Explorations

We are actively exploring a range of projects that we believe hold significant potential not only for Microsoft but also for the broader society.

Large Action & Behavioral Models

Games, Robotics, Self-driving OS

Large Action & Behavioral Models (LAM) are AI foundational models that can predict how to act on human environments and interfaces, like LLM are capable to interpret human language and generate coherent text. We develop LAMs that will be capable to perform complex tasks in a multitude of scenarios, like how to play a videogame, how a to move in the real world, and how to use an Operating System (OS).

Transformer Enhancements

EMA (Exponential moving average), LASER, Learning from Text feedback, feedback driven learning in a RAG

We develop techniques that can dramatically improve the performance or reduce the costs of Transformer-based Large Language Models (LLMs) by applying simple interventions after training requiring minimal or zero parameters or data.

Foundational Models for latent state and latent dynamics in LLM

A latent state or a latent dynamic can improve planning, exploration, and credit assignment of AI agents by keeping task-relevant information and discarding distractions. We develop techniques that autonomously discover such latent representations directly from real-world signals.