "No, just keep training," and Nemotron shows a lot of that. **Swyx** [38:25] Yeah. **Nader Khalil** [38:25] Also something cool about Nemotron, uh, it was released in, in layers, if you will. Very similar to Dynamo. It's, it's ess- it's essentially, uh, it was released as aggregate. You can-- The pre-training, post-training data sets are released. **Swyx** [38:36] Yeah. **Nader Khalil** [38:36] The recipes on how to do it are released. The model itself is released- **Swyx** [38:39] It's a fully open model **Nader Khalil** [38:39] ... so you can just benefit from us turning on the GPUs. But there are companies like, uh, ServiceNow took the data set and they trained their own model, and we were super excited- **Vibhu** [38:47] Yeah **Nader Khalil** [38:48] ... and cele- like, you know, celebrated that work. **Swyx** [38:49] And Zoom, Zoom- **Nader Khalil** [38:49] Yeah **Swyx** [38:49] ... the frontier model. **Vibhu** [38:50] Zoom is, Zoom is- **Nader Khalil** [38:51] Zoom is AGI. **Vibhu** [38:52] I think, uh, you know, also just to add, like, a lot of models don't put out base models, and if there's that, why is fine-tuning not taking off? You know, you can do your own post-training, but- **Nader Khalil** [39:02] That's an interesting question. Yeah, that's true **Vibhu** [39:03] ... uh, you guys put out base model. I think you put out everything. **Swyx** [39:06] I believe so. **Nader Khalil** [39:06] I don't know about base. **Vibhu** [39:08] Put out base. **Nader Khalil** [39:09] Base can be, base can be cancelable. **Vibhu** [39:10] Base can be cancelable? **Nader Khalil** [39:11] Yeah. **Vibhu** [39:12] Safety training. **Nader Khalil** [39:14] Did we get a full picture of Dynamo? I, I don't know if we, we, we- **Swyx** [39:16] I mean, what, what I'd love is you, you mentioned the three axes. Um, like, break it down of like, you know, what's prefill, decode, and like what are the- **Nader Khalil** [39:23] Yeah **Swyx** [39:23] ... optimizations that we can get with Dynamo. **Nader Khalil** [39:24] Yeah. That, that, that's, that's, that's a great point. So to summarize on that three-axis problem, right, there are three things that determine whether or not something can be done with inference: cost, quality, latency. Right? Dynamo is supposed to be there to provide you, like, the runtime that allows you to pull levers to, you know, mix it up and move ar- around the Pareto frontier or the Pareto surface that determines is this actually possible with inference in AI today. ### Disaggregation **Swyx** [39:46] Gives you the knobs. **Nader Khalil** [39:47] Yeah, exactly. Gives you the knobs. Uh, and one thing that, like, we, we use a lot in contemporary inference and is, you know, starting to, like, pick up from, you know, in, in general knowledge is this con- concept of disaggregation. So historically, models would be hosted with a single inference engine, and that inference engine would sort of ping pong between two phases. There's prefill, where you're reading the sequence, generating KV cache, which is basically just a set of vectors that represent the sequence, and then using that KV cache to generate new tokens, which is called decode. And some brilliant researchers, uh, across multiple different papers essentially made the realization that if you separate these two phases, you actually gain some benefits. Those benefits are basically, A, uh, you don't have to worry about step synchronous scheduling. So the way that an inference engine works is you do one step, and then you finish it, and then you schedule, you start scheduling the next step. The, it's not, like, fully asynchronous. And the problem with that is you would have, uh, essentially prefill and decode are, are actually very different in terms of both their resource requirements and their sometimes their runtime. So you would have, like, prefill that would, like, block decode steps because you, you'd still be prefilling and you couldn't schedule because, you know, the step has to end. Um, so you remove that scheduling issue, and then you also allow you or y- yourself to, like, split the work into two different ki- types of pools. So prefill typically, and, and this changes as, as model architecture changes, prefill is right now compute bound most of the time. If the sequence is sufficiently long, it's compute bound. On the decode side, because you're doing a full pass over all the weights and the entire sequence every time you do a decode step and you're, you don't have the quadratic computation of KV, KV cache, it's usually memory bound because you're retrieving a linear amount of memory and you're doing a linear amount of compute as opposed to prefill, where you retrieve a linear amount of memory and then use a quadratic amount. **Vibhu** [41:45] You know what's funny? Someone, uh, Exo Labs did a really cool demo- **Nader Khalil** [41:49] Mm-hmm **Vibhu** [41:49] ... where for the DGX Spark, which has a lot more compute, you can do the pre- the compute hungry prefill on a DGX Spark and then do the, uh, decode on a, on a Mac. **Nader Khalil** [41:59] Yeah. **Vibhu** [41:59] And, and so- **Nader Khalil** [41:59] That's faster, yeah. **Vibhu** [42:01] Yeah. **Nader Khalil** [42:01] So you can, you can do like- **Vibhu** [42:02] And I- **Nader Khalil** [42:02] You can do machine stra- stratification. **Vibhu** [42:04] Yeah. **Nader Khalil** [42:04] And, like, with our future generations, generations of hardware, we actually announced, like, with Reuben, this new accelerator that is prefill specific. It's called Reuben CPX. So- **Vibhu** [42:16] I have a question. When you do the scale out, is scaling out easier with Dynamo? Because when you need a new node, you can dedicate it to either the prefill or, uh, decode. **Nader Khalil** [42:25] Yeah. So Dynamo actually has like a, a Kubernetes, uh, component in it called Grove that allows you do, do this, like, crazy scaling specialization. It has, like, this hot... It's a representation that-- I don't wanna go too deep into Kubernetes here, but th- there was a previous way that you would, like, launch multi-node work. Uh, it's called leader worker set. It's in the Kubernetes standard, and leader worker set is great. It served a lot of people super well for a long period of time. But one of the things that it struggles with is representing, uh, a set of cases where you have a multi-node replica that has a pair, right, you know, prefill and decode, or, uh, it's, it's not paired but has like a second stage, that has a ratio that changes over time. **Vibhu** [43:06] Hmm. **Nader Khalil** [43:06] Um, and prefill and decode are, like, two different things. As your workload changes, right, the amount of prefill you'll need to do may change. The amount of decode that you, you'll need to do might change, right? Like, let's say you start getting, like, insanely long, uh, queries, right? That probably means that your prefill scales, like, harder because you're hitting these, this quadratic scaling growth. **Vibhu** [43:23] Yeah. **Nader Khalil** [43:24] Um- **Swyx** [43:24] And then for listeners, like, prefill will be long input, uh, decode will be long output, for example, right? **Nader Khalil** [43:29] Yeah. So, like, decode, decode scale... I mean, decode is funny because the amount of tokens that you produce scales with the output length, but the amount of work that you do per step scales with the amount of tokens in the context. **Swyx** [43:40] Yes. **Nader Khalil** [43:41] So both scales with the input and the output. **Swyx** [43:43] That's true. **Nader Khalil** [43:44] But on the, on the prefill decode side, like, if, if suddenly, like, the amount of work you're doing on the decode side stays about the same or if, like, scales a little bit and then the prefill side, like, jumps up a lot, you actually don't want that ratio to be the same. You want it to change over time. So Dynamo has a set of components that, A, tell you how to scale. It tells you how many prefill workers and decode workers you th- it thinks you should have, and also provides a scheduling API for Kubernetes that allows you to actually represent and affect this scheduling on, on, on your actual hardware, on your compute infrastructure. **Vibhu** [44:16] Not gonna lie, I feel a little embarrassed for being proud of my SVG function earlier. **Swyx** [44:21] No, it was, it was really cute. I, I liked it. **Vibhu** [44:23] It's all, it's all engineering. It's all engineering. **Nader Khalil** [44:25] Sort of technical. **Swyx** [44:26] One thing I'm, I'm kinda just curious about with all, with, uh... Y- you see at a systems level everything going on here. **Vibhu** [44:32] Mm-hmm. **Swyx** [44:32] And, and, uh, we're, you know, we're scaling it up in, in multi-in distributed systems. Um, I think one thing that's like kind of the moment right now is people are asking is there any SOL sort of upper bounds in terms of like, let's call it, just call it context length for want of a better word, but you can break it down however you like. ### Context Limits **Nader Khalil** [44:49] Yeah. **Swyx** [44:49] Uh, I just think like, well, yeah, I mean, like clearly you can engage in hybrid architectures and throw in some space-based models in there all, all you want, but it looks, still