Jeb: Turn any OpenAI API into a decision model (github.com)

Jeb, a new tool designed for the AI/ML community, enables users to convert any OpenAI API into a decision-making model. By sending a state along with various questions—including choices, scores, or binary (noul) queries—Jeb prompts an OpenAI-compatible model, retrieves token log probabilities, and delivers structured JSON responses. This innovation allows developers to create decision frameworks that leverage advanced language models' probabilistic outputs, enhancing the precision of automated decision-making processes. The significance of Jeb lies in its ability to facilitate more complex decision-making scenarios using existing OpenAI APIs, making it accessible to a wider array of applications from risk assessment to resource management. It supports multiple question types, including choices with descriptions, risk scores, and binary decisions, and utilizes log probabilities to derive probabilities for each outcome. With detailed configuration options and support for various OpenAI-compatible providers, Jeb empowers developers to better harness language models for critical decision-making tasks, potentially leading to improved outcomes in fields like finance, healthcare, and logistics.