Large language models (LLMs) are highly effective for many natural language processing tasks when given the right prompts. However, finding an optimal prompt is challenging due to LLMs' notable sensitivity to the prompt input. Additionally, the frequent updates of proprietary models and the emergence of unforeseen edge cases after deployment often requires continuous manual prompt refinement.

In this talk, we explore recent methods in prompt optimization and their challenges. We introduce a novel method for automatic prompt engineering that interactively refines prompts based on user intent through synthetic data generation, and mitigates the scarcity of high-quality benchmarks for real-world applications. We will also present our new open-source system for prompt optimization, featuring powerful capabilities such as prompt distillation, prompt squashing, and synthetic benchmark creation.

AutoPrompt: https://github.com/Eladlev/AutoPrompt

You can find an overview of the shared documents, chat and QnA here: https://drive.google.com/drive/folder...

Signup for coming meetups here: https://lu.ma/d720wy9f