This paper was accepted at IEEE Symposium on Visible Languages and Human-Centric Computing (VL/HCC) 2024
Programmers steadily have interaction with machine studying tutorials in computational notebooks and have been adopting code era applied sciences primarily based on massive language fashions (LLMs). Nonetheless, they encounter difficulties in understanding and dealing with code produced by LLMs. To mitigate these challenges, we introduce a novel workflow into computational notebooks that augments LLM-based code era with a further ephemeral UI step, providing customers UI scaffolds as an intermediate stage between person prompts and code era. We current this workflow in BISCUIT, an extension for JupyterLab that gives customers with ephemeral UIs generated by LLMs primarily based on the context of their code and intentions, scaffolding customers to grasp, information, and discover with LLM-generated code. By a person research the place 10 novices used BISCUIT for machine studying tutorials, we discovered that BISCUIT affords customers representations of code to assist their understanding, reduces the complexity of immediate engineering, and creates a playground for customers to discover totally different variables and iterate on their concepts.
