DSPy – Programming – not prompting – LMs (dspy.ai)

DSPy, a new declarative framework for AI software development, has been announced to streamline the process of building modular systems for language models (LMs). This innovative approach shifts the focus from creating fragile prompts to programming in a structured manner, using natural-language modules that can be easily integrated with various models and inference strategies. DSPy enables developers to iterate quickly, making AI software more reliable, maintainable, and portable, akin to the transition from assembly language to higher-level programming languages like C. Significantly, DSPy provides tools for optimizing prompts and weights in LM applications, allowing for more efficient and effective model performance. Its design fosters an open-source community where users can contribute to and refine both architectural components and optimization methods. With support for a range of language model providers and the ability to run sophisticated optimizers, DSPy aims to enhance the compositionality of AI software, empowering developers to leverage the latest research advancements while effectively managing the complexity of AI programming.