Probably – a programming language for LLM workflows, powered by Jev (probably-lang.southpolesteve.workers.dev)

A new programming language named "Probably," powered by Jev, has been introduced to streamline large language model (LLM) workflows. This language allows users to create interactive applications by handling model calls and responses in a structured yet playful way. Noteworthy features include a built-in capability to assess probabilities for different judgments and responses, supporting conditional logic and probabilistic outcomes. Users can direct LLMs to generate text, draft replies, and even manipulate the tone of the generated content, all while adhering to a unique syntax that leverages TypeScript for interpretation. The significance of Probably lies in its innovative approach to programming with LLMs, emphasizing flexibility and ease of use. It encourages experimentation by allowing users to replay previous runs without calling models again, preserving the context and decisions made. While the language is still experimental and does not support complex data structures or arithmetic, it offers substantial potential for simplifying LLM-driven workflows. Developers can run it locally or deploy it using Cloudflare Workers, ensuring that the language can be integrated into existing tech stacks for enhanced productivity in AI applications.