Data-centric disambiguation for data transformation with programming-by-example
summary
A system called FlashAttention highlights ambiguous outputs from a programming-by-example engine (FlashFill/FlashProg family), shows orange ambiguity markers, offers checkbox confirmations and an enhanced scrollbar, and lets users correct outputs instead of editing programs. A lab study with 26 participants performing 12 Kaggle-derived tasks shows FlashAttention cuts median completion time across three dataset sizes and reduces task-error rate to about 4.7% versus 15.3% for FlashFill. Limitations include dependence on the DSL expressiveness and a learning curve.
AI-written summary · about
chapters (4)
related talks
25mFlashFill++: Scaling Programming by Example by Cutting to the Chase
27mDefault Disambiguation in Online Parsers
01h 04mIn Defense of "Little Code"
30mCognitive Shortcuts: Models, Visualizations, Metaphors, and Other Lies
52mResistance Is Futile
20mChoose, Don’t Label: Multiple-Choice Query Synthesis for Program Disambiguation
42mBuild Without Limits - Lessons Learned Working with Data and AI Coding Assistants
22mCodeMend: Assisting Interactive Programming with Bimodal Embedding