Kyle Morris, founder of Democratize Intelligence, gives opening remarks of the Summit celebrating the Democratize Intelligence as a community dedicated to making AI open and accessible.
Simeon Bochev, CEO of Compute Exchange, argues relying on a few companies for compute and data stifles AI innovation—unlike the early internet’s open hardware boom. He highlights only 20 million GPUs exist instead of the billions needed, while lack of shared standards wastes capacity and drives up costs.
Jacob Buckman, CEO of Manifest AI, calls expanded context length the next major breakthrough in AI model architecture because transformers’ costly quadratic scaling limits how much data they retain. He proposes “power attention,” a sub-quadratic approach that he expects most will adopt.
Fireside chat with Amjad Masad recounting the early days of how Replit got started, and why he made a big bet on AI agents as Replit’s core product.
A panel on democratizing compute featuring Chris Lattner, John Barrus, Simeon Bochev, and Carmen Li.
Richard Ngo, an AI governance researcher, talked about how AI will give a lot of power to a few people, which could be dangerous. We can either align AI or rely on law enforcement to stop bad actors, but both have downsides. He suggested ways to keep AI safe without too much control—like better transparency, safety checks, and making sure no one group has too much power.
Josh Albrecht, CTO of Imbue, advocates “human in the loop” for AI agents writing software over pure automation. His vision is to revive the idea of a personal computer — empowering domain experts who don’t write software (like scientists) to build custom AI software for their niche use cases.
Jan Liphardt, founder of OpenMind, talked about an open-source robotics system, OM1, that leverages LLMs to convert multisensory inputs—such as vision, sound, and location—into natural language, which are then translated into actionable commands via a hardware abstraction layer.
Div Garg, co-founder of The AGI Company, describes his journey from a hackathon winning project to developing commercial-grade AI agents, highlighting three open-source projects: an agent protocol, a real-world benchmark framework, and self-improving training methods like Agent Q.
Chris Lattner, CEO of Modular AI, presents Max, an AI inference platform that works without CUDA, he shows it can do a from scratch compilation of llama 3.1 in 20 seconds on an Apple M3 laptop compared to other standard methods costing as much as an hour for the same task
Srivastava Satyam, Chief AI SW architect at d-Matrix, presented Corsair— an accelerator that delivers 2–2.5× higher throughput and 10–30× lower latency than top GPUs to lower training costs.
Carmen Li, CEO of SiliconData, brings transparency to the global GPU markets through tools like SiliconNavigator, SiliconMark, SiliconCarbon to track pricing, performance, and carbon emission metrics.
Fireside chat with Bill Tai, TSMC’s first badge and early Zoom/Canva investor.