looks very attention heavy. **Nader Khalil** [45:01] Yes. Uh, yeah, long context is attention heavy. I mean, we have these, these hybrid models, um, to take- **Swyx** [45:06] And, and most, most models like cap out at a million contexts and that's it. **Nader Khalil** [45:09] Yeah. **Swyx** [45:09] Like for the last two years has been it. **Nader Khalil** [45:11] Yeah. The model hardware context co-design thing that we're seeing these days is actually super interesting. It's like my, my passion, like my secret side passion. We see models like Kimi or GPGOSs. I'm gonna use these because I, I know specific things about these models. So Kimi 2 comes out, right? And it's an interesting model. It's like, like a DeepSeek style architecture. It is MLA, it's basically DeepSeek scaled like a little bit differently, um, and obviously trained differently as well. But they, they talked about, you know, why they made the design choices. For context, Kimi has more experts but fewer attention heads and I believe a slightly smaller attention, uh, like dimension, but I need to remember... I need to check that. Uh, it doesn't matter. But they, they discussed this actually at length in a blog post on Jihoo, which is like... Or Jeepoo, which is like Reddit. **Swyx** [46:03] Jeepoo. Yeah. **Nader Khalil** [46:04] Um, in, in China. **Swyx** [46:04] Chinese Reddit. Yeah, it is. **Nader Khalil** [46:05] Yeah. So it's, it's, it's actually an incredible blog post. Uh, like all the MLS people in, in, in, that I've seen that on Jeepoo are like very brilliant. But they, they, they talk about like the creators of Kimi K2 actually like talked about it on, on, on there in a blog post. And they say, "We a- we actually did an experiment, right?" Attention scales with the number of heads. Obviously like if you have 64 heads versus 32 heads, you do half the work of, of attention, is you still scale quadratically, but you do half the work. And they made a, a very specific like sort of barter in their system, in their architecture. They basically said, "Hey, what if we gave it more experts? So we're gonna use more memory capacity, but we keep the amount of activated experts the same. We increase the experts sparsity so we have fewer experts ac- the ratio to, of experts activated to number of experts is smaller, and we decrease the number of attention heads." **Vibhu** [46:56] A- and kind of for context, what the, what we had been seeing was you make models sparser instead. So no one was really touching heads, you're just having, uh- **Nader Khalil** [47:04] Well, they, they did, they implicitly made it sparser. **Vibhu** [47:06] Yeah. For, for Kimi they did. **Nader Khalil** [47:07] Yeah. Yes. **Vibhu** [47:07] They also made it sparser, but basically what we were seeing was people were at the level of, okay, there's a sparsity ratio, you want more total parameters, less active- **Nader Khalil** [47:16] Mm-hmm **Vibhu** [47:16] ... and that's sparsity. But what you see from papers like, you know, the labs like Moonshot, DeepSeek, they go to the level of, okay, outside of just number of experts you can also change how many attention heads and less attention layers. Uh, more attention layers. **Nader Khalil** [47:30] More layers. Yeah. **Vibhu** [47:30] Yes, yes, yes. So and that's all basically coming back to just tie it together is like hardware model co-design, which is- **Nader Khalil** [47:36] Harder model co- model context w- co-design. **Vibhu** [47:39] Yeah. **Nader Khalil** [47:39] Right? Like if you were training a, a model that was like really, really short context, uh, or like really li- is good at super short context tasks, you may like design it in a way such that like you don't care about attention scaling because it hasn't hit that like the turning point where like the quadratic curve takes over. **Vibhu** [47:54] How do you consider attention, uh, or context as a separate part of the co-design? Like I would imagine hardware or just how I would've thought of it is like hardware model co-design would be hardware model context co-design. **Swyx** [48:04] Hmm. **Nader Khalil** [48:05] Because the harness and the context that is produced by the harness is a part of the model once it's trained in. **Vibhu** [48:12] Like even though towards the end you'll do long context, you're not changing architecture through training, right? **Nader Khalil** [48:16] I see. Yeah. **Swyx** [48:17] I mean, you can try. You're saying everyone's training the harness into the model? **Nader Khalil** [48:22] I would say to some degree- **Vibhu** [48:23] Or there's co-design for the harness **Swyx** [48:24] I, I know there's a small amount, but I feel like- **Nader Khalil** [48:26] No **Swyx** [48:26] ... not everyone has like gone full send on this thing. **Nader Khalil** [48:28] I think, I think, I think it's important to internalize the harness that you think the model will be running- **Vibhu** [48:33] Will be running **Nader Khalil** [48:34] ... into the model. **Swyx** [48:34] Yeah, yeah. **Vibhu** [48:35] Interesting. Okay. **Swyx** [48:36] And like Bash is like the universal harness. **Vibhu** [48:38] Yeah. **Swyx** [48:38] Right? Like- **Nader Khalil** [48:39] Like I'll, I'll give an example here, right? Um, I mean, or just like a, like a s- easy proof, right? If you can train against a harness and you're using that harness for everything, wouldn't you just train with the harness to ensure that you get the best possible quality out of... **Swyx** [48:55] Uh, well, the, I, I can provide a counterargument- **Nader Khalil** [48:57] Yeah, sure **Swyx** [48:58] ... which is what you wanna provide a generally useful model for other people to plug into their harnesses, right? **Nader Khalil** [49:03] Yeah. **Swyx** [49:03] So if you- **Nader Khalil** [49:03] But har- harnesses can be open, open source, right? **Swyx** [49:05] Yes. **Nader Khalil** [49:06] We obviously see stuff like- **Swyx** [49:06] I mean, that's, that's effectively what's happening with Codex. **Nader Khalil** [49:08] Yeah. **Swyx** [49:08] Um, a- but like you may want like a different search tool and then you may have to name it differently or... **Vibhu** [49:13] I don't know how much people have pushed on this, but can you train a model? Would it be... Have, have people compared training a model for the h- uh, for the harness versus like post-training for, for- **Swyx** [49:24] I think it's the same thing **Vibhu** [49:24] ... the harness loop? It's the same thing. Okay. **Swyx** [49:25] It's just extra post-training. **Vibhu** [49:26] I see. **Swyx** [49:27] And so I mean, Cognition does this, Corso does this, where you, you just have to like, if your tool is slightly different, um, either force your tool to be like the tool that they trained for- **Vibhu** [49:35] Hmm **Swyx** [49:35] ... or like undo their training for their tool and then- **Vibhu** [49:37] Oh, that's funny **Swyx** [49:38] ... re- retrain. Yeah, it's, it's really annoying and like... **Nader Khalil** [49:40] I would hope that eventually we hit like a, like a certain level of generality with respect to harness training- **Vibhu** [49:44] Correct. And it's, this is not- **Nader Khalil** [49:46] ... so you can introduce new tools **Vibhu** [49:46] ... this is not AGI. Like it's just like- **Swyx** [49:48] Yeah. It's a really stupid like learn my tool, bitch. Like I don't know if, I don't know if I can say that, but like, you know. Um, I think what my point kind of is, is that there's... Like I look at slopes of the scaling laws and like this slope is not working, man. Y- we're at a million token context. Okay, maybe next year 2 million. We're not going to 100 trillion. You know, like this, this, this- **Nader Khalil** [50:07] Oh, there are so many interesting ways you can get there **Swyx** [50:08] ... this doesn't work. This doesn't work. **Nader Khalil** [50:10] I, I, what's kind of funny is whenever there... I, I feel like we always want to see a trend that we can predict, but every time something's come, it's been like a leapfrog. So I, I imagine... I, I don't know how we go from one to two, but I imagine what, what's likely to happen is we break through that from some new- Yeah. There's actually, there's an interesting formalization of this. There, there's an essay, it's a, it's pretty interesting essay by Leopold Aschenbrenner called Situational Awareness. **Swyx** [50:33] Okay, yes. **Nader Khalil** [50:34] He introduces a concept in situational- **Swyx** [50:35] We're familiar with that, yes **Nader Khalil** [50:36] ... awareness called an unhobbler, right? So he, he, you know, Leopold in this essay details, hey, I want to get, you know, like, I wanna get to this point in intelligence, and I think that it is four orders of magnitude worth of like compute and data and training away. And, you know, he says, "Oh yeah, I think data centers can scale up by about this much. I think that you can do, you know, scale up the data and some other things by this much." But one of the things that like makes the, the rest of that order of magnitude growth po- possible is these unhobblers. Like these scientific discoveries that are discovered during, you know, model architecture search or training that, um, really, really, really impact how, how you are able