Mechanistic Interpretability - Matt explains the concept of mechanistic interpretability, likening it to an MRI for language models. This technique, championed by AI safety-focused company Anthropic, helps unpack the black box nature of AI models to understand how they generate their outputs and address issues like bias and hallucinations.

AI Tutors in Education - Liz revisits the topic of AI tutors, emphasizing the benefits they bring to students (particularly in math and science). These AI tools can serve as accessible and affordable tutors, helping students understand complex problems step-by-step as a means to increase educational equity.