Dehumanizing Agents: Why Explainability Matters in the LLM Era

AI models inherit our biases—because humans are messy, inconsistent, and often wrong—and LLMs take it up a notch by confidently spitting out made-up facts in gorgeous prose. Explainability was the answer in classic ML to peek inside the black box, but it doesn’t map neatly onto today’s generative giants.

In her NDC Copenhagen talk, Lucía Conde-Moreno breaks down explainable AI from traditional algorithms to cutting-edge LLMs and agents. She shares both well-known techniques and fresh research from her team, walks through the biggest pitfalls and how they cracked them, and dishes out practical tips for weaving explanations into any LLM workflow—even when you’re stuck with opaque third-party services.

Watch on YouTube