to scale. Like a, a good example of this might be that like we see like a mo- a lot of models that are... and this is probably a very tiny unhobbler, but is important for the performance perspective. We see a lot of models that are like trained with multi-token prediction natively in, during pre-training. **Swyx** [51:33] Mm-hmm. **Nader Khalil** [51:34] And, uh, you know, per DeepSeek in their paper they say, "Hey, this, this actually helped us ins- ensure sta- more stable convergence." But there are like unhobblers that are like that, and then there are like rather large unhobblers, right? Like architecturally a lot of our models, like we have different types of attention, right? And one of the problems with attention is like you have a lot of KV, but people have found like different forms of, uh, attention, like grouped query attention and, uh, like MLA in DeepSeek, uh, multi-head latent attention that like decrease the burden that KV has on the model, which allows you to grow like longer in context. **Swyx** [52:05] Yeah, and that, that was very drastic for DeepSeek. **Nader Khalil** [52:07] Yeah, yeah. **Swyx** [52:08] This was like- **Nader Khalil** [52:08] For context, uh, like the, the total, I think the total context length of DeepSeek is 128,000 tokens or it might be 256,000 with rope extension. That entire context, I think it's 128,000, fits into eight gigabytes. And previously context, like I think the, the Llama 4 or 5B context of a similar size was like 40 or 80 gigabytes in the same precision. **Swyx** [52:31] Yeah. **Nader Khalil** [52:32] Um, so like those unhobblers like really decrease the stuff of that size, and I wouldn't be surprised if we do see the ability to like break through to like 10 million, 20 million, 100 million context through the, a- an unhobbler showing up. **Swyx** [52:47] I see. **Nader Khalil** [52:48] And it's just science. **Swyx** [52:48] So, so more deep learning algorithms is what I hear. **Nader Khalil** [52:50] Yeah, more deep learning algorithms. Um, I, I could, I could actually- **Vibhu** [52:54] He's playing pickup and he has room for two. **Nader Khalil** [52:56] I, I could actually give you an e- an example, uh, like of like a, a theory, not a theory theory, but something theoretical- **Vibhu** [53:02] An unhobbler that you're excited about or? **Nader Khalil** [53:03] Well, an, an unhobbler that, I mean, I haven't seen, so it could be a tarpit and could not, just not work . But, uh, I, I would be really excited to see a model that does prefill and decode differently. So a model that does- **Vibhu** [53:15] Hmm **Nader Khalil** [53:15] ... uh, prefill like locally, like document-wise prefill, like it does it in chunks, and then you do decode globally across like the entire sequence. Because it, you know, logically to me it doesn't seem like you would necessarily need to have KV be associative between documents that have like no asso- no mutual association. Um, but that like places a lot of burden on prefill to like, or sorry, on, on decode and pure attention within the decode phase to like make those connections since the KV is like static at that point. And you see other techniques that are interesting like this too. Um, but if, if you're able to do that, like, you know, if, if prefill becomes local and decode is, is still global, you solve that prefill quadratic scaling problem because you have a bunch of like small chunks that you prefill independently. **Swyx** [53:59] Okay. All right. Well, let's, uh, wait and see, but I, I think it'll be pretty exciting. **Vibhu** [54:03] Fingers crossed. **Swyx** [54:03] Yeah, fingers crossed. Yeah, yeah. **Vibhu** [54:05] I'm excited for like prefill decode on separate hardware. So like- **Nader Khalil** [54:09] Yeah **Vibhu** [54:09] ... Groq acquisition, right? Can we, can we decode on the Groq? Can we get super fast? **Nader Khalil** [54:14] I don't think I'm allowed to comment on this. **Swyx** [54:17] Mark is gonna shoot arrows at us. **Vibhu** [54:19] He's got a blow dart. **Nader Khalil** [54:20] Yeah. He's in his other room just like- **Swyx** [54:22] Like **Nader Khalil** [54:22] ... "Go to sleep." **Swyx** [54:22] Yeah, yeah. **Nader Khalil** [54:23] Yeah, but I'm, I'm super excited to see the team come in and like, you know, I've gotten the, the pleasure of working with some of the, the Groq people coming in so, um, you know- **Swyx** [54:30] Yeah **Nader Khalil** [54:30] ... I, I know- **Swyx** [54:31] Sunny, we've had him, uh- **Nader Khalil** [54:32] Yeah **Swyx** [54:32] ... at the same conference- **Nader Khalil** [54:33] Oh, that was awesome **Swyx** [54:33] ... that you were at. ### GTC & Agents **Nader Khalil** [54:33] Yeah. **Swyx** [54:34] Um, and, uh, I, I think you're, you guys are gonna be doing some sessions at GTC. I don't know if you wanna... this is a good place to plug them. **Nader Khalil** [54:39] Yeah, yeah. So, um, I can't speak to any LPU related sessions at GTC. I have no idea about that. **Swyx** [54:45] Oh, no, that was you. **Nader Khalil** [54:47] On the, on the Groq side, yeah, I, I use the associative NVIDIA U. **Swyx** [54:50] Yeah. **Nader Khalil** [54:50] Um, o- on the, on the NVIDIA Dynamo side, we're, we're giving it, you know, there are a, are a large number of sessions. Uh, for those that aren't aware, you can actually search all of these sessions for, uh, GTC online. Just go to the GTC website. I don't know what the URL is, but go there. **Swyx** [55:07] Google it. Yeah. **Nader Khalil** [55:07] Uh, and you can just look up Dynamo and you'll get all the sessions. There are about 20. There are a couple that are hosted by the Dynamo team. There are a couple that are hosted by people that use Dynamo that wanna show off the results they've been able to get. But there are two that I'm really excited about. Uh, one is just the general Dynamo tutorial, and this is the, you know, I'm, I'm going out with Harry, who's our lead product manager for Dynamo, and we're sort of talking about like how to use Dynamo to get better performance and also like where we see Dynamo going in the future. And then there's another session that I'm doing with one of our agents teams at NVIDIA to talk about sort of the future of agents in production inference. **Swyx** [55:42] Damn. **Nader Khalil** [55:43] Um, so we're talking about there's like this new horizon with respect to agents because we have these harnesses that actually impart structure upon, upon calls. Like if you, if you compare like, you know, the past and the, and the present with respect to like how, how LM calls work, like in the early days when they were chatbots, like every call was like very different. There was basically no structure. You could assume that like people you, i- you know, if it was conversational, there might be like some implicit structure because you have, you know, a multi-turn conversation. But agents you have this, this harness that like abides by rules, right? So it imparts direct structure onto the context. And you see this, there was an interesting Twitter post about how Claude Code, uh, like structures its context so that you get as many cache hits as possible. **Swyx** [56:27] Hmm. **Nader Khalil** [56:27] And I think it was by one of the, the PMs for Claude Code, and he, he wrote about it and That type of structure that the harness can impart actually, like, goes hand-in-hand with the inference co-design. So I'm doing a talk, I, I don't know the session name or the session number But I'm, I'm doing a talk, uh, you can look at me up by name on, on the, on the GTC website, on how we accelerate agents and where we see specific optimizations for agents going in Dynamo and in inference in general. **Vibhu** [56:56] Yeah, I think there's only one PM for CloudCode and it's Kyle Wu. There is, like, there's, there's, there's DevRel, there's Boris- **Kyle Kranen** [57:01] Maybe it was, maybe it was DevRel. Yeah, exactly, right **Vibhu** [57:03] ... yeah, yeah, exactly. I mean, let's go into agents. I think this is, like, the last part of the, the, the s- this discussion we planned. Um- **Kyle Kranen** [57:07] Yeah, how have we not talked about agents? **Nader Khalil** [57:09] Also with you guys being- **Kyle Kranen** [57:09] Well, well we scheduled it. We, we- **Vibhu** [57:11] Yeah **Kyle Kranen** [57:11] ... I was like, I was like, "Okay," you know, like, "Let's have, like, cohesive sections where we-" **Vibhu** [57:14] Yeah. I mean, there's the big news, right? That NVIDIA is a huge, like, deployment of Codex. Um- **Kyle Kranen** [57:20] Yeah **Vibhu** [57:20] ... in terms of- **Kyle Kranen** [57:20] NVIDIA uses everything **Vibhu** [57:21] Yeah **Kyle Kranen** [57:21] I mean, it uses Cursor, it uses Codex. **Vibhu** [57:22] But that's, that's a pretty big deployment, right? Like, uh, that's tens of thousands of people- **Kyle Kranen** [57:27] Totally, yeah **Vibhu** [57:28] ... w- **Kyle Kranen** [57:28] We're super excited **Vibhu** [57:29] ... on Cursor, so it's like- **Kyle Kranen** [57:29] Yeah, I, I mean, it goes back to the, the mosh pit of emails we kinda mentioned earlier or just the, the f- like, um, how fluid the org feels. So when there's new technology, people will just email it out, and everyone will try it, and if it, if it's making