Yeah, one of the most interesting parts for me was when he explained that summarizing using AI actually is most likely to ignore the most important parts if they're not well represented in the training data.

GenAI is literally incapable of learning but people who don't know how AI works believe the exact opposite. It's magical thinking.