The first programming language designed for LLM (github.com)

MIT researchers have announced SPELL, the first programming language specifically designed for large language models (LLMs). SPELL utilizes an AI-native dataflow paradigm, allowing computations to be represented as graphs of explicit dependencies rather than following traditional sequential reasoning. Key features include explicit type definitions and a structured JSON format that aligns well with LLM training data, making it inherently suitable for AI applications. This pre-alpha version serves as a proof-of-concept by validating the architecture needed for future enhancements, such as file I/O and network operations. The significance of SPELL lies in its potential to transform how LLMs interact with programming constructs. Traditional programming languages often require a more human-centric approach, but SPELL is tailored for AI, emphasizing dependency clarity and type explicitness to facilitate data transformations. The language currently supports basic operations like addition and filtering, which are foundational for more complex processing tasks. By providing a framework that mirrors LLM operation logic, SPELL promises to streamline AI programming and push the boundaries of what is possible in machine learning development.