User Interaction Models for Disambiguation in Programming by Example
summary
Presents two concrete interaction models: program viewer paraphrases code into plain English so users inspect and switch among ranked alternative programs, and active-learning disambiguation asks targeted example questions that show the first disagreements across many candidate programs. Built a system integrated with a PBE engine (Flash Fill/Flash Extract SDK) and ran a lab with 29 participants; both models cut errors substantially with no time penalty, and users slightly prefer the proactive active‑learning model.
AI-written summary · about
chapters (10)
- 0:00Introduction to problem
- 1:50Programming by example in Excel
- 3:18Demo overview and first issue
- 4:03Program viewer interaction model
- 5:12Disambiguation interaction model
- 6:58How disambiguation works under the hood
- 8:01Simplifying program paraphrasing
- 9:46User study results
- 11:52SDK release and availability
- 12:22Q&A session
related talks
18mGrounded Copilot: How Programmers Interact with Code-Generating Models
13mInteractive Learning and User‑Centered AI: A Retrospective
23mAn Interactive Learning Environment for Program Design
26mGrappling with the Bitter Lesson
37mInteractive Explainable AI
01h 02mAI Effects on Research and Education: A Programming and Software Language Perspective
45mWhere AI Meets Code
22mCodeMend: Assisting Interactive Programming with Bimodal Embedding