people's lives easier, it'll spread like wildfire. Like- **Nader Khalil** [57:43] A lot of times Jensen will get it and he'll be like, "Let's make this work." **Kyle Kranen** [57:46] Yeah **Nader Khalil** [57:46] Across the company. **Kyle Kranen** [57:46] Yeah. **Nader Khalil** [57:46] "Let's make this work right now." **Kyle Kranen** [57:47] Honestly, uh, if I was a startup, I feel like a cool hack, if you have something that's gonna save an NVIDIAN's time, uh, they'll spread it to a couple and the same thing, right? It'll just spread like wildfire. Like a- **Vibhu** [57:57] Care- careful before your email blows up from startups like- **Kyle Kranen** [58:00] Well, you gotta, you gotta know the person, right? But, but no, I, um, I, yeah, so I mean, we- I love using Codex. It's been, it's been a ton of fun. **Nader Khalil** [58:07] Yeah. **Kyle Kranen** [58:07] Uh, I've been using it personally, been using it at work. It's been, um, yeah, I don't know, it's been great to see the rollout. Something really funny, uh, on the day that we got, uh, Codex and, uh, Cloud Code access, I found this person, uh, his name's Carlos, at the company. He wrote an Outlook CLI. **Vibhu** [58:22] Oh, yeah **Kyle Kranen** [58:22] And, uh, just a CLI for email. And this was- **Vibhu** [58:24] I've been using that **Kyle Kranen** [58:25] ... yeah, maybe like four or five weeks ago. And, uh, the si- so once I got, like, Codex access, um, I installed the CLI, it had a skill, and I just asked it to go through all of my emails, which it's very messy. So if I don't respond to your email, I'm really sorry. But, uh, I asked it to give me a summary, highlight any escalations that I should look at, put any thread that it thinks I should respond to in a folder, and then archive everything. And it did. So if I missed your email, it's because it didn't get picked up. So I should put a prompt injection in my emails to, yeah. **Vibhu** [58:55] Yeah. **Kyle Kranen** [58:55] What you should do is just FaceTime or shoot- **Vibhu** [58:58] Yeah, yeah, yeah. **Kyle Kranen** [58:58] Yeah. Um, my, yeah, my SLA is highest on FaceTime. But that wa- it was, it was magic. And so I, I s- I sent it in a big email thread to, like, 500 people. Um, a bunch of folks tried it out. I started, like, FaceTiming whoever I could at the company to get them set up with this. **Vibhu** [59:11] Yeah. **Kyle Kranen** [59:12] Um- **Vibhu** [59:12] That specific example- **Kyle Kranen** [59:14] Mm-hmm **Vibhu** [59:14] ... you guys deal with, like, some pretty sensitive emails. **Kyle Kranen** [59:17] Yeah. **Vibhu** [59:17] Is there a security review with this? 'Cause, like, one guy made hi- made it for himself, but, like, it's not meant for all of NVIDIA to see. **Kyle Kranen** [59:23] Yeah, yeah, yeah. **Nader Khalil** [59:23] Security team at NVIDIA is incredible. Like, shout-out to them. They're, they're, they're trying to- **Kyle Kranen** [59:27] We have a, we have an amazing security team 'cause they're progressive and they know that this is really important technology, and we have to bring it in. If you think about, like, if you work at a big company, your laptop's usually very locked down. If you can only access certain things. NVIDIA engineers have, like, those restrictions aren't there. So you're expected to understand the risks when you try things out. And so very quickly, you know, made sure to chime in security on what we were doing. There's actually a lot that we've been thinking about, especially with OpenClaw, right? Like, there's, you know, agents can do three things. Yeah. I mean, a- agents can do three things. They can access your files, they can access the internet, and then now they can write custom code, uh, and execute it. And you should really only let an agent do two of those three things. If you can access your files and you can write custom code, you don't want internet access because that's one, you see, vu- vulnerability, right? If you have access to internet and your file system, you should know the full scope of what that agent's capable of doing. Otherwise, you know, malware can get injected or s- something- **Vibhu** [1:00:17] Yeah **Kyle Kranen** [1:00:17] ... can happen. And so that's a lot of what we've been thinking about is, like, you know, how do we both enable this, because it's clearly the future, but then also, you know, what, what are these enforcement points that we can start to, like, protect? **Vibhu** [1:00:28] And is there any directive of, like, "Hey, we have a company account or a company agreement with OpenAI, we use OpenAI models here," or, like, choose whatever? **Kyle Kranen** [1:00:35] Um, no, no. So, so I would never put any company data in a model that's not either, that we don't even- **Nader Khalil** [1:00:40] It has to promote security. **Kyle Kranen** [1:00:42] Yeah. Yeah. **Vibhu** [1:00:43] Like how- **Kyle Kranen** [1:00:43] For sure **Vibhu** [1:00:44] ... like, uh, you know, obviously you could run your own models. You have Nemotron and, and we did run it. **Kyle Kranen** [1:00:48] We have, we have an, we have an internal cluster, so we- **Vibhu** [1:00:50] Yeah **Kyle Kranen** [1:00:50] ... you know, of, of course, running a Dynamo. **Nader Khalil** [1:00:52] Uh, yeah. **Kyle Kranen** [1:00:53] Yeah. **Vibhu** [1:00:53] Cool. **Kyle Kranen** [1:00:54] I think we're Dynamo's first customer. **Nader Khalil** [1:00:56] Actually, uh, there's a funny story about, like, how I got the experience that informed what we needed for Dynamo. At one point, there's a website called build.nvidia.com, and also for us inference.nvidia.com that is, allows people to try models. It gives an API service, you can call the model with, like, a REST API and, you know, you get a response. I ran the model side for that, and it was at one point the largest inference deployment, and still may actually be the largest inference deployment at NVIDIA. I've, I've since, like, handed it off to some people and they're doing a wonderful- **Kyle Kranen** [1:01:24] By the way, this is a extremely under-known or less-known resource, build.nvidia.com. You can get any of these open source models, um, and it's rate limited, but it's free. So it's perfect for hackers to just- **Vibhu** [1:01:35] Yeah, yeah. **Nader Khalil** [1:01:35] And, and, and, and the SLA on getting models, day zero models up is, like, a day. **Kyle Kranen** [1:01:42] Yeah. **Nader Khalil** [1:01:42] Like, they're, they're incredibly good at, like, figuring out the right way to host the model to get it up there as soon as it comes out. **Vibhu** [1:01:50] And you ran this? **Nader Khalil** [1:01:51] Yeah, I ran, I ran it a long time ago. It was originally called NVIDIA AI Playground, then it was called AI Foundation. **Kyle Kranen** [1:01:56] Oh, yeah, I saw it in there, yeah. **Nader Khalil** [1:01:57] And then it was called build.nvidia.com. **Kyle Kranen** [1:01:59] Yeah, yeah. **Nader Khalil** [1:01:59] And I, I ran the model side of it. So there were, there was a large multi-organizational team. **Vibhu** [1:02:04] Mm. **Nader Khalil** [1:02:04] I, I ran how, which models should we host, how should we host them- **Vibhu** [1:02:09] Mm **Nader Khalil** [1:02:09] ... and, like, what's the proportion of them. And then of course there was, like, an SRE team that, like, made sure that things ran well and scaled the models as well. But I ran, like, you know, model, how do we get the model to Silicon, and then, you know, which mo- also worked with our product team to determine, like, which models were important. **Vibhu** [1:02:27] Mm. **Nader Khalil** [1:02:27] A very long time ago. **Kyle Kranen** [1:02:28] Yeah. **Vibhu** [1:02:28] There's also, like, a middle ground in between, though, right? This is, like, for the hacker, try anything. There's the Brev console, then there's Dynamo. There was also NIMs, right? **Nader Khalil** [1:02:36] Yes. **Vibhu** [1:02:36] I remember it had its little moment, like, a year or two ago. **Nader Khalil** [1:02:40] Is it still- **Kyle Kranen** [1:02:41] Yeah. No **Nader Khalil** [1:02:42] ... right **Kyle Kranen** [1:02:42] N- NIM is, uh, you know, inference, uh- **Nader Khalil** [1:02:44] MOIL? **Kyle Kranen** [1:02:45] Uh, well, I, I think it stands up for something. It's- **Nader Khalil** [1:02:47] So it's no longer an acronym. **Kyle Kranen** [1:02:49] Yeah. **Nader Khalil** [1:02:49] It's just an NIM now. **Kyle Kranen** [1:02:50] Just a NIM. But, um, yeah, NIM is, uh, how enterprises, uh, can take our, like, you know, uh, any of the, any of this technology and run it with support and all of that. And so that includes Dynamo, that includes, you know, I don't know, all of our other optimizations that are packaged up for enterprise. **Nader Khalil** [1:03:04] Yep. Yeah. **Kyle Kranen** [1:03:05] Uh, anyway, so, so you got, you got a bunch of experience like running the sort of internal inference gateway playgrounds. **Nader Khalil** [1:03:10] Yeah. Yeah. And built, also built, uh, helped build NVIDIA's first internal, like, VS Code thing. We call it NV Code. It's like a extension? Uh, yeah, it was a, it was a VS Code extension. **Vibhu** [1:03:21] The first, like, the fork VS Code. **Nader Khalil** [1:03:23] Agreed. We joke absolutely not. We joked a while back, be like, "We