{
 "episode_id": "all-in-2026-07-10-open-source-wins-agi-is-here-and-scorsese-s-ai-toolkit-with-",
 "title": "Open Source Wins, AGI Is Here, and Scorsese's AI Toolkit with CEOs of Cerebras & Black Forest Labs",
 "podcast": "all-in",
 "published_at": "2026-07-10",
 "speakers": [
  "Chamath Palihapitiya",
  "Andrew Feldman",
  "Robin Rombach"
 ],
 "summary": "In a two-part interview episode, Chamath Palihapitiya sits down separately with Cerebras founder Andrew Feldman and Black Forest Labs founder Robin Rombach. Feldman describes an unprecedented AI infrastructure build-out — data centers that will soon draw more power than the previous 50 years combined and a $25B Cerebras backlog — and argues the demand reflects real, enormous value despite experimentation; he champions more domestic open-source models, defends the government's staged-rollout ask on Anthropic's Fable model as reasonable, and contends that by any older definition AGI has already been reached, with recursive 'loop-maxing' gains compounding exponentially toward an abundant future (curing cancer, personalized AI tutors). Rombach explains how latent diffusion and Black Forest Labs' Flux models are converging into multimodal 'world models' that generate images, video, and audio and predict actions — the same model that can make a movie can serve as a robot's brain — and recounts working with Martin Scorsese, framing the tools as a creative medium best used with a human in the loop. Both guests land on the episode's thesis that open source has caught up to frontier models for most tasks and that AI sovereignty and control are now a defining enterprise and national trend. The conversation closes on IP and content, with Rombach favoring new creation over replicating existing IP and Chamath predicting a future where studios license their characters so fans can tell their own stories.",
 "topics": [
  {
   "name": "The AI build-out and Cerebras's demand backlog",
   "timestamp": "0:00",
   "summary": "Feldman describes the physical scale of the AI infrastructure build-out — data centers the size of football fields drawing more power than mid-sized cities, being built worldwide — and a $25B Cerebras backlog chasing demand that is already booked, not speculative."
  },
  {
   "name": "Token maxing: value vs. experimentation and disciplined deployment",
   "timestamp": "4:30",
   "summary": "Chamath and Feldman debate whether the compute spend is creating value; Feldman argues the net value is enormous despite waste, likening early unconstrained token use to AWS/Costco, and describes enterprises maturing into tiered deployment that routes ordinary work to cheaper open-source models."
  },
  {
   "name": "Reasoning as inference, intent understanding, and unlimited tokens",
   "timestamp": "6:19",
   "summary": "Chamath recounts running the Hermes agent on GLM 5.2 with unlimited compute and watching a reasoning model debate itself; the two discuss the leap from next-word prediction to understanding intent, and how unlimited tokens implies unlimited reasoning."
  },
  {
   "name": "Cerebras chips and breaking Moore's law",
   "timestamp": "12:45",
   "summary": "Feldman explains why fast inference makes token-heavy reasoning tractable, and claims Cerebras broke the traditional 18-month Moore's-law doubling, carving a new trajectory he expects to exceed 2x in the next 18 months."
  },
  {
   "name": "Custom silicon, dependency, open source, and sovereignty",
   "timestamp": "14:04",
   "summary": "Feldman argues hyperscalers build their own chips to avoid dependency and control their destiny; he says open source is having a moment, the US needs more domestic open-source models, and Cerebras already runs GLM, Kimi, Qwen, OpenAI, and sovereign customer models."
  },
  {
   "name": "Fable/5.6, AI safety, and red teaming",
   "timestamp": "20:05",
   "summary": "On the government's action against Anthropic's Fable model, Feldman calls a staged rollout with government red-teaming reasonable (analogous to powerful pharmaceuticals), cites Palo Alto Networks finding critical vulnerabilities, and laments how partisanship clouds clear thinking."
  },
  {
   "name": "AGI, superintelligence, and recursive learning",
   "timestamp": "29:14",
   "summary": "Chamath and Feldman agree AGI has effectively been reached by any older definition and discuss 'loop maxing' / recursive learning producing exponential, vastly-better answers, using multi-generational stonemason and Drosophila-genetics analogies for compressed learning."
  },
  {
   "name": "Feldman's optimist's ledger: abundance, cancer, education",
   "timestamp": "37:44",
   "summary": "Feldman argues that weighing pros and cons honestly, AI is a massive net benefit — a real shot at eliminating cancer deaths, personalized AI tutors, and abundance — with some economic dislocation comparable to cars displacing carriage makers."
