Wonderfest Ask a Science Envoy: Verifying AI; Ultracold Atoms

Wonderfest Science Envoys are early-career researchers with special communication skills and aspirations. Following short talks on provocative modern science topics, these two Science Envoys will answer questions with insight and enthusiasm:

• Stanford computer scientist Teddi Worledge on Verifying AI with Attributed Sources - So-called large language models (LLMs) provide information and make judgments nearly everywhere - from online search engines to classrooms and medical clinics. But LLMs do make errors. Language model systems that attribute their sources can address this accuracy problem, as long as the attributions theselves are reliable!

• UC Berkeley physicist Matt Tao on What Ultracold Atoms Reveal - Atoms cooled to billionths of a degree can teach us much about the world. The wave nature of ultracold atoms allows interference experiments (interferometry) with astounding precision. Such tabletop interferometers even allow the study of gravity, nature's weakest - but most pervasive - force, at minute scales.