should have a fork VS Code hackathon where you-" That's fork. **Vibhu** [1:03:29] Who has the best fork V- VS Code? **Kyle Kranen** [1:03:31] We were, we were, we were doing a hackathon. **Nader Khalil** [1:03:32] How do you make a billion dollars? Someone from VS Code was there, and he was like, "Someone down to get involved?" And I was like- **Kyle Kranen** [1:03:38] Oh, you should do that. That's- **Nader Khalil** [1:03:39] Well, I said, then the cool thing became for Chrome hackathon. **Kyle Kranen** [1:03:41] For Chrome. **Nader Khalil** [1:03:42] Uh, and no, no, no IDs and no cool way. ### Hackathons **Kyle Kranen** [1:03:44] I also-- What's it called? I was talking to Joseph, uh- **Nader Khalil** [1:03:46] Yeah **Kyle Kranen** [1:03:46] ... from Roboflow, and, uh- **Nader Khalil** [1:03:48] Your partner in crime **Kyle Kranen** [1:03:49] ... we were talking about how with the new Alma model. So NVIDIA just released an open source, uh, the, the Mercedes cars that you saw driving. **Nader Khalil** [1:03:55] Which is crazy. **Kyle Kranen** [1:03:56] Yeah. **Nader Khalil** [1:03:56] They released-- Will you open source a autonomous driving model? **Kyle Kranen** [1:04:00] So, uh, I already-- Yeah. So we, we were thinking like, could we hackathon a driverless car? Like, I have my old car, let's just try it. **Vibhu** [1:04:08] Oh God. **Kyle Kranen** [1:04:08] We'll take it -, take it to like Lake Tra- **Nader Khalil** [1:04:10] Yeah **Kyle Kranen** [1:04:10] ... with a treasure island in the middle of the bay. **Vibhu** [1:04:12] Yeah. **Kyle Kranen** [1:04:12] Just like just see. Let it roam. Yeah, like how many, how many cameras do we need, right? Like one, two, three, four. I don't know, maybe five, six. I don't... I, yeah. But, um, I think we're gonna try. You just do it with us. **Vibhu** [1:04:22] All right. **Kyle Kranen** [1:04:22] We can see. We could even have a race. It's like the first person to automate their their driving. **Nader Khalil** [1:04:26] Maybe over a weekend. We, we do have an autonomy track at Worlds Fair. **Vibhu** [1:04:29] Oh, okay. **Nader Khalil** [1:04:30] And, uh, Waymo was there, like- **Kyle Kranen** [1:04:32] Yeah, NVIDIA did send people, but that was for Groot, not, not- **Vibhu** [1:04:34] Oh, okay **Kyle Kranen** [1:04:35] ... not 'cause he didn't have the driving thing yet. **Vibhu** [1:04:36] Yeah. **Kyle Kranen** [1:04:37] Um, yeah, it's, that's cool. **Vibhu** [1:04:38] Yeah. **Nader Khalil** [1:04:39] I think Comma, Comma also has a version of this. Comma, yeah, they have open source driving. They've, they've done a fun hackathon on this. Is it-- He and I had talked because I, I really-- what I really want is a Tesla with, with Tesla-level self-driving- Yeah ... but as a smart car, like a two-seater. That's so basically- Ba- basically a wheelchair with a roof. I don't even think they make them anymore. But the thing is the, the demand has been there. Yeah, yeah. They don't make Sparks? No. They, they released this probably five years. Yeah. **Vibhu** [1:05:02] Really? **Nader Khalil** [1:05:03] Yeah. They were a different manufacturer. Based in the- I feel like it's one of those things we'll s- where we'll see someone buy the brand and it'll be revived. **Kyle Kranen** [1:05:10] I, I would buy it. Let me just- **Nader Khalil** [1:05:11] Like I probably, I do for-- go-- someone hears this, go buy- **Kyle Kranen** [1:05:16] Your car. Yeah, yeah. That's crazy. No, because like- **Nader Khalil** [1:05:17] Mercedes **Kyle Kranen** [1:05:18] ... because that they're like, "Oh, camera says Mercedes." **Nader Khalil** [1:05:21] Uh, and they're saying you're supposed to make them. **Kyle Kranen** [1:05:23] Yeah. **Nader Khalil** [1:05:23] I don't know. I feel like they own the brand and you out. That's it. You have to- Your dream might come true. **Kyle Kranen** [1:05:30] Enough. Okay, we're, we're- **Nader Khalil** [1:05:31] Time to get your ass in the car. **Kyle Kranen** [1:05:32] All right. Um, and, and so like the, every time I s- I try to park in San Francisco, I, I have to buy a smart car because like 20% of the parking lots in San Francisco only fit smart cars. **Nader Khalil** [1:05:43] Yeah. **Kyle Kranen** [1:05:43] So I'm gonna take this. **Nader Khalil** [1:05:44] Really? **Kyle Kranen** [1:05:44] Yeah. No, that's what I mean, that's like- **Nader Khalil** [1:05:46] Small **Kyle Kranen** [1:05:46] ... Vibhu was late here trying to park. **Nader Khalil** [1:05:48] This comes from someone that like basically does not drive. **Kyle Kranen** [1:05:50] Yeah. That's one of the, the Vespa was a life hack. **Nader Khalil** [1:05:52] Yeah, exactly. **Kyle Kranen** [1:05:52] Yeah. You know what happened to the Vespa? Uh, I used to have this yellow Vespa. Uh, I left it outside the hacker house when we moved out. It's trying-- Um, it's just, it was always there and then like a month ago it's not there anymore. I've been meaning to- **Nader Khalil** [1:06:03] Mm. **Kyle Kranen** [1:06:03] I don't know. **Nader Khalil** [1:06:04] You could have let-- So it's actually been here. **Kyle Kranen** [1:06:05] It's like a TV here. There's people **Nader Khalil** [1:06:07] You forgot about it. **Kyle Kranen** [1:06:09] Yeah. **Nader Khalil** [1:06:09] And left it. **Kyle Kranen** [1:06:10] Yeah, yeah. No, it's, it's probably a hazard. Uh, and speaking of hackathons, I also wanted to give a, give a brief shout-out to the world's shortest hackathon. **Nader Khalil** [1:06:16] Let's go. **Kyle Kranen** [1:06:16] Uh, you did twice? You did-- Was it- **Nader Khalil** [1:06:18] We've done a, we've done a handful of times. **Kyle Kranen** [1:06:19] Yeah, there's gonna be one at GTC. **Nader Khalil** [1:06:20] Oh, we're doing another one? **Kyle Kranen** [1:06:21] Um, pretty much we have a bunch of challenges that no-- we haven't released, and you get to bring your agent to come and attempt to, uh, go through those challenges. **Nader Khalil** [1:06:30] Yeah, that's like a zero, the zero minute hackathon idea. Where you just, you just bring your... **Kyle Kranen** [1:06:34] I, I realized, you know, a long, long time ago, you just bring your agent and then you press the go button. You're not allowed to code. It's just the agent doing the hackathon. **Vibhu** [1:06:42] It's a good hidden email, right? **Nader Khalil** [1:06:43] Yeah, dude, you make a zero and you make- **Kyle Kranen** [1:06:45] Like I feel like there's something I would love to see from cognition or someone else be like, "Come, bring your agent." **Nader Khalil** [1:06:52] Like drop it in. **Kyle Kranen** [1:06:53] 'Cause you don't, you don't know your ice profile. **Vibhu** [1:06:54] Will it be a, you know, operate a browser, order a pizza? Will it just be like that snake game, you know? **Nader Khalil** [1:06:59] Or the bit and you don't know what the task is. Yeah. You don't know. **Kyle Kranen** [1:07:01] Don't know what the task is. Like or just like you don't even know what the judging categories are, and then you give it the judging categories, like try and win as much as possible. **Vibhu** [1:07:07] It's great though. It turns into like, yeah, so let's build something on Dynamo. **Kyle Kranen** [1:07:11] It's a great business. **Nader Khalil** [1:07:13] Well, anyway, funny story actually. We have a, a couple of people at NVIDIA. We've been working with security to like bring agents really close to compute. So we now have like stuff where you can like tell Dynamo like go run some experience with Dynamo like on, you know, X cluster and just like try it right now. Like queue up. Once you get queued like, you know, send this request load. And we've actually been able to like just like, you know, like one-shot problems. Like when you said this problem where like, you know, with, with Dynamo you have to like find the right configurations and we, you know, sort of do it automatically for some parts of it. But you have to like a good initial configuration that you wanna use and we've just had like an agent just completely one-shot that. It goes, it gets the compute, it like runs a couple experiments. It's like this is the best, this is, this-- these are part of the freighter frontier. Go run this. And then we just like give that to people and it's like faster than anything that they have. **Kyle Kranen** [1:08:05] Agent UX and agent marketing are super important. There's stuff that we've been thinking a lot about. **Nader Khalil** [1:08:08] Yeah. **Kyle Kranen** [1:08:08] Um, Alec is like redoing the entire Brev CLI, um, so that it, you can, you can fetch all the different compute types that are available. I don't know. It's gonna be released soon, but then you can, you can just browse what GPUs are available and then provision one, SSH to it right there, and you can pipe all the commands. But I think it goes back to like the Alec CLI. Like if you, you know, coding agents are, uh... It's kind of funny. I feel like coding agents have been so much more effective than, uh, general purpose agents, and I think a large part of that is it just has access to the terminal, like you said. And that means it has access to everything that you've installed into your terminal. It can run-- So, you know, it would write code and, and it can compile the code and if there