  },
  {
   "name": "Black Forest Labs and multimodal world models",
   "timestamp": "41:03",
   "summary": "Rombach introduces Black Forest Labs (Freiburg + San Francisco), recounts inventing latent diffusion behind stable diffusion, and describes converging image/video/audio generation with action prediction so a single multimodal model can also serve as a robot's brain."
  },
  {
   "name": "Controllability, Scorsese, and AI in film production",
   "timestamp": "44:46",
   "summary": "Rombach explains exposing manipulation layers to defeat the 'slot machine' critique, recounts Martin Scorsese exploring the Flux models in person, frames generative AI as a creative medium best used with a human in the loop, and discusses AI-scenery film production (e.g., a Bitcoin movie)."
  },
  {
   "name": "World models, robotics, and action prediction",
   "timestamp": "54:17",
   "summary": "Rombach says the same multimodal model that makes a movie can be deployed as a robot brain, discusses combining perception, prediction, and simulation, and the goal of prompting robots in context rather than task-specific fine-tuning."
  },
  {
   "name": "IP holders, open source, and fan films",
   "timestamp": "58:02",
   "summary": "Rombach advises IP owners like Disney that new creation beats replicating existing IP and that blocking certain IP generation on public tools is sensible; Chamath predicts a future of licensed, fan-created stories, citing AI Star Wars fan films."
  }
 ],
 "entities": [
  {
   "name": "Chamath Palihapitiya",
   "type": "person",
   "salience": "central",
   "role_in_episode": "Host and sole interviewer for both segments; frames the open-source, sovereignty, AGI, and AI-film themes"
  },
  {
   "name": "Andrew Feldman",
   "type": "person",
   "salience": "central",
   "role_in_episode": "Cerebras founder/CEO; guest one on the AI build-out, inference chips, open source, AI safety, and AGI/abundance"
  },
  {
   "name": "Robin Rombach",
   "type": "person",
   "salience": "central",
   "role_in_episode": "Black Forest Labs co-founder/CEO; guest two on latent diffusion, Flux, multimodal world models, robotics, and AI film production"
  },
  {
   "name": "Martin Scorsese",
   "type": "person",
   "salience": "discussed",
   "role_in_episode": "Filmmaker collaborating with Black Forest Labs; explored the Flux models in person with Rombach"
  },
  {
   "name": "Sam Altman",
   "type": "person",
   "salience": "discussed",
   "role_in_episode": "OpenAI CEO ('Sam'); credited with foreseeing reasoning; cited in the Dario-vs-Sam competition point"
  },
  {
   "name": "Ilya Sutskever",
   "type": "person",
   "salience": "discussed",
   "role_in_episode": "Cited ('Ilya') for early warnings on AI safety and foreseeing exponential recursive gains"
  },
  {
   "name": "Dario Amodei",
   "type": "person",
   "salience": "discussed",
   "role_in_episode": "Anthropic CEO ('Dario'); central to the Fable safety/rollout discussion and frontier-lab competition"
  },
  {
   "name": "Jensen Huang",
   "type": "person",
   "salience": "discussed",
   "role_in_episode": "NVIDIA CEO ('Jensen'); described as reluctant to promote NVIDIA's own open-source models against its customers"
  },
  {
   "name": "Elon Musk",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Cited ('Elon') on rockets/near-zero launch cost and as a frontier-lab leader (Sam, Dario, Elon, Sergey)"
  },
  {
   "name": "Demis Hassabis",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Cited ('Dennis') among leaders who foresaw powerful recursive gains"
  },
  {
   "name": "Nikesh Arora",
   "type": "person",
   "salience": "discussed",
   "role_in_episode": "Palo Alto Networks CEO ('Nikesh') whose security test found critical vulnerabilities in a new model"
  },
  {
   "name": "Gal Gadot",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Actress; her AI-scenery Bitcoin movie is cited as an example of AI in film production"
  },
  {
   "name": "Warren Buffett",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Cited via a reinsurance analogy for inevitable, unpredictable data breaches"
  },
  {
   "name": "George Lucas",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Cited for permitting non-commercial Star Wars fan films and releasing lightsaber how-tos"
  },
  {
   "name": "Yuri Milner",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Host of the Breakthrough Prize event where Chamath spoke with Gal Gadot"
  },
  {
   "name": "AOC",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Invoked as a hypothetical 'President AOC' in the point about depoliticizing AI policy"
  },
  {
   "name": "Donald Trump",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Invoked as 'President Trump' in the same partisanship-of-AI-policy point"
  },
  {
   "name": "Cerebras",
   "type": "company",
   "salience": "central",
   "role_in_episode": "Andrew Feldman's inference-chip company; $25B backlog, wafer-scale chips, claims to have broken Moore's law"
  },
  {
   "name": "Black Forest Labs",
   "type": "company",
