Lightning Talk – Lee McAlilly
summary
Claims a multi-agent GPT synthetic survey can emulate human choices and validate auction and behavioral economics results researchers at MIT observe. Describes an Expected Parrot DSL in Python, using the OpenAI Assistance API to spin up many agents, funneling data from Facebook or Google Analytics. Reports building a synthetic focus group for pre-flighting ad headlines that yields a 20–30% lift faster and far cheaper than live A B tests, argues for open source tools and Ruby-friendly pipelines.
AI-written summary · about