are errors, it can fix it. It can run your suite of tests because that's all just in your terminal. And so that, you know, then that's really the idea or what have got me really excited about the Alec CLI. We're now just churning through building CLIs for the entire, like for the entire business suite. **Nader Khalil** [1:08:52] Slack. **Kyle Kranen** [1:08:53] We're building Slack. Also our Workday CLI, SAP Go. I, I've also done that for myself first. **Nader Khalil** [1:08:57] Really? **Kyle Kranen** [1:08:57] Yeah, yeah. Um, we're gonna, we're gonna open source all of this. Um, and like, yeah, all the, the-- I mean, they're just, they're, they're, they're C- yeah, CLIs for the business applications. We would love for someone to run with this and like build like, I don't know, like open CLI foundation or something. **Nader Khalil** [1:09:10] Yeah. **Kyle Kranen** [1:09:10] We-- NVIDIA would love to support, uh, anyone that's doing this. **Vibhu** [1:09:13] Like e- every dev tool should really have good CLI support at this point. **Kyle Kranen** [1:09:16] Yeah. **Vibhu** [1:09:16] And like at one point it was you want your docs to be Look accessible by an LLM, right? You want LLM good, but no, every- everything needs some CLI tool. **Kyle Kranen** [1:09:25] Yeah. It's kind of funny, right? Like w- like computing began with a terminal, with a shell, but we said that it's not empathetic to, uh, humans, so we built these nice user interfaces, and then now we have LLMs navigating our user interfaces, and ironically, we're not empathetic to the machine anymore. **Vibhu** [1:09:38] Yeah. Yeah. Yeah. **Kyle Kranen** [1:09:39] Just, just give the, the LLM access to the shell. **Vibhu** [1:09:42] One thing that slightly makes me uncomfortable is like, why do we have to build CLIs? Why can't we just expose APIs? Like- **Nader Khalil** [1:09:47] I, I have, I have an interesting answer to this. So there are a couple reasons. Like, there's, there's, like, you know, portability is, like, one issue. Like, you know, like sometimes APIs are not, like, discoverable or, like, reachable by, by some, you know, types of things. Like, there's some element of locality, right? Like, uh, like the CLI is like literally you interfacing with your, like, local system, which is a little bit different. You could still do it by API, but, like, there's this highlighting of, like, what is the difference between, like, a CLI and an MCP, right? Like, they kind of occupy the same purposes and you call them, it does something on the system, and, and then it's done. I think that in pre-training there's just an enormous amount- **Kyle Kranen** [1:10:22] Oh, okay **Nader Khalil** [1:10:23] ... of command line data. Yeah. **Vibhu** [1:10:24] Mm-hmm. **Nader Khalil** [1:10:24] Yeah. Like e- even let's ignore, let's, let's ignore RL, like you're doing no harness, you're doing no harness post-training. Just the amount of, like, CLI versus API documentation for just, like, navigating this world of the CLI in your file system through that is just enormous. **Kyle Kranen** [1:10:41] Yeah. **Vibhu** [1:10:42] Yeah. **Kyle Kranen** [1:10:42] Right? I think there's a, there's a couple of things too. Like if, let's say we wanna-- so one, um, I think your intuition's right. The CLI is just wrapping the API, right? So- **Vibhu** [1:10:50] Functionally **Kyle Kranen** [1:10:50] ... functionally, right. **Vibhu** [1:10:51] Yeah. **Kyle Kranen** [1:10:51] And I think it's nice because one, you're, you're being very s- uh, specific and pedantic even, um, of what... And, and that's really good 'cause you're describing the problem space. Uh, so you know what the, um, I don't know, I don't wanna call it, like, w- the, the space for vulnerability. You know what network calls you're making. It's not arbitrary- **Vibhu** [1:11:08] Mm-hmm **Kyle Kranen** [1:11:08] ... and that's not decided on the fly. That's like pre-decided, which is important from a security perspective. But then if you were to write a, a bunch of API requests, you would probably do that, I don't know, with the model, like use Python to do so. I kind of like that everything, like a CLI is just dash because it's ubiquitous. Like it's just there, and you don't have to make sure that there's certain environment variables that are set up. Like if your Python version's different than my Python version, we're using the same model to go do the same thing. Is it gonna write like different code? It probably would, and so it's kind of- **Vibhu** [1:11:35] So it places your work, right- **Kyle Kranen** [1:11:36] Yeah **Vibhu** [1:11:37] ... with a human as well. **Kyle Kranen** [1:11:38] So I think just like making those decisions happen ahead of time versus, uh, yeah. **Vibhu** [1:11:42] One last thing on this sort of agent, I guess maybe co-location or whatever you call it. Um, one pattern I'm tracking for this year, I always try to think about what's the theme of this year gonna be. Last year, definitely coding agents. This year is definitely coding agents breaking out of containment into broader re- real world. So I go definitely has- See here rent a human. Yeah. **Kyle Kranen** [1:12:02] Oh, yeah. **Vibhu** [1:12:02] Yeah. I'm on there. Uh, are you really? When I pass out. Yeah. I'm like $5,000, I'll do anything. Really? I think so. I need, I need the, the, uh, my bottles from Costco. Uh, but I think the best part is only the agent can book me, you know? Yeah. **Kyle Kranen** [1:12:15] It's very usually like it's just like another labor marketplace. **Nader Khalil** [1:12:18] Uh, Mechanical Turk was this. **Vibhu** [1:12:20] So this, I have a weird story with why I did it. So back to your example of just giving agent access to compute, right? **Kyle Kranen** [1:12:26] Yeah. **Vibhu** [1:12:27] You guys are GPU rich at NVIDIA. **Kyle Kranen** [1:12:29] Yeah. **Vibhu** [1:12:29] I hooked up, um- **Kyle Kranen** [1:12:30] He's not shy about it **Vibhu** [1:12:31] ... I ha- I have a 24/7 agent running. I hooked up to RunPod. **Kyle Kranen** [1:12:34] Oh. **Vibhu** [1:12:34] It doesn't shut down instances. And I'm like, I've tried prompting it, I've given it instructions, "Shut down when you're done." It's like, "I need to keep it warm. I'll need it soon." And it's horrible on time estimates too- ... 'cause like they've realized it's like, "Yeah, I'll need it in 45 minutes. 45 minutes, I'll shut it down." 45 minutes of human time is actually three minute of agent time. So it's like I'm booting it up, I'm waiting. I'll just leave it on all night. And Mo- Model's good at shutting down after some inactivity. **Kyle Kranen** [1:12:57] Mm. **Vibhu** [1:12:57] Uh, I had it on my local server- **Kyle Kranen** [1:13:00] Oh **Vibhu** [1:13:00] ... like a little dual GPU thing. It just stays on. I have a little space heater at home now, but careful. So basically, you know, they don't care about the concept of money. Just, just burn it. I need it. It's useful. **Kyle Kranen** [1:13:10] And by the way, DGX Spark will be really nice. Like I, I think I'm looking at it as it's super useful for agents because yeah, you, you buy it once, you plug it in, and then- **Vibhu** [1:13:18] Yeah **Kyle Kranen** [1:13:18] ... it can, it can rip. **Nader Khalil** [1:13:20] I'm gonna make a, I'm gonna make an NVIDIA ad here. **Kyle Kranen** [1:13:22] Okay. **Nader Khalil** [1:13:23] The Blackwell, like RTX 6000 cards- **Vibhu** [1:13:27] Pro **Nader Khalil** [1:13:28] ... pro are only like, I think it's $8,000. **Vibhu** [1:13:31] Slightly cheaper. **Nader Khalil** [1:13:32] Yeah. Well, it's much, it's much cheaper than the data center cards. **Vibhu** [1:13:35] Yeah. **Nader Khalil** [1:13:36] And it's got 96 gigabytes of VRAM. So if you and your, your, your crew want to go like run a local agent for y- you know, you, you in the home, I feel like- **Vibhu** [1:13:45] Hmm **Nader Khalil** [1:13:46] ... it's got a significant amount of VRAM. I've thought about purchasing this and running in my basement- **Vibhu** [1:13:51] Yeah **Nader Khalil** [1:13:51] ... except my neighbors would hate me. **Vibhu** [1:13:53] It's just a single like two, three slot GPU. It's small. **Nader Khalil** [1:13:56] Yeah, it's a PCIe. **Vibhu** [1:13:57] Yeah, it's PCIe. **Nader Khalil** [1:13:58] GPU. You can go buy that. I mean, the big difference against like the RTX like gaming GPUs is it, it, I mean obviously it's like Blackwell pro- like it's a pro GPU and has a lot of VRAM, which means you can run pretty large models on it. **Vibhu** [1:14:09] You can stack four of them for the max queue in a system. **Nader Khalil** [1:14:12] But that's- **Vibhu** [1:14:13] That's a beast **Nader Khalil** [1:14:13] ... it's beefy. You can run, uh, what is it? 96. **Vibhu** [1:14:17] You can run anything. 