MurmurCast
TechnicalDiscussion

Open Source Wins, AGI Is Here, and Scorsese's AI Toolkit with CEOs of Cerebras & Black Forest Labs

All-In with Chamath, Jason, Sacks & Friedberg1h 3m

Andrew Feldman (Cerebras CEO) and Robin Rombach (Black Forest Labs co-founder) discuss the massive AI infrastructure buildout happening globally, the emergence of reasoning capabilities in AI models, and the potential of generative AI for creative applications including filmmaking with Martin Scorsese, while debating the balance between open-source and proprietary models.

Summary

The episode explores the unprecedented scale of AI infrastructure investment, with Feldman noting that the capital, time, and talent dedicated to AI buildout rivals historical efforts like the Great Wall of China or war efforts. Cerebras is at the center of this with a $25 billion backlog of orders from companies like OpenAI, Anthropic, and Google, who are insatiably ordering chips before they're even finished manufacturing. Feldman explains that unlike typical boom cycles, demand isn't speculative—companies have already booked capacity and are worried about how to keep suppliers from leaving.

The conversation shifts to reasoning capabilities in frontier models. Feldman describes how newer AI systems understand user intent beyond mere prompt-following, exemplified by the host's experience with reasoning models that can debate themselves on where to find information and engage in extended thinking over 24-48 hours. Cerebras's inference chips enable faster computation for these reasoning-intensive tasks, with Feldman claiming their architecture breaks Moore's Law by achieving more than 2X improvements every 18 months—far exceeding the traditional 18-month doubling cycle.

On AGI and superintelligence, Feldman argues that by any definition proposed 20-50 years ago, we've already achieved artificial general intelligence. The challenge now isn't capability but deployment and alignment with human goals. He discusses the government's decision to slow O1/O5 release for red-teaming, comparing it to FDA oversight of pharmaceuticals—a reasonable precaution given the technology's power, as evidenced by Palo Alto Networks discovering previously unknown vulnerabilities in their security software within hours of model exposure.

Rombach presents Black Forest Labs' vision of multimodal generative models that combine image, video, audio, and action prediction—the same models serving dual purposes as creative tools for filmmakers and as foundational systems for robotic AI. He describes working with Martin Scorsese to visualize scenes for film production, positioning generative models as a medium for translating mental imagery into visual communication. He emphasizes the importance of human-in-the-loop workflows rather than fully automated content generation.

The discussion highlights the emerging trend of using generative AI in high-end film production, citing examples like a $30 million Bitcoin movie that would have cost $150 million with traditional sets. Rombach discusses licensing approaches for IP holders and the rise of fan-created content using AI—recreating untold Star Wars stories with millions of views. The episode concludes with both speakers optimistic about AI's potential to solve major problems like cancer while acknowledging economic disruption is inevitable but historically manageable.

About this episode

<p>(0:00) The AI Buildout: Datacenters Bigger Than Cities (Andrew Feldman)</p> <p>(1:50) Reasoning, Inference, and Breaking Moore's Law</p> <p>(16:28) Open Source, AI Sovereignty, and the Road to AGI</p> <p>(40:54) The Innovation Behind Generative Video (Robin Rombach)</p> <p>(47:31) Martin Scorsese, Robots, and the Future of Hollywood IP</p> <p>Thanks to our partners for making this possible!</p> <p>AppLovin Ads - AppLovin's AI advertising platform reaches over a billion daily active users across mobile games. Full-screen video ads with a 35-second median watch time. Advertisers are profitably spending hundreds of thousands of dollars a day and advertiser access is still in closed beta. The window is open at <a href="https://applovin.com/ALLIN">https://applovin.com/ALLIN</a></p> <p>Nasdaq - Industries, capital, and intelligence are converging into a single, interconnected system, and the infrastructure behind it needs to evolve just as quickly. Nasdaq was built for this moment: Powering more than 135 marketplaces and regulators globally and connecting capital to the companies shaping the future. As the innovation economy accelerates, connectivity becomes the critical asset. Nasdaq is the technology platform that makes it possible, and scalable. Learn more at <a href="https://Nasdaq.com">https://Nasdaq.com</a></p> <p>Follow Andrew:</p> <p><a href="https://x.com/andrewdfeldman">https://x.com/andrewdfeldman</a></p> <p>Follow Robin:</p> <p><a href="https://x.com/robrombach">https://x.com/robrombach</a></p> <p>Follow the besties:</p> <p><a href="https://x.com/chamath">https://x.com/chamath</a></p> <p><a href="https://x.com/Jason">https://x.com/Jason</a></p> <p><a href="https://x.com/DavidSacks">https://x.com/DavidSacks</a></p> <p><a href="https://x.com/friedberg">https://x.com/friedberg</a></p> <p>Follow on X:</p> <p><a href="https://x.com/theallinpod">https://x.com/theallinpod</a></p> <p>Follow on Instagram:</p> <p><a href="https://www.instagram.com/theallinpod">https://www.instagram.com/theallinpod</a></p> <p>Follow on TikTok:</p> <p><a href="https://www.tiktok.com/@allin">https://www.tiktok.com/@allin</a></p> <p>Follow on LinkedIn:</p> <p><a href="https://www.linkedin.com/company/allinpod">https://www.linkedin.com/company/allinpod</a></p> <p>Intro Music Credit:</p> <p><a href="https://rb.gy/tppkzl">https://rb.gy/tppkzl</a></p> <p><a href="https://x.com/yung_spielburg">https://x.com/yung_spielburg</a></p>

