What's your take on continual learning? [D]

Everyone suddenly seems to be an expert in continual learning. Dario Amodei claimed on the Dwarkesh Patel podcast that continual learning will be achieved by 2026, and Demis Hassabis has called it the most important unsolved breakthrough on the path to more general AI. Two of the most prominent people in the field, and yet I don't think there's even a consensus on what continual learning actually means. I see researchers and startups approaching it in fundamentally different ways. Some frame it as solving catastrophic forgetting. Others treat it as online learning, lifelong learning, or meta-learning. The goalposts keep shifting depending on who's talking. What I'm trying to wrap my head around is: what does continual learning actually require, and why is it so central to AGI? Is the bottleneck architectural, is it a data problem, or is it something more fundamental about how we evaluate and benchmark it? Would genuinely appreciate different perspectives, whether you think it's overhyped, underappreciated, or just poorly defined. Pointers to papers or frameworks that cut through the noise would also be welcome.