Automated Program Generation from Input-Output Examples
summary
A demo of example-driven program synthesis shows a production-grade engine used in Excel as Flash Fill (since 2013) with about one million weekly invocations, and an enhanced Flash Fill++ that handles numbers, dates, table extraction and zero-example splits. The system compiles examples into programs in a domain-specific language, searches with inverse semantics to hit ~0.1s responses, ranks multiple candidate programs to surface ambiguities, and emits readable Python extraction/parsing code (codex.read) usable in pandas and PowerShell, demonstrated on 9-1-1 and semi-structured logs.
AI-written summary · about
chapters (15)
- 0:00Welcome and introduction
- 1:03Background of PROSE team
- 3:15Problem with input-output examples
- 4:23Demo of Flash Fill in Excel
- 7:39Limitations and customer feedback
- 9:10Flash Fill Plus Plus features
- 10:32Extracting tabular data from logs
- 14:26How program synthesizers work
- 16:53Program ranking and disambiguation
- 21:31Innovations in program synthesis
- 24:52Zero-example program synthesis
- 27:27Generating readable code
- 28:31Demo with Python notebooks
- 36:24Future of programming
- 39:18Q&A session
related talks
41m"Do What I Mean": The History of AI and Program Synthesis
54mA World Without Programmers?
28mScooPy: Enhancing Program Synthesis with Nested Example Specifications
15mSemantic programming by example with pre-trained models
31mLetting users do the undecidable: lessons from program synthesis
01h 01mC++hatGPT: How AI Tools Impact Your Life as a Systems Programmer
41mThe future of AI and generative code Q&A
14mExample-Based Live Programming for Everyone: Building Language-Agnostic Tools for Live Programming w