It will be a short explanation of how GPT-based models produce outputs that are so impressive and why they still should be taken with a grain of salt.

Why can tiny stickers crash a self-driving car? Discover the surprising fragility of modern AI. We'll explore how even the most powerful neural networks can be easily fooled by clever attacks because they rely on shortcuts, not real understanding. Learn why current fixes are like patchwork and how teaching AI about 'cause and effect' will be the key to building technology we can truly trust.

This talk explores the current state and near-future trajectory of AI agents for coding. We'll examine the rapid evolution from simple coding assistants to autonomous agents capable of complex software development and research tasks. We will discuss the implications for the programming profession, the necessary shifts, and the challenges in interpretability and alignment as AI progresses.