Prompting Is Programming: A Query Language for Large Language Models

Explore a groundbreaking approach to language model programming in this 30-minute conference talk from PLDI 2023. Discover how researchers from ETH Zurich introduce Language Model Programming (LMP), a novel concept that combines text prompting with scripting to enhance the capabilities of large language models. Learn about LMQL (Language Model Query Language), an implementation of LMP that optimizes inference procedures and minimizes costly language model calls. Gain insights into how this innovative method can capture state-of-the-art prompting techniques, facilitate interactive flows, and potentially reduce computation costs by 26-85%. Delve into the practical applications and benefits of this new paradigm in language model interaction, supported by evaluations demonstrating improved accuracy on various downstream tasks.