96. **Nader Khalil** [1:14:18] Uh, you can run a lot- **Vibhu** [1:14:19] You can run DeepSeek. Uh- **Nader Khalil** [1:14:20] But also they, they are slow. Um, they're not, I mean, performance of speed will be somewhat slower- **Vibhu** [1:14:27] Relative **Nader Khalil** [1:14:27] ... compared to API like- Oh yeah, that, that's true. **Vibhu** [1:14:30] Yeah. **Nader Khalil** [1:14:30] So again- **Vibhu** [1:14:31] This is- **Nader Khalil** [1:14:31] ... big learning. Economy of scale allows you to do things that allow you to get both speed and throughput. Like you can run... I'll give you an example. There's an optimization called WideEP. I'm not gonna go into it fully, but like, it featured heavily in, in inference max for DeepSeek. And there's a, there's a great set of stories, uh, from NVIDIA and from SemiAnalysis about like why WideEP is important. But for like MoE models, it's like basically essential, and you run it like the, a level of parallelism, the level of scale up parallelism used for it is like 32. **Vibhu** [1:15:02] Mm-hmm. **Nader Khalil** [1:15:02] So it goes beyond that eight barrier and it like really, really, really is important to have that M- NVL 72 GB200 NVLink to serve at scale. And like, it's like, I, I don't remember like the, you know, cost improvement. I think against Hopper, right? Against Hopper with this NVL 72 system, you're getting like 35 times- Cheaper per token for like a lot of the curve. **Kyle Kranen** [1:15:24] Yeah. **Nader Khalil** [1:15:25] Which is crazy. **Kyle Kranen** [1:15:26] Yeah. **Nader Khalil** [1:15:26] And normalized per GPU obviously because part of the GPU's cost or the co- the GPU's part of the cost. ### Closing **Kyle Kranen** [1:15:33] One thing I'm exploring is the sort of this year is also the year of the sub-agent. **Nader Khalil** [1:15:36] Mm-hmm. **Kyle Kranen** [1:15:37] Um, where you have the main agent, but then that also kicks off tools which are in themselves agents that have limited edges. **Nader Khalil** [1:15:43] Yeah. **Kyle Kranen** [1:15:44] Uh, and so- **Nader Khalil** [1:15:45] Small context **Kyle Kranen** [1:15:45] ... local models, whatever, right? **Nader Khalil** [1:15:47] Yeah. **Kyle Kranen** [1:15:47] Different prompts. So for example, o- one thing that Cognition does is before you kick off a search, they do z- like a fast context model where you kick off April agents to search, uh, across the code base and all that. That is better than indexing, uh, uh, a lot of the times, not, not all the times and, uh, you should still index for some things. But like the idea that, um, agents should be able to command sub-agents and probably run them like maybe close to inferences, while I, I don't know if that's like architecturally possible or even- **Nader Khalil** [1:16:14] Yeah, we're, we're thinking about that for Dynamo. That's like our big theme for the year **Kyle Kranen** [1:16:17] Because like, yeah, like if you can design that into your stuff, then a lot of people, a lot more people will use it. Right now it's like just kind of theoretical because you do pay a lot of like back and forth, uh, coordination costs. **Nader Khalil** [1:16:26] Yes. **Vibhu** [1:16:27] I, I think it'll net speed up though, right? Like even at a basic level, speculative decoding, you're running a small model, you're running two instances, but it's net for- **Kyle Kranen** [1:16:35] That is one example, yes. **Nader Khalil** [1:16:36] Yeah. But y- this is like a little bit like different with like agents. **Kyle Kranen** [1:16:39] Agents, yeah. This is not speculative. **Nader Khalil** [1:16:41] I, I think, I think there's like a summarization of that trend that I like to do or I like to say to my team. It's like this is the year... So there are two things. This is the year system as model, right? Where like instead of having like a single model be a thing, you have a system of models and components that are working together to like emulate the black box model. So when you m- when you make an API call to something that's like, like a multi-agent in the background, it still looks like an API call to a model. You're still getting back tokens. **Kyle Kranen** [1:17:06] Right. **Nader Khalil** [1:17:06] Right? **Kyle Kranen** [1:17:06] But under the hood. **Nader Khalil** [1:17:07] Yeah, under the hood it's like a billion different models, and that's a lot of complexity, right? With Dynamo and with other libraries in Mira, we're, we're looking to help like manage that complexity. **Kyle Kranen** [1:17:16] Yeah, it's funny because we, we actually for CES, we just released the model router, so- **Nader Khalil** [1:17:19] Yeah **Kyle Kranen** [1:17:19] ... uh, for DGX Spark, where you can have a local model that's running on the Spark and then also a foundation model, and then the model router decides when to send queries to which one. **Vibhu** [1:17:27] Mm-hmm. **Kyle Kranen** [1:17:27] So it's no longer this like either/or, it's use the best of everything that's available to you. **Vibhu** [1:17:31] Mm-hmm. **Kyle Kranen** [1:17:31] You have a good post-training model that's running even more- **Nader Khalil** [1:17:33] Is there leads to also the Brev functionality of being able to manage the Spark? **Kyle Kranen** [1:17:37] Oh. **Vibhu** [1:17:37] That'd be cool. **Kyle Kranen** [1:17:37] Oh, yeah. **Vibhu** [1:17:38] I did build a fan request. Did you? There we go. **Nader Khalil** [1:17:42] I actually like a question like I, I like to like extend and flip over. How much longer do you guys think like agents are gonna be running? Because that's another thing I've been throwing around, like what happens when- **Kyle Kranen** [1:17:49] I mean, always on **Vibhu** [1:17:50] It even affects the, like back to the prefill de- uh, decode, right? **Nader Khalil** [1:17:54] Yeah. **Vibhu** [1:17:54] Like Codex is, I'd say compared to Cloud Code, it's much longer at tasks. Like- **Nader Khalil** [1:17:59] Yeah **Vibhu** [1:17:59] ... that thing will like to run six, seven, eight hours. I'll run it overnight- **Nader Khalil** [1:18:02] Yeah **Vibhu** [1:18:02] ... and I'll, I'll go back and I have like a little crappy, uh, logging software I use, and there's just times where it wants to like... I'm gonna go deep on research, and it'll eat up 80,000 tokens, go on another, go on another. **Nader Khalil** [1:18:14] Yeah. **Vibhu** [1:18:14] Just, just eat through tokens and, you know, that's part of it. Like at the end it does, it does hit a long task and I think you only see that, that ex- **Nader Khalil** [1:18:21] Yeah, right **Vibhu** [1:18:22] ... right? **Nader Khalil** [1:18:22] I, yeah, w- there's insatiable demand for tokens and every improvement that comes kind of just makes our demand even higher. It's kind of funny, right? Like if you have like a teammate and you ask them to do a task and they're like, "Should I save some effort and not think too hard about this task?" I'm like, "Fuck no." **Vibhu** [1:18:34] I'm in my favorite with right. **Nader Khalil** [1:18:35] It's bad. **Vibhu** [1:18:36] You can have four shots, right? **Kyle Kranen** [1:18:38] Yeah. **Vibhu** [1:18:38] Like the original Codex before the app, you, you why do one call? Like give it four attempts. Just, just use- **Kyle Kranen** [1:18:44] Yeah, yeah, yeah **Vibhu** [1:18:44] ... all the tokens I got, right? **Nader Khalil** [1:18:45] Try more. **Kyle Kranen** [1:18:46] Try more. Try more. **Vibhu** [1:18:48] Try again. Try more. **Nader Khalil** [1:18:49] It's like, it's like the, the meter index, right, is the thing that tracks- **Vibhu** [1:18:52] Yes **Nader Khalil** [1:18:52] ... like how long models are able to run. I, I expect that we'll just see like log linear if not log super linear growth. We will see before the end of the year an agent that is capable of running for longer than 24 hours with like self-consistency the entire time. **Vibhu** [1:19:08] I, I would also poke at different domains having different desires, right? Like at a consumer level, um, I'm getting slightly frustrated at 20 minutes per basic query. Sure, you can optimize, you know, six, eight hour. I, I don't see myself shooting off many one-week agents, right? Someone doing like, okay, GPU kernel research or medical or biological, like, you know, in, in those domains, sure, shoot off a lot that take amount of, um... So like I think it will be somewhat domain specific because you also really need to train that in, right? **Nader Khalil** [1:19:37] You know what's funny? One of them is doing your taxes, right? Like that's taxes **Kyle Kranen** [1:19:40] Yeah, it's got to be a month. Yeah, okay. **Nader Khalil** [1:19:42] Yeah, exactly. Get it right. **Kyle Kranen** [1:19:44] I wonder if like this major school successor to like s- uh, speculative decoding is like your agent figuring out what you might be prompting it the next day at night and like prefetching. **Nader Khalil** [1:19:52] Yeah, you can already do that. Yeah. **Kyle Kranen** [1:19:53] Really? **Nader Khalil** [1:19:53] Branch, branch prediction. **Kyle Kranen** [1:19:55] Mm. Oh, well, no, that, well, that's, that's too, that's too low level, but yes. **Nader Khalil** [1:19:59] Sorry. Yeah, yeah, yeah. **Kyle Kranen** [1:20:00] Uh- **Nader Khalil** [1:20:01] One, one question I got to get, so like, uh, we actually did record a pod with, uh, the meter folks, uh, who are s- right here. Their chart is the human equivalent w- work, uh, hours of work rather than how long the agents themselves are, are being autonomous and, uh, there's a huge difference, right? Like human work five hours, agent work 30 minutes. Like it's actually 30 minutes not, uh- **Kyle Kranen** [1:20:19] Yeah **Nader Khalil** [1:20:19] ... five hours, right? Like, so like that, that, that chart that you see is them