# Foundation Models for Decision Making: Problems, Methods, and Opportunities

**Authors:** Sherry Yang, Ofir Nachum, Yilun Du, Jason Wei, Pieter Abbeel, Dale Schuurmans  
**arXiv:** [2303.04129](https://arxiv.org/abs/2303.04129)  
**Date:** 2023-03-07  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.04129v1/

## Key points

- The field of foundation models for decision making is rapidly evolving
- New paradigms are emerging to train these models to interact effectively with other agents and perform long-term reasoning
- These models are pretrained on diverse data at scale and have demonstrated extraordinary capabilities in a wide range of vision and language tasks
- When deployed in real-world environments, they must interface with other entities and agents
- Researchers are leveraging ever-larger datasets curated for multimodal, multitask, and generalist interaction to address these challenges
- This research holds tremendous promise for creating powerful new systems that can interact effectively across a diverse range of applications such as dialogue systems, autonomous driving, healthcare, education, and robotics
- Sherry Yang et al. examine the scope of foundation models for decision making in their manuscript
- They provide conceptual tools and technical background for understanding the problem space and exploring new research directions.

## Summary

The field of foundation models for decision making is rapidly evolving, with new paradigms emerging to train these models to interact effectively with other agents and perform long-term reasoning. These models, which are pretrained on diverse data at scale, have demonstrated extraordinary capabilities in a wide range of vision and language tasks. However, when deployed in real-world environments, they must interface with other entities and agents. For example, language models are often used to interact with human beings through dialogue, while visual perception models are used for autonomous navigation. To address these challenges, researchers are leveraging ever-larger datasets curated for multimodal, multitask, and generalist interaction. This research holds tremendous promise for creating powerful new systems that can interact effectively across a diverse range of applications such as dialogue systems, autonomous driving, healthcare, education, and robotics. In this manuscript by Sherry Yang et al., the authors examine the scope of foundation models for decision making and provide conceptual tools and technical background for understanding the problem space and exploring new research directions.