Keynote Speech "Assured LLM-Based Software Engineering"
Abstract
This talk addresses the question: "How can we use Large Language Models for reliable Software Engineering when the models are inherently non-deterministic and worse, may hallucinate, thereby leading to more, not fewer bugs?
We outline the approach known as "Assured Large Language Model Software Engineering" (Assured LLMSE), which addresses the twin challenges of:
1. Ensuring LLM-generated code does not regress the properties of the original code.
2. Quantifying the improvements achieved in a verifiable and measurable way.
In so doing, the Assured LLMSE approach tackles the problem of LLMs' tendency to hallucinate, as well as providing confidence that generated code improves on the existing code base. Using Assured LLMSE, not only avoids problems of non-determinism and hallucination, but can even generate code that is superior-by-design to human-authored code, in well-defined ways. The keynote will outline the overall approach to Assured LLMSE, and will present results from its application at Meta platforms in the development of large systems of tens to hundreds of millions of lines of code.
This is joint work with Nadia Alshahwan, Andrea Aquino, Jubin Chheda, Anastasia Finegenova, Inna Harper, Mitya Lyubarskiy, Neil Maiden, Alexander Mols, Shubho Sengupta, Alexandru Marginean, and Eddy Wang.
Short Bio
Mark Harman is a full-time Research Scientist at Meta Platforms in the Instagram Product Performance team, working on software engineering automation. He was previously in the Simulation-Based Testing (SBT) team at Meta, which he co-founded. The SBT team developed and deployed both the Sapienz and WW platforms for client- and server- side testing. Sapienz grew out of Majicke (a start up Mark co-founded) that was acquired by Facebook (now Meta Platforms) in 2017. Prior to working at Meta Platforms, Mark was head of Software Engineering at UCL and director of its CREST centre, where he remains a part time professor. In his more purely scientific work, he co-founded the field Search Based Software Engineering (SBSE) in 2001. He received the IEEE Harlan Mills Award and the ACM Outstanding Research Award in 2019 for his work on Software Engineering automation, and was awarded a fellowship of the Royal Academy of Engineering in 2020.