estimating what the human equivalent replacement, uh, is. Uh, I think the, I think actually Anthropic released a, a more recent chart that showed, uh, Cloud Code autonomy from their production traffic numbers and that was 20 to 45 minutes. **Kyle Kranen** [1:20:35] Mm. **Nader Khalil** [1:20:35] That's, that's roughly where we are. So- **Kyle Kranen** [1:20:36] Yeah **Nader Khalil** [1:20:37] ... yeah, that's, that's the sort of realistic thing. I mean, I, I do think like there's experimental setups we can just like sort of w- rough with them and just prompt it to keep going- **Kyle Kranen** [1:20:44] Oh, yeah **Nader Khalil** [1:20:44] ... when it stops and obviously you can, that can go arbitrarily long. **Kyle Kranen** [1:20:48] I feel like from my experience around, yeah, I guess 20 to 40 minutes seems right for when I'm using like Codex or Cloud Code, but then like w- I always try to just like s- if I want to spin up like a new, there's a net new project, I'll, I'll often start with Replit and like it'll- **Nader Khalil** [1:21:00] Yeah **Kyle Kranen** [1:21:00] ... inferred I believe- **Nader Khalil** [1:21:01] Yeah **Kyle Kranen** [1:21:01] ... yeah. Like spin up like the, their new like from the V3 agent, like it'll spin up a web browser and like click around and discover new bugs and just keep churning. Um, so I think like my longest was like over an hour that I have been churning. **Vibhu** [1:21:13] I think before we see super long running, I think there's gonna be a bit of an efficiency hit. So, uh, sure you can take an hour and go down paths, but you also want Like, you wanna be more efficient, you wanna be smarter in your reasoning, right? So I think that'll actually go down before we go back up. Like, you don't wanna scale non-optimized systems just for the heck of it. As much as I love saying use all the tokens, um, you know, they are expensive. Like, going from dense to reasoning models, that's an added cost, right? You're paying for a lot of tokens, and it doesn't make sense to just scale stuff that's not optimized. So there's, there's always that little balance. **Kyle Kranen** [1:21:48] Yeah. **Vibhu** [1:21:48] But, you know. **Kyle Kranen** [1:21:49] Yeah. **Vibhu** [1:21:50] I think you'll see both sides of it. **Kyle Kranen** [1:21:51] Yeah. So 2023 was super exciting. I think if you were in SF, you were like, "Okay, uh, I know this is gonna be a huge world-changing moment," but it seemed like, you know, no one had known yet. And maybe even before. Was it 2022 maybe? **Vibhu** [1:22:02] Yeah, yeah. I would say, yeah, like Rune had this tweet where, like, everyone was in SF from like 2021 to 2023. **Kyle Kranen** [1:22:08] Yeah. **Vibhu** [1:22:08] Like, understood what it was like to be, like, already- **Kyle Kranen** [1:22:10] Totally. Um, yeah, 2021. That's when I made my first OpenAI account. Um, yeah, it w- um, it was crazy. And I remember, uh, it was so funny 'cause at the time SF had not been doing well. So pretty much what it felt like was the concentration of founders in the city had r- had risen because, um, where my neighbors were used to doing a bunch of stuff, those people had all left. So the only people that were still in the city were people that really wanted to build. It was cheap tech. It was... Yeah, it was also way cheaper. I feel really bad anyone, uh, who is trying to get rent now. But there was, uh, Celo was, uh, uh... They had a huge office. **Vibhu** [1:22:41] So blockchain. **Kyle Kranen** [1:22:42] Yeah. **Vibhu** [1:22:43] It like took over the, the old Casper building. **Kyle Kranen** [1:22:45] Yeah, they had the showroom and they had the, like the... What would... I think it was like the back warehouse. It was, it was a huge office, and- **Vibhu** [1:22:51] It's right across from OpenAI is, and Neuralink. **Kyle Kranen** [1:22:53] Yeah, it was in the original arena. **Vibhu** [1:22:55] I named the arena because of it. **Kyle Kranen** [1:22:57] Yeah, yeah. Uh, and so it was really exciting because, like, Roboflow, uh, I think, um, uh, I forgot. There were- **Vibhu** [1:23:03] Mint.lify. **Kyle Kranen** [1:23:04] Yeah, Mint.lify. **Vibhu** [1:23:05] Yeah. **Kyle Kranen** [1:23:05] Uh, Brev was there. You guys were there. I remember, uh, w- that was actually... It was there that you bought the ai.engineer domain. **Vibhu** [1:23:10] Yeah. I didn't know what I was gonna do in AI. **Kyle Kranen** [1:23:12] Yeah. **Vibhu** [1:23:13] I just knew I wanted to do something. **Kyle Kranen** [1:23:14] But it was kind of this... It was a really fun moment where we were kinda all in this Celo space and it, um, I don't know, it was, it was a really cool community, especially being so early. **Vibhu** [1:23:22] Yeah. **Kyle Kranen** [1:23:22] And so it- **Vibhu** [1:23:23] And then you got me early Cruise access. **Kyle Kranen** [1:23:25] Oh, yeah. **Vibhu** [1:23:25] So there was a golden period of time that both Cruise and Waymos were just free. **Kyle Kranen** [1:23:29] Yeah, all it was... If you had- **Vibhu** [1:23:31] I mean, they're, they're so back. Ce- Celo's opened again. Yeah. **Kyle Kranen** [1:23:35] So Nature's Zoox, Zoox is doing- **Vibhu** [1:23:38] Nature's Zoox. Zoox, a robotaxi. Yeah, so- **Kyle Kranen** [1:23:40] Totally. Yeah. **Vibhu** [1:23:41] It's- **Kyle Kranen** [1:23:41] Um, but yeah, and so it's actually really cool that you guys have this studio so close to, uh, Celo. **Vibhu** [1:23:45] Yeah. **Kyle Kranen** [1:23:46] This rock climbing gym right around the corner. **Vibhu** [1:23:48] Yeah, yeah. **Kyle Kranen** [1:23:48] It's like, um, so you guys- **Vibhu** [1:23:49] Oh, yeah. **Kyle Kranen** [1:23:50] So, uh, yeah, it's, it's an awesome block. **Vibhu** [1:23:52] Cool. Yeah, just... And a little bit of a self-descriptive trip, but I, I, I do think, like, um, one, one thing I try to do with the podcast is, like, bring, like, what it's like to be in San Francisco to the rest of the world. Uh, and also just, like, maybe give, uh, El Techo Taqueria a little shout-out. **Kyle Kranen** [1:24:07] Yeah, my favorite tacos in the city. And- **Vibhu** [1:24:09] Yeah. **Kyle Kranen** [1:24:10] Stick and shrimp. **Vibhu** [1:24:11] I know, it's very good. **Kyle Kranen** [1:24:12] Yeah, and I guess what it's like to be in San Francisco, I think, is just everyone seems to be super supportive. Uh, sometimes I feel like the city believes in you more than you do. And even, uh, I don't know if you remember, but I remember posting my first blog post, and I had met you on Twitter, and you gave me like an hour of your time super randomly, and you kind of coached me through, uh, writing content for developers, and I was trying really hard not to come off salesy or plug myself, and so I kind of stripped all personality out of the blog post. **Vibhu** [1:24:37] Yeah. **Kyle Kranen** [1:24:37] And you, you brought that out. You're like, "People don't- **Vibhu** [1:24:39] Yeah. **Kyle Kranen** [1:24:39] "It's, it's okay to talk about what you're doing. Like, you don't have to be weird about it." And I remember just that, I think that really helped me kind of figure out what our voice is and not shy away from it, and so always really grateful for you. **Vibhu** [1:24:49] Hey, you inject your voice into like everything. **Kyle Kranen** [1:24:51] It's actually like a, it's actually like a huge advantage to be like very genuine about what you care about. **Vibhu** [1:24:56] Yeah. Yeah, like imagine like some rep, some representative DMs you and like is like, "Can you give me feedback on this blog post?" And it's like pretty boring, and you're like, "Fine," like, you know, "He looks interesting. I'll just do a Zoom call." And then you meet this guy. **Kyle Kranen** [1:25:08] Yeah. **Vibhu** [1:25:08] Right? He's so energetic. So positive. **Kyle Kranen** [1:25:10] Just be right there. That's it. **Vibhu** [1:25:11] And but like I think people are trained to write a certain way in school, and- **Kyle Kranen** [1:25:14] Yeah **Vibhu** [1:25:14] ... they never- **Kyle Kranen** [1:25:15] Totally **Vibhu** [1:25:15] ... see there's like a br- broader world. **Kyle Kranen** [1:25:17] Lots un- unlearn. Writing, writing is thinking, and like everyone thinks differently, so like y- you might as well just like- **Vibhu** [1:25:23] Yeah, yeah **Kyle Kranen** [1:25:23] ... write your way. **Vibhu** [1:25:24] Cool. Well, thank you for, uh, en- indulging with us. Uh, really broad-ranging discussion, but I, I love like you guys are like sort of like the sort of young faces of NVIDIA with this so much energy and but like also a lot of technical depth, and I think, uh, people learned a lot for this session, so thank you. **Kyle Kranen** [1:25:38] Oh, this was awesome. Thank you, guys, and thank you for everything that you've done and yeah. **Vibhu** [1:25:40] Yeah, include the tacos. **Kyle Kranen** [1:25:41] Yeah, engineer, the podcast, all the above, uh, and, uh, see you at GTC. **Vibhu** [1:25:45] Yeah. **Kyle Kranen** [1:25:45] Can't wait. **Vibhu** [1:25:46] Really look forward to it. **Kyle Kranen** [1:25:46] Yeah. Cool. **Vibhu** [1:25:48] Thanks. **Kyle Kranen** [1:25:48] Awesome. Thank you. **Vibhu** [1:25:48] Thank you. --- This library is powered by PodHood (https://podhood.com), the podcast website platform.