   "salience": "central",
   "role_in_episode": "Robin Rombach's open-source image/video/world-model company; makers of Flux and the Scorsese collaboration"
  },
  {
   "name": "OpenAI",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "Frontier lab; released OSS 120b, the jalapeno chip, model 5.6, and had a Sora/Disney relationship"
  },
  {
   "name": "Anthropic",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "Frontier lab whose Fable model triggered the government safety/rollout action discussed with Feldman"
  },
  {
   "name": "NVIDIA",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "GPU maker; its open-source models and the dependency/neo-cloud dynamic are discussed"
  },
  {
   "name": "Amazon",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "AWS analogy for early unconstrained cloud use; cited making its own chips amid insatiable data-center demand"
  },
  {
   "name": "Google",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Cited among the insatiable data-center buyers and frontier competitors (Gemini)"
  },
  {
   "name": "Microsoft",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Cited among the hyperscalers demanding more data centers"
  },
  {
   "name": "Palo Alto Networks",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "Nikesh Arora's security firm that found critical vulnerabilities when testing a new model"
  },
  {
   "name": "Disney",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "IP holder used as the example for how content owners should approach open-source vs. proprietary models"
  },
  {
   "name": "G42",
   "type": "organization",
   "salience": "mentioned",
   "role_in_episode": "UAE partner (with MBZ-UAI) whose sovereign models Cerebras runs"
  },
  {
   "name": "GlaxoSmithKline",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Pharma customer that wrote its own models and runs them on Cerebras"
  },
  {
   "name": "SpaceX",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Listed among the insatiable buyers of AI compute capacity"
  },
  {
   "name": "BitTensor",
   "type": "organization",
   "salience": "mentioned",
   "role_in_episode": "Distributed crypto project hosting the GLM 5.2 capacity Chamath used for the Hermes agent"
  },
  {
   "name": "Perplexity",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Cited ('Perplexi') for automatically suggesting a user's next prompts"
  },
  {
   "name": "Intel",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Cited as the x86 dependency the hyperscalers learned to avoid"
  },
  {
   "name": "Flux",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Black Forest Labs' open-source image model, used by startups and explored by Scorsese"
  },
  {
   "name": "latent diffusion",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "The algorithm Rombach and co-founders invented, foundational to modern generative image/video models"
  },
  {
   "name": "Stable Diffusion",
   "type": "product",
   "salience": "mentioned",
   "role_in_episode": "Prior generative image model Rombach and co-founders worked on before Black Forest Labs"
  },
  {
   "name": "GLM",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Open-source model (GLM 5.2 / ZAI); run by Chamath via BitTensor and among the models Cerebras hosts"
  },
  {
   "name": "Kimi",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Open-source model Chamath uses; he says it is indistinguishable from Claude for his tasks"
  },
  {
   "name": "Qwen",
   "type": "product",
   "salience": "mentioned",
   "role_in_episode": "Open-source model set among those Cerebras runs"
  },
  {
   "name": "OSS 120b",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "OpenAI's open-source model Feldman praises as a good domestic option"
  },
  {
   "name": "Fable",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Anthropic frontier model (with '5.6') at the center of the government safety/staged-rollout discussion"
  },
  {
   "name": "Sora",
   "type": "product",
   "salience": "mentioned",
   "role_in_episode": "OpenAI video model tied to a now-defunct Disney character-licensing relationship"
  },
  {
   "name": "Optimus",
   "type": "product",
   "salience": "mentioned",
   "role_in_episode": "Humanoid robots referenced ('a thousand optimists') in Chamath's Palace-of-Versailles thought experiment"
  },
  {
   "name": "Claude",
   "type": "product",
   "salience": "mentioned",
   "role_in_episode": "Anthropic model; Chamath contrasts 'blowing out' Claude tokens with the cheaper, comparable Kimi"
  },
  {
   "name": "open source",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "The episode's central thesis ('Open Source Wins') — open models closing the gap with frontier labs"
  },
  {
   "name": "AI sovereignty",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "Enterprises and nations retaining control of models, data, and compute — a recurring trend across both interviews"
  },
  {
   "name": "superintelligence",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "The race and definitional waypoints discussed with Feldman"
