Collaboration aims to create powerful AI systems that learn from experience and discover new knowledge independently.
NVIDIA and Ineffable Intelligence have announced a major engineering partnership focused on building advanced infrastructure for reinforcement learning, a branch of artificial intelligence where systems improve through trial and error.
The collaboration brings together NVIDIA CEO Jensen Huang and reinforcement learning pioneer David Silver, whose research helped shape modern AI systems.
According to Huang, the next stage of artificial intelligence will involve “superlearners” — systems that continuously improve by learning from real-world experience instead of relying only on human-created data.
Silver explained that AI researchers have already made major progress in creating systems that can understand and use existing human knowledge. However, he said the bigger challenge now is building AI that can independently discover new ideas and solutions through experience-based learning.
Unlike traditional AI pretraining, reinforcement learning systems generate new data while they operate. These systems constantly act, observe results, score outcomes, and update themselves in real time. This creates heavy demands on computing infrastructure, including memory bandwidth, interconnect speed, and data-serving capabilities.
To solve these challenges, engineers from NVIDIA and Ineffable Intelligence are working together to design highly optimized reinforcement learning pipelines capable of operating at massive scale.
The project will begin using NVIDIA’s NVIDIA Grace Blackwell architecture and will later expand to the upcoming NVIDIA Vera Rubin platform. The companies aim to explore the future hardware and software needed for AI systems that learn through simulation and experience rather than human-generated datasets.
Both companies believe the new infrastructure could enable reinforcement learning agents to operate in highly complex environments and potentially drive breakthroughs across science, technology, and other fields of knowledge.
Source: NVIDIA
