Research
We introduce Large Behavioral Models (LBMs)—foundation models trained on chronological sequences of human actions. By framing behavioral data as sequences of discrete events, LBMs learn powerful representations complementary and often orthogonal to LLMs. BehaviorGPT, a frontier LBM, delivers double-digit sales uplift for retailers and payment companies.
We introduce the first behavioral foundation model for visual art and aesthetics. Trained on 215 billion human interactions (4.7 trillion tokens) across major art and design platforms, BehaviorGPT-v3 is a 0.5B-parameter Transformer model that learns semantic meaning not from pixels or text-image pairs but from sequences of user behavior.
We applied language modeling techniques to detailed workforce behavioral data, creating BehaviorGPT-v2: a foundation model treating employee behaviors as a language to predict future actions and outcomes.
We applied language modeling to rich grocery consumption data to build BehaviorGPT-v1, treating user consumption history as a language to predict future events.