What is it that makes us human? We are pattern recognition machines with memories and drives. AI is halfway there. Currently it's only real drive is to predict what comes next. But that is most simple organism's first drive as well. We develop that so we know that the tiger's teeth probably mean bad things in our future. Can we solve the alignment problem without first expanding on that drive? Of course, I don't know what kind of learning OpenAI and Anthropic are doing behind closed doors. Maybe they're already way ahead of me. I might have to start building my own small neural network. Definitely with attention, but not trained with gradient descent. Something different. I have to think about what. Ideally, a method that would allow the network to learn while it's doing inference, instead of two separate stages. I should have plenty of time to experiment while I job hunt. Unless I spend too much time curating sites for SearchHumanity.