NVIDIA has announced a strategic engineering collaboration with Ineffable Intelligence, a London-based AI laboratory founded by David Silver, the architect of AlphaGo. The partnership aims to develop the necessary infrastructure for large-scale reinforcement learning, a field where AI systems acquire knowledge through continuous trial and error rather than static datasets. Ineffable Intelligence recently emerged from stealth mode to pursue this ambitious research direction. Jensen Huang, NVIDIA CEO, described the collaboration as a move toward creating superlearners. He stated that the next frontier of artificial intelligence involves systems capable of learning continuously from experience. By working together, the two companies intend to codesign the hardware and software foundation required to push these intelligent systems forward. David Silver emphasizes a shift in the AI landscape. He argues that researchers have largely mastered the initial challenge of building systems that replicate human knowledge. The more difficult problem now is creating systems that can discover entirely new knowledge independently. This evolution requires a fundamental change in approach, moving from passive data consumption to active learning from experience. Implementing this vision presents unique technical challenges. Unlike pretraining, which relies on fixed historical data, reinforcement learning generates data dynamically in real time. These workloads require systems to constantly act, observe, score outcomes, and update models in tight loops. This process places immense pressure on network interconnects, memory bandwidth, and model serving capabilities in ways that traditional training does not. Furthermore, the data generated in these simulated environments often differs significantly from human language or imagery, potentially necessitating novel model architectures and training algorithms. To address these challenges, engineers from NVIDIA and Ineffable Intelligence are collaborating to build a high-performance training pipeline. This initiative will begin on NVIDIA's current Grace Blackwell platform and serve as a foundational test for the upcoming Vera Rubin system. The objective is to define the next generation of technology needed as the industry transitions from relying on human data to models that learn through simulation. Successfully establishing this infrastructure is expected to unlock an unprecedented scale of reinforcement learning. By enabling agents to operate effectively in highly complex and rich environments, the partnership hopes to facilitate breakthroughs across various fields of knowledge. This effort marks a critical step in developing AI that can evolve and innovate independently, moving the technology beyond mere data replication into the realm of autonomous discovery.