  },
  {
   "name": "AGI",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "Claimed by both Chamath and Feldman to already be effectively reached by any older definition"
  },
  {
   "name": "reasoning",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "Inference-heavy reasoning models and the shift from next-word prediction to understanding intent"
  },
  {
   "name": "recursive learning",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "'Loop maxing'; exponential compounding gains from iterating a model on its own outputs"
  },
  {
   "name": "token maxing",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "Unconstrained token consumption and the discipline of tiered model deployment"
  },
  {
   "name": "AI build-out",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "The unprecedented global data-center and compute mobilization Feldman details"
  },
  {
   "name": "Moore's law",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "The 18-month doubling Feldman says Cerebras broke and expects to exceed"
  },
  {
   "name": "world models",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "Multimodal action-prediction models bridging media generation and robotics — Rombach's core thesis"
  },
  {
   "name": "physical AI",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "Robotics powered by the same generative multimodal models"
  },
  {
   "name": "AI film production",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "Generative image/video for filmmaking — Scorsese, launch videos, and AI-scenery movies"
  },
  {
   "name": "AI safety",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "Cyber-weapon risk, red teaming, staged rollout, and the inevitability of a massive breach"
  },
  {
   "name": "export controls",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "The government's action on Anthropic's Fable/5.6 model that Feldman calls a wake-up call in Europe"
  },
  {
   "name": "job displacement",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "Economic dislocation from AI, which Feldman frames as real but outweighed by the benefits"
  },
  {
   "name": "education",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "Personalized AI tutoring as an example of a long-known-better method AI can finally deliver"
  },
  {
   "name": "healthcare",
   "type": "concept",
   "salience": "mentioned",
   "role_in_episode": "Curing cancer cited as a concrete near-term prize of the technology"
  },
  {
   "name": "generative video",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "Text-to-video and multimodal video generation, from 64x64 images to multi-minute HD"
  },
  {
   "name": "content licensing",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "How IP owners should license and control AI-generated use of their characters"
  },
  {
   "name": "Germany",
   "type": "place",
   "salience": "mentioned",
   "role_in_episode": "Black Forest Labs' home (Freiburg, in the Black Forest mountain range)"
  },
  {
   "name": "San Francisco",
   "type": "place",
   "salience": "mentioned",
   "role_in_episode": "Black Forest Labs' second office and hiring location"
  },
  {
   "name": "Texas",
   "type": "place",
   "salience": "mentioned",
   "role_in_episode": "Cited as a site of gigantic AI data-center build-outs"
  },
  {
   "name": "Europe",
   "type": "place",
   "salience": "mentioned",
   "role_in_episode": "Data-center build-out region; the Fable action described as a European wake-up call"
  },
  {
   "name": "China",
   "type": "place",
   "salience": "mentioned",
   "role_in_episode": "Source of Chinese open-source models and cheap/free-energy compute capacity"
  },
  {
   "name": "UAE",
   "type": "place",
   "salience": "mentioned",
   "role_in_episode": "Home of the G42/MBZ-UAI sovereign-model partnership run on Cerebras"
  }
 ],
 "claims": [
  {
   "holder": "Chamath Palihapitiya",
   "claim": "The scale of AI demand proves massive real value is being created, not just a bubble.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "AI build-out",
    "token maxing"
   ],
   "timestamp": "4:30",
   "quote": "There is clearly massive value happening."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "The net value of the AI build-out is enormous even though some of the spend will go nowhere.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "AI build-out",
    "token maxing"
   ],
   "timestamp": "5:11",
   "quote": "But it doesn't mean that the net value isn't enormous. It means some of it is going to go nowhere."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "Enterprises are maturing from unconstrained token use to disciplined, tiered deployment — routing ordinary work to cheaper open-source models and reserving frontier models for hard problems.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "open source",
    "token maxing"
   ],
   "timestamp": "5:11",
   "quote": "Over here, we can use maybe an open source model, maybe a cheaper model over here. And now we're sort of running like a business."
