Fluid Concepts is an AI data lab founded by former Meta Superintelligence Labs (FAIR) researchers.

We build datasets to solve some of the hardest and most meaningful capability challenges for the future of human<>AI interaction:

• multimodal understanding and generation

• expressivity and personality

• realtime conversational coordination

• asynchronous agency and tool use

• social reasoning and inference

• contextual grounding and situated behavior

We will look back on today’s primarily text-bound, turn-based interactions the same way we do early command lines: functional, transformative for their time, but ultimately limited.

Humans bring a deeply refined suite of communicative tools to any interpersonal interaction: nuanced expressivity, gesture and reactivity, multimodal perception and production, contextual grounding, social inference, and the ability to jointly coordinate activity.

Our foundational datasets are designed to help models gain traction on these capabilities.

Previously, our team contributed to some of the world’s most advanced efforts in conversational AI, multimodal interaction, and embodied intelligence — leading large-scale data initiatives for products used by billions as well as research efforts featured in leading scientific venues.