Key Insights

  • Cerebras has a $25 billion backlog of unfilled orders, and unlike speculative tech booms, this demand is already booked—companies worry about suppliers leaving rather than whether demand exists.
  • Feldman argues that reasoning models consuming massive amounts of tokens internally makes inference speed critical; slower computation would make extended reasoning impractical, creating a direct business case for fast chips.
  • The host demonstrated using a reasoning model with unlimited compute that autonomously debated where to source trending information, showing AI systems can now self-organize their approach to problems without explicit instruction.
  • Feldman claims Cerebras's new architecture achieves improvements greater than 2X every 18 months, breaking the historical Moore's Law trajectory of 2X every 18 months, because newer architectures have more room for optimization than mature designs.
  • By definitions proposed 20-50 years ago (Turing test, general intelligence benchmarks), artificial general intelligence has already been achieved; current debates are about deployment rather than capability.
  • The government's decision to stage O1/O5 release for red-teaming is framed as reasonable precaution comparable to FDA pharmaceutical trials—Palo Alto Networks discovered critical unknown vulnerabilities in hours of model exposure.
  • Rombach positions generative video models as dual-purpose tools serving both creative filmmaking (ideation with directors) and robotic AI (action prediction from the same trained model).
  • A $30 million Bitcoin film production used generative AI for scenery instead of building physical sets, reducing what would have been a $150 million production cost to something greenlit-worthy—demonstrating economic viability of AI-assisted production.
  • Feldman describes initial token overallocation at organizations similar to early AWS credit practices, where engineers generated waste until companies adopted strategic token management and cost discipline.
  • Black Forest Labs works with IP holders to develop custom models while implementing guardrails on public tools preventing generation of copyrighted IP, creating a licensing business model for content companies.
  • Fan-created Star Wars content using generative AI (Star Wars Stories Untold) is achieving millions of views per video, suggesting licensing fan creativity with IP protection could become a significant market.
  • Both speakers frame potential economic disruption from AI through historical precedent: horse-shoers and carriage builders faced displacement when cars emerged, but the net benefit was massively positive and should be weighed against gains like eliminating cancer deaths.

Topics

AI Infrastructure Buildout and Compute DemandReasoning Capabilities in Large Language ModelsCerebras Chip Architecture and Moore's LawAGI Definition and AchievementGovernment Oversight of Advanced AI ModelsGenerative AI for Film and Creative ProductionBlack Forest Labs' Multimodal ModelsOpen Source vs. Proprietary AI ModelsIP Protection and Licensing for Generative AIRobotics and Physical AI ApplicationsToken Maximization and Computational EfficiencyEconomic Displacement vs. Technological Benefits

Transcript

We are in the race for super intelligence. And Andrew Feldman is back. And obviously, CEO and founder of Cerebras doing inference chips, pioneered the space had a successful IPO. We've talked about this a couple of times we got to see each other in January at Davos IPO happens. The boys and I got to sit with you recently. That was fun at liquidity. It's really really fun. I had a great discussion with the boys and i got to sit with you recently that was fun uh at liquidity that was really really fun i had a great discussion with the boys but i wanted to deep dive with you about a couple of topics the first one…

Full transcript available for MurmurCast members

Sign Up to Access
View original source →

More from All-In with Chamath, Jason, Sacks & Friedberg

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.