N/A

Most code is now written as a collaboration between humans and code-assistive technologies, which are based on foundation models, trained on text and source code. How can we ensure that this collaboration is most effective? In this ambitious project, we will develop methods based on machine learning, cognitive science, and software engineering that improve the effectiveness of human-AI collaboration in the domain of code generation. An example of a concrete research question of interest is how to adapt the model output based on inferences about the user's cognitive state. We will also validate our developed methods via large-scale crowdsourced human experiments as well as in-person studies with a more focused population of programmers.