  },
  {
   "holder": "Chamath Palihapitiya",
   "claim": "Unlimited tokens effectively means unlimited reasoning.",
   "stance": "opinion",
   "conviction": 0.7,
   "subjects": [
    "reasoning",
    "token maxing"
   ],
   "timestamp": "9:44",
   "quote": "Because unlimited tokens, I believe, means unlimited reasoning"
  },
  {
   "holder": "Chamath Palihapitiya",
   "claim": "Open-source models have closed the reasoning gap with frontier models this year — Kimi is now indistinguishable from Claude for his use.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "open source",
    "Kimi"
   ],
   "timestamp": "16:09",
   "quote": "this open source started to figure out reasoning and the gap has suddenly closed this year."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "Hyperscalers build their own chips mainly to avoid dependency and control their destiny, not necessarily to make the fastest chip.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "NVIDIA",
    "Amazon",
    "OpenAI"
   ],
   "timestamp": "15:46",
   "quote": "mostly it's about an opportunity to control at least an important part of your destiny."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "AI demand is so high right now that no silicon will go unused.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "Cerebras",
    "AI build-out"
   ],
   "timestamp": "14:50",
   "quote": "there is so much demand right now that there is no silicon that will go unused."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "For the large volume of ordinary, non-inventive work (much of G&A), rock-solid open-source capability is sufficient — you don't need the most sophisticated frontier agents.",
   "stance": "opinion",
   "conviction": 0.7,
   "subjects": [
    "open source"
   ],
   "timestamp": "17:29",
   "quote": "What this needs is sort of rock solid open source capabilities."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "The US needs more domestic open-source models to give the world a real choice beyond OpenAI's OSS 120b and Chinese models.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "open source",
    "AI sovereignty"
   ],
   "timestamp": "18:37",
   "quote": "in the US, we need more domestic open source models. We need to give the world a choice"
  },
  {
   "holder": "Andrew Feldman",
   "claim": "OpenAI's release of the open-source OSS 120b was a good call and a good open-source model.",
   "stance": "opinion",
   "conviction": 0.7,
   "subjects": [
    "OpenAI",
    "OSS 120b"
   ],
   "timestamp": "18:37",
   "quote": "I think OpenAI made a good call releasing OSS 120b some months back. That was a good open source model."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "For a sufficiently powerful model, the government asking labs to stage the rollout and allow red-teaming is reasonable — analogous to how powerful pharmaceuticals are handled.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "AI safety",
    "Fable",
    "export controls"
   ],
   "timestamp": "21:56",
   "quote": "with a powerful new technology, it certainly doesn't seem unreasonable to say, hey guys, let's at least do some red teaming at the government. So we know our defenses can block this."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "Political polarization badly damages clear thinking about AI policy.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "AI safety",
    "export controls"
   ],
   "timestamp": "22:54",
   "quote": "I think the polarization hurts a great deal. It hurts clear thinking"
  },
  {
   "holder": "Andrew Feldman",
   "claim": "Intense competition between frontier labs (Dario vs. Sam) is good for the technology, and regions shouldn't reflexively rush to regulate it first.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic",
    "OpenAI"
   ],
   "timestamp": "23:12",
   "quote": "we want Dario and Sam competing like crazy."
  },
  {
   "holder": "Chamath Palihapitiya",
   "claim": "AGI has effectively been reached — we already have artificial general intelligence, just not yet fully deployed.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "AGI",
    "superintelligence"
   ],
   "timestamp": "29:28",
   "quote": "AGI, I think, I suspect, you'll agree with me that we've hit it."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "By any definition of AGI from decades ago, the current systems have already blown past it.",
   "stance": "assertion",
   "conviction": 0.85,
   "subjects": [
    "AGI",
    "superintelligence"
   ],
   "timestamp": "29:49",
   "quote": "By any definition we had 20 years ago, we've hit it."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "Recursive, loop-based AI gains are exponential — iterating a model on its own results produces vastly better answers, with no clear ceiling in sight.",
   "stance": "assertion",
   "conviction": 0.75,
   "subjects": [
    "recursive learning",
    "superintelligence"
   ],
   "timestamp": "31:52",
   "quote": "powerful recursive gains are exponential."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "AI will cause real economic dislocation, but it is outweighed by the benefits — like cars displacing carriage-makers, the honest pros/cons ledger comes out positive.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "job displacement",
    "superintelligence"
   ],
   "timestamp": "38:21",
   "quote": "There was dislocation when cars came and it was a bad deal to be a guy who shoot horses, right, build carriages."
  },
  {
   "holder": "Andrew Feldman",
   "claim": "We have long known personalized tutoring beats factory-model teaching but chose not to do it; AI can finally deliver individualized teaching at scale.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "education",
    "superintelligence"
   ],
   "timestamp": "39:13",
   "quote": "if you give a child a tutor and the tutor modifies the teaching for the child, they learn better. That's not how we do teaching classes."
