Completing Programming Adventures with Jev (tomasp.net)

A recent announcement from TypeSafe AI regarding their "System One Model" named Jev aligns perfectly with a paper being presented at the SPLASH conference, focused on a unique programming interaction model known as the Choose-Your-Own-Adventure Calculus. This calculus allows users to create programs by selecting from a series of choices presented by the programming system, akin to F# type providers. Significantly, the paper explores the integration of large language models (LLMs) to automate the choice-making process, thereby enhancing user experience and learning. Jev is designed to facilitate this interaction, enabling users to not only receive code suggestions but also understand the underlying options thanks to its capacity to assign probabilities to multiple choices. Jev's introduction is noteworthy for the AI/ML community as it stands to deepen the intersection of programming and AI-assisted choices, particularly for applications involving data exploration and manipulation. The experimental results suggested that Jev outperformed existing LLMs in terms of accuracy for these types of tasks while being significantly more cost-effective—up to 50 times cheaper than high-end models when using system prompts. With its ability to sort and present options based on probability, Jev not only aids users in selecting the right code snippets but also encourages a more interactive and oversight-friendly programming experience, potentially reshaping how developers engage with AI in coding environments.