Ep 11: Is AI Model Distillation the new U.S. vs China frontier? Ft. Travis Good x Ben Fielding x Bailey Wang

On February 12th, OpenAI sent a memo to the US House Select Committee on China claiming DeepSeek systematically distilled its models through third party routers and unauthorised resellers. Two weeks later, Anthropic followed with allegations that DeepSeek, Moonshot, and Minimax created over 24,000 fraudulent accounts to conduct 16 million exchanges with Claude in what they call industrial scale distillation campaigns. In this episode, we strip away the geopolitics and ask whether distillation is a fundamental feature of intelligence or IP theft. Bailey argues companies are stealing post-training data worth $400 an hour by recovering OpenAI's expert annotations through model prompts, not just compute costs. Travis has a conspiracy theory based on the Platonic model hypothesis that models trained on 16 trillion tokens with five trillion parameters start looking very similar, and you can infer exact differences by running variations of a single prompt. Ben compares this to digital rights management and music distribution, where the world shifted what was valuable instead of fighting to maintain the old system. The moat is not in training anymore but in productization and user experience. When agents become autonomous and selection pressure strips away their humanity to create the best financial report parser, where does accountability lie if something goes wrong? Is software with neural nets bolted on for resilience fundamentally different from the agent discourse about human-like generalizability?•••About ⁠Gensyn:⁠Gensyn is building decentralised compute infrastructure for AI training and inference at scale. The protocol enables global GPU owners to contribute computational power to machine learning workloads through cryptographic verification, making AI development more accessible and cost-effective while reducing dependence on centralized cloud providers.Find Gensyn on ⁠X⁠.•••About ⁠Ben Fielding:⁠Ben Fielding is the CEO and Co-founder of Gensyn. He focuses on building verifiable compute protocols that democratize access to AI infrastructure, addressing supply-demand dynamics in machine learning markets, and creating economic incentives that align decentralized compute networks with real-world AI development needs.Find Ben on ⁠X.⁠•••About ⁠PrismaX:⁠PrismaX develops AI-powered robotics solutions for physical labor automation. The company addresses disruption across all labor categories, not just white collar work, by creating tools that help people transition from repetitive assembly and cleaning tasks to higher-value work while maintaining earning opportunities through expert annotations and supervision of robotic systems.•••About⁠ Bayley Wang:⁠Bayley Wang is the CEO of PrismaX. His work focuses on how AI and robotics can serve as equalizers in labor markets, enabling workers to create more value in less time while addressing the ethical challenges of physical automation and ensuring displaced workers have pathways to new earning opportunities in the AI economy.Find Bayley on ⁠X.⁠•••Hosted by Travis Good, CEO & Co-Founder of ⁠⁠⁠⁠⁠⁠⁠⁠⁠Ambient⁠⁠⁠⁠⁠⁠⁠⁠⁠: an SVM-compatible PoW L1 that will serve as a cornerstone of the agentic economy. Ambient has raised $7.2M from a16z, CSX, Delphi Digital and Amber Group. It is 10x more efficient than incumbent crypto AI systems. Ambient believes AI is money and is working to democratize it. Find us on: ⁠⁠⁠⁠⁠⁠⁠⁠⁠X⁠⁠⁠⁠⁠⁠⁠⁠⁠ | ⁠⁠⁠⁠⁠⁠⁠⁠⁠Youtube⁠⁠⁠⁠⁠⁠⁠⁠⁠ | ⁠⁠⁠⁠⁠⁠⁠⁠⁠LinkedIn⁠⁠⁠⁠⁠⁠⁠⁠⁠ | ⁠⁠⁠⁠⁠⁠⁠⁠⁠Ambient Chat⁠⁠⁠⁠⁠⁠⁠⁠⁠Find Travis on: ⁠⁠⁠⁠⁠⁠⁠⁠⁠X⁠⁠⁠⁠⁠⁠⁠⁠⁠ | ⁠⁠⁠⁠⁠⁠⁠⁠⁠Linkedin⁠⁠⁠⁠⁠⁠⁠⁠⁠Find our Litepaper ⁠⁠⁠⁠⁠⁠⁠⁠⁠here.