  },
  {
   "holder": "Chamath Palihapitiya",
   "claim": "When you weigh AI's costs and benefits honestly, the ledger is heavily weighted toward abundance.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "superintelligence"
   ],
   "timestamp": "39:58",
   "quote": "the ledger, as you describe it, is heavily weighted towards abundance."
  },
  {
   "holder": "Chamath Palihapitiya",
   "claim": "To generate a convincing image or video of the world, a model must by default understand the world and the objects in it.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "world models",
    "generative video"
   ],
   "timestamp": "43:27",
   "quote": "if you can make the image and you can train the model, that means by default you understand the world. In order to make a video of the world, you have to understand the world"
  },
  {
   "holder": "Robin Rombach",
   "claim": "A single multimodal model can generate images, video, and audio and also predict actions, so the same model can ultimately be deployed as a robot's brain in the real world.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "world models",
    "physical AI",
    "Black Forest Labs"
   ],
   "timestamp": "42:09",
   "quote": "you can actually use the same model to make images, to make videos, to make audio, and to predict actions which means you can ultimately deploy it on a robot in the real world."
  },
  {
   "holder": "Robin Rombach",
   "claim": "Generative AI models are a creative medium, not a prescribed tool — creators like Scorsese should decide how to use them, not be told how.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "AI film production"
   ],
   "timestamp": "47:39",
   "quote": "We don't want to tell anyone, especially not someone like Martin Scorsese, how is he supposed to use his model?"
  },
  {
   "holder": "Robin Rombach",
   "claim": "The most interesting and valuable use of these tools comes with a human in the loop iterating, not fully automated end-to-end movie generation.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "AI film production"
   ],
   "timestamp": "49:34",
   "quote": "the real interesting use cases, they come when you have like a human in the loop who iterates and uses it as a medium."
  },
  {
   "holder": "Robin Rombach",
   "claim": "The most interesting use of generative AI is creating something new rather than replicating existing IP, and blocking certain IP generation on public tools is a sensible approach.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "content licensing"
   ],
   "timestamp": "58:59",
   "quote": "the most interesting use cases of this, like if you think about content creation, is in generating something, making something that hasn't been there before."
  }
 ],
 "predictions": [
  {
   "holder": "Andrew Feldman",
   "prediction": "Cerebras's chip performance will improve well over 2x in the next 18 months, continuing to beat the traditional Moore's-law trajectory.",
   "timeframe": "the next 18 months",
   "subjects": [
    "Cerebras",
    "Moore's law"
   ],
   "timestamp": "13:18",
   "quote": "in the next 18 months will be way over 2x.",
   "due": "2028-01-10"
  },
  {
   "holder": "Andrew Feldman",
   "prediction": "A massive AI-related data leak/breach is inevitable — we know it will happen, we just don't know when.",
   "timeframe": "unspecified",
   "subjects": [
    "AI safety"
   ],
   "timestamp": "26:33",
   "quote": "I think we can also know that there will be a massive data leak. Of course. We know this."
  },
  {
   "holder": "Andrew Feldman",
   "prediction": "This technology gives a real shot at a future where our children and the people they love do not die of cancer.",
   "timeframe": "within a generation",
   "subjects": [
    "healthcare",
    "superintelligence"
   ],
   "timestamp": "38:21",
   "quote": "There's a shot that our children, none of them nor their people they love will die of cancer."
  },
  {
   "holder": "Andrew Feldman",
   "prediction": "AI will create massive abundance.",
   "timeframe": "unspecified",
   "subjects": [
    "superintelligence"
   ],
   "timestamp": "40:15",
   "quote": "I think it'll create abundance."
  },
  {
   "holder": "Robin Rombach",
   "prediction": "Generative image/video technology will keep improving rapidly and unlock even more high-end film use cases.",
   "timeframe": "unspecified (hard to predict)",
   "subjects": [
    "generative video",
    "AI film production"
   ],
   "timestamp": "53:19",
   "quote": "it's not going to stop there, it's going to continue to improve."
  },
  {
   "holder": "Chamath Palihapitiya",
   "prediction": "The future of content IP is owners licensing their characters to fans — charging a fee or a share of output — so fans can create their own stories, as AI Star Wars fan films already show.",
   "timeframe": "unspecified",
   "subjects": [
    "content licensing",
    "AI film production"
   ],
   "timestamp": "1:00:23",
   "quote": "I think that's really the future is letting the customer base pay a licensing fee or pay a fee, maybe rent software or maybe based on the output, and let them be creative with the characters, let them make their own stories."
