Letterpaths: Or how LLMs can be good even when they're bad (www.robinlinacre.com)

Letterpaths, a newly developed library, utilizes LLMs to create cursive handwriting tools aimed at assisting schoolchildren in the UK. While initial attempts with frontier AI models struggled with generating accurate letter shapes and natural joins, the project ultimately showcased how AI can significantly enhance human productivity. The developer employed a combination of techniques, including defining a JSON schema for letter geometry and leveraging the File System Access API for easier file management. Despite the limitations of AI in visual tasks, LLMs contributed to accelerating the library's development within a short timeframe. The significance of Letterpaths lies in its demonstration of how AI can effectively serve as a supportive tool, even when not directly suited for specific tasks. Key aspects of the library include a join algorithm that closely mimics traditional cursive writing, with careful adjustments made through an interactive visualization that allowed for real-time tweaks. Furthermore, the development structure, using a pnpm monorepo for efficient tracking and updates, highlights a repeatable strategy in software design for optimizing educational tools and other applications. This project exemplifies the potential of combining LLM insights with human creativity and problem-solving to navigate complex challenges in AI development.