Audrey Roy Greenfeld - Rethinking AI: Building What Matters, Not Just What's Possible

Filmed at dotAI on October 17, 2024 in Paris. More about the conference on https://www.dotai.io In this talk, Audrey M. Roy Greenfeld dives into the creative and entertaining possibilities of AI through an unexpected lens: generating parody news articles! Together with PyDanny, she developed a parody-generation system using LLMs, leveraging advanced techniques like function/tool calling and dual AI agents (a writer and an editor). Highlights of the talk include: A behind-the-scenes look at building and refining a parody-generation pipeline using LLMs. How adopting diverse personas enhances creativity and adds distinct voices to the parodies. The critical role of quality filtering and selecting the best examples to improve generation outputs. Using FastHTML to build an engaging platform for showcasing AI-generated parodies. While the focus is on humor, the talk also poses a deeper question: Beyond their comedic value, do AI-generated parodies have broader implications? The same tools used for fun can be repurposed for more serious challenges, like improving adult literacy or creating disinformation—underscoring the importance of ethical considerations in AI development. Viewers will leave inspired to think more creatively about AI and encouraged to explore how they can build their own meaningful and enjoyable AI-driven projects. Who is Audrey Roy Greenfeld? Audrey is a Research and Development staff member at Answer.AI. She is co-author of the Two Scoops of Django book series, and the Cookiecutter project templating system. Previously, she was the cofounder and first president of PyLadies, and has worked in engineering and operations leadership roles in nonprofit and insuretech. Her perspective is further informed by early exposure to computer vision and robotics projects at MIT (2002-2005).