  }
 ],
 "facts": [
  {
   "statement": "Cerebras has a $25 billion backlog.",
   "subjects": [
    "Cerebras"
   ],
   "value": "$25B backlog",
   "timestamp": "3:34",
   "quote": "we have a $25 billion backlog.",
   "source": "Andrew Feldman"
  },
  {
   "statement": "AI data centers being built in the next several years will use more power than the previous 50 years on Earth consumed.",
   "subjects": [
    "AI build-out"
   ],
   "value": "> previous 50 years of Earth's power",
   "timestamp": "2:03",
   "quote": "data centers that are in the next several years going to use more power than the previous 50 years on Earth took.",
   "source": "Andrew Feldman"
  },
  {
   "statement": "Individual AI data-center buildings the size of football fields draw more power than mid-sized cities.",
   "subjects": [
    "AI build-out"
   ],
   "timestamp": "2:27",
   "quote": "individual buildings, the size of football fields that have more power coming into them than mid-sized cities.",
   "source": "Andrew Feldman"
  },
  {
   "statement": "A Cerebras chip costs about half a billion dollars to make.",
   "subjects": [
    "Cerebras"
   ],
   "value": "~$500M per chip",
   "timestamp": "11:44",
   "quote": "when one costs half a billion to make, you bring it everywhere with you.",
   "source": "Andrew Feldman"
  },
  {
   "statement": "Cerebras runs GLM, Kimi, the Qwen models, OpenAI's closed-source models, GlaxoSmithKline's own models, and its UAE partners' (G42 / MBZ-UAI) sovereign models.",
   "subjects": [
    "Cerebras",
    "GLM",
    "Kimi",
    "G42",
    "GlaxoSmithKline"
   ],
   "timestamp": "19:35",
   "quote": "we run GLM, we run Kimi, we run the Qwenn set of models and we run open AI models, the closed source ones. We run models for, say, GlaxoSmithKline, which they wrote and developed. We run models for our partner in the UAE, G42 and MBZ-UAI that are their models that they designed.",
   "source": "Andrew Feldman"
  },
  {
   "statement": "OpenAI released an open-source model called OSS 120b some months earlier.",
   "subjects": [
    "OpenAI",
    "OSS 120b"
   ],
   "timestamp": "18:37",
   "quote": "I think OpenAI made a good call releasing OSS 120b some months back.",
   "source": "Andrew Feldman"
  },
  {
   "statement": "Palo Alto Networks (Nikesh Arora) ran a new model against its own software, found critical vulnerabilities it was unaware of, and had to stop everything to patch for six weeks.",
   "subjects": [
    "Palo Alto Networks",
    "AI safety",
    "Nikesh Arora"
   ],
   "timestamp": "25:35",
   "quote": "we had to stop everything we're doing and do patches for six weeks.",
   "source": "Nikesh Arora (via Chamath Palihapitiya)"
  },
  {
   "statement": "Black Forest Labs is based in Freiburg (in the Black Forest, Germany) and San Francisco, with two offices.",
   "subjects": [
    "Black Forest Labs",
    "Germany",
    "San Francisco"
   ],
   "timestamp": "41:30",
   "quote": "It's a town called Freiburg and in San Francisco. We have two offices.",
   "source": "Robin Rombach"
  },
  {
   "statement": "Black Forest Labs was founded about two years ago.",
   "subjects": [
    "Black Forest Labs"
   ],
   "value": "~2 years old",
   "timestamp": "41:42",
   "quote": "We started a company two years ago.",
   "source": "Robin Rombach"
  },
  {
   "statement": "Rombach and his co-founders invented latent diffusion (as PhD students in Munich), the algorithm underlying stable diffusion and modern generative image/video (and physical-AI) models.",
   "subjects": [
    "latent diffusion",
    "Black Forest Labs",
    "Stable Diffusion"
   ],
   "timestamp": "41:42",
   "quote": "we invented an algorithm called latent diffusion, which is basically like the fundamental algorithm behind all of like generative models that are being deployed for image generation, video generation, even like physical AI now.",
   "source": "Robin Rombach"
  },
  {
   "statement": "When Rombach began his PhD, image models could only produce 64x64-pixel images; today they produce multi-minute high-resolution video.",
   "subjects": [
    "generative video",
    "Black Forest Labs"
   ],
   "value": "64x64 px -> multi-minute HD video",
   "timestamp": "53:19",
   "quote": "the only thing that you could do was images of 64 by 64 pixels. How you can do multi-minute videos at a high resolution",
   "source": "Robin Rombach"
