Research
We pursue fundamental research to understand and build intelligent systems. Our work spans multimodal learning, reasoning, safety, and scalable architectures.
Research Areas
The core areas that define our research agenda.
Understanding across modalities
Multimodal Learning
We develop systems that can perceive, understand, and reason across text, images, audio, and video. Our goal is to build unified representations that capture the rich structure of the multimodal world.
Thinking beyond pattern matching
Reasoning & Planning
We investigate how to build systems capable of complex, multi-step reasoning and long-horizon planning. This includes both neural and neuro-symbolic approaches to logical and mathematical reasoning.
Building trustworthy AI
AI Safety & Alignment
Ensuring AI systems remain aligned with human values and intentions is central to our mission. We research interpretability, robustness, and alignment techniques to build AI we can trust.
Efficient frontier models
Scalable Architectures
We design novel neural network architectures that scale efficiently to frontier capabilities. Our work spans attention mechanisms, state-space models, and efficient training methods.
AI that acts in the world
Autonomous Agents
We build AI systems that can interact with environments, use tools, and accomplish complex tasks autonomously. This research bridges language models with real-world action.
Accelerating discovery
AI for Science
We apply AI to accelerate scientific discovery across domains—from biology to materials science. Our goal is to build AI systems that can contribute meaningfully to human knowledge.
Open Source
We believe in open research. We release our code, models, and datasets to accelerate progress across the AI community.
Explore on GitHub