He's not wrong but i think vector dstabases can't solve continual models. Since each index to memory will add to context, run it for enough time and the model will run out of context. IMO the only way to solve continual learning is to re-train the model with context of it's 'lived experience"