CoreWeave's Fully Connected 2026 Day 1 keynote centers on the AI loop of run, observe, curate, improve and evaluate, and the launch of CoreWeave Forge. It covers production agent operations, agent observability and RL rollouts, Vera Rubin NVL72 and the Vera CPU, and enterprise evaluation practices from Capital One, Caterpillar and Cognition.
Denis Blessing presents scalable training of diffusion samplers via fixed point score matching: a unifying framework for learning Boltzmann-type distributions from energy evaluations alone, with applications to amorphous materials, alloy phase diagrams, and reinforcement learning with flow policies.
Dyna Robotics is pioneering physical AI with Dyna-2, a world-action model trained on over a million hours of human video, predicting future states and generating robot actions. Co-founder Jason Ma explains how this enables robots to learn new tasks from minimal demonstrations, tackles the robotics data bottleneck with synthetic data and self-improving models, and details their full-stack approach, including custom semi-humanoid hardware for real-world deployments at Din Tai Fung. The discussion delves into diffusion transformers, agentic harnesses for hour-long workflows, the "Bitter Lesson" of scaling data, and the distinction between world-action models and VLAs, all driven by a deployment-first research culture.
David Silver, co-founder of Ineffable Intelligence, outlines his bold vision for building a "superlearner"—an AI that discovers science, mathematics, and language autonomously, free from human data or explicit guidance. Drawing parallels to AlphaGo Zero's success, Silver argues that human-language reasoning is a ceiling, not a foundation, for advanced AI. He details their approach using complex simulated "crucibles" and simple rewards, aiming to develop the "science of superintelligence" in simulation for safety and accelerated progress. With a powerful GPU cluster and a small, elite team, Silver forecasts strong evidence within a year and superintelligence within three.
Snorkel AI and UC Berkeley's Sky Computing Lab show that a 4B Qwen3 model, trained with reinforcement learning for under $500 using the rLLM framework, beats its 235B sibling on expert-validated financial question answering built from SEC 10-K filings. The talk covers the FinQA dataset construction and verification, common failure modes of frontier models, transfer to multi-table and general tool-use tasks, and ablations showing that simple data and binary rewards outperform more elaborate setups, since the real bottleneck is disciplined tool use rather than reasoning depth.
Misha Laskin, co-founder and CEO of Reflection AI, discusses building frontier open-weight models, the launch of the 500B-parameter Beam reasoning model, the economics of renting versus owning intelligence, and why he believes open models will capture most token demand. The conversation covers RL-driven scaling efficiency, Chinese open model ecosystems and their geopolitical implications, and the safety case for openness, including cybersecurity as a lens for defense and offense. He closes with optimism about AI-accelerated science and the evolving research workforce.