7:15 - 8:30 p.m.: “The Emerging Risk of Code-Writing ML Models” by Nick Doiron, a software engineer at HP Enterprise
Advanced machine learning models for language have inspired a new wave of code-generating and code-explaining models from GitHub, OpenAI, HuggingFace, and other research teams. Co-pilot, CodeParrot, and other code-generation models are emerging as an upgrade to code auto-complete tools. Large neural models succeed on benchmarks and produce lifelike outputs in human and programming languages. The downsides of language models have been discussed in papers including “Stochastic Parrots” and their regulation / alignment discussed in “CTRL: A Conditional Transformer Language Model”. There isn’t equivalent research into code generation yet. Most criticism of these models focuses on a real (but separate) issue of training on public code. In this talk, Nick will cover these issues of IP and privacy, adversarial examples which trick code-explaining models, and lessons learned from the natural language world about testing and controlling the model outputs.
8:30 - 8:45 p.m.: Door Prizes
8:45 - 9 p.m.: Networking and General Discussion
About IndyPy:
Organized by Six Feet Up, the Indianapolis Python Meetup (“IndyPy”) is a leading national group promoting the Python programming language and connecting the Python community. IndyPy draws Python enthusiasts from the entire Midwest (and beyond via Zoom) for regular events. In 2022, we celebrate 15 years of success with over 2,400 members.
