Code models improve interfaces by learning from user feedback
Revision-Based Learning from Interactive Experience and Feedback
Artificial Intelligence
Summary
Interfaces like websites need to work well during real use, but it’s hard to teach computer models how to make better ones just by looking at code. The authors introduce a method called RELIEF that lets a coding model fix its own websites based on feedback from users interacting with them. By collecting data on how people use these sites and what they think, the method helps the models learn from their mistakes and make better, more usable interfaces. Their tests showed that feedback from real human interactions helped improve the models the most.
What this means in practice
- •For web developers: Improve generated websites by incorporating user interaction feedback to make sites easier and more intuitive.
- •For software engineers: Enhance code generation tools for user interface design by training models with real user critiques and experience data.
Tested on one dataset.
Authors
Jiawen Li, Chenming Ge, Alice Gao, Rishab Vempati, Yi-Hao Peng, Jason Wu
Abstract
Code generation models are increasingly used to produce interactive artifacts, such as websites and user interfaces. Many properties that determine whether these interfaces are useful only become apparent through realistic interaction, making them difficult to incorporate into model training. In this paper, we introduce RELIEF (REvision-based Learning from Interactive Experience and Feedback), a method for improving code generation models using feedback on their generated interfaces. RELIEF uses a coding agent to revise these interfaces, then incorporates the improvements into their original generation traces to produce supervision for fine-tuning. To evaluate RELIEF, we collected a dataset of 200 synthetically generated websites paired with 355 interaction recordings and accompanying feedback from 24 users. We used critiques derived from code inspection, automated interaction, explicit human feedback, and human interaction recordings to fine-tune variants of two code generation models. Our results show that feedback effectiveness varies with base model capabilities, with critiques inferred from human interaction recordings leading to the largest usability improvements overall under our tested models.