  },
  {
   "statement": "Black Forest Labs has crossed 100 people and is hiring in Germany and San Francisco.",
   "subjects": [
    "Black Forest Labs"
   ],
   "value": "100+ people",
   "timestamp": "1:02:12",
   "quote": "we just crossed 100 people, we're hiring in Germany and in San Francisco.",
   "source": "Robin Rombach"
  },
  {
   "statement": "Gal Gadot's Bitcoin movie was shot on a sound stage without green screens, with AI-generated scenery, on a ~$30M budget versus an estimated $150M with physical sets (which would have kept it from being greenlit).",
   "subjects": [
    "Gal Gadot",
    "AI film production"
   ],
   "value": "$30M budget vs. ~$150M with physical sets",
   "timestamp": "52:38",
   "quote": "That's a $30 million budget movie. She said it would have cost $150 million if they had to build sets, and the film would have never been greenlit.",
   "source": "Gal Gadot (via Chamath Palihapitiya)"
  },
  {
   "statement": "Startups historically paid $100,000-$250,000 for a launch video; some now spend just a week or two working with a director using AI tools.",
   "subjects": [
    "AI film production"
   ],
   "value": "$100k-$250k launch videos",
   "timestamp": "50:35",
   "quote": "people used to make a launch video for their startup for $100,000, $250,000.",
   "source": "Chamath Palihapitiya"
  },
  {
   "statement": "OpenAI had a Sora licensing relationship with Disney for certain characters that is no longer active.",
   "subjects": [
    "OpenAI",
    "Disney",
    "Sora"
   ],
   "timestamp": "58:02",
   "quote": "OpenAO had a relationship with them for Sora that's no longer happening",
   "source": "Chamath Palihapitiya"
  },
  {
   "statement": "The 'Star Wars Stories Untold' AI fan-film channel gets millions of views per video on YouTube.",
   "subjects": [
    "AI film production",
    "content licensing"
   ],
   "value": "millions of views per video",
   "timestamp": "1:00:23",
   "quote": "Star Wars stories untold is I think the biggest one. It's getting millions of views per video already.",
   "source": "Chamath Palihapitiya"
  },
  {
   "statement": "Black Forest Labs is working with filmmaker Martin Scorsese, who explored the models in person with Rombach multiple times.",
   "subjects": [
    "Martin Scorsese",
    "Black Forest Labs"
   ],
   "timestamp": "47:39",
   "quote": "sitting in the same room with him multiple times and actually him seeing like exploring our models.",
   "source": "Robin Rombach"
  }
 ],
 "notable_quotes": [
  {
   "speaker": "Andrew Feldman",
   "timestamp": "2:03",
   "quote": "data centers that are in the next several years going to use more power than the previous 50 years on Earth took.",
   "why": "Captures the almost incomprehensible physical scale of the AI infrastructure build-out that opens the episode."
  },
  {
   "speaker": "Chamath Palihapitiya",
   "timestamp": "16:09",
   "quote": "this open source started to figure out reasoning and the gap has suddenly closed this year.",
   "why": "The episode's title thesis ('Open Source Wins') stated plainly — open models catching frontier labs on reasoning."
  },
  {
   "speaker": "Andrew Feldman",
   "timestamp": "16:45",
   "quote": "You don't want to take your Ferrari to the grocery store.",
   "why": "The memorable metaphor for reserving frontier models for hard problems and using cheaper open-source models for ordinary work."
  },
  {
   "speaker": "Andrew Feldman",
   "timestamp": "29:49",
   "quote": "By any definition we had 20 years ago, we've hit it.",
   "why": "The flat 'AGI has arrived' claim that anchors the superintelligence discussion."
  },
  {
   "speaker": "Andrew Feldman",
   "timestamp": "38:21",
   "quote": "There's a shot that our children, none of them nor their people they love will die of cancer.",
   "why": "Crystallizes Feldman's optimist's ledger — framing AI as a massive net benefit for humanity."
  },
  {
   "speaker": "Robin Rombach",
   "timestamp": "54:17",
   "quote": "you can use the same kind of AI model to make a movie and deploy that as a brain on a robot.",
   "why": "Rombach's core thesis that generative-media models and robotics are converging into one multimodal world model."
  }
 ],
 "extracted_by": "agent",
 "prompt_version": "episode-extract@3",
 "_file": "all-in-2026-07-10-open-source-wins-agi-is-here-and-scorsese-s-ai-toolkit-with-.json"
}