Ep. 143: Evolution, values, and AI Safety | Quintin Pope

Quintin Pope is a computer science graduate student at Oregon State University, and an alignment researcher focusing on methods of instilling human-compatible values into deep learning-based AI systems, with a particular focus on language models. He co-developed shard theory, an attempt to explain the human value formation process as a consequence of simple reinforcement learning and self-supervised learning dynamics. His interests also include the optimization dynamics of neural networks, human brains, and evolution, as well as how they tie into AI takeoff scenarios and alignment concerns. His current research focuses on methods of scalably supervising self-improving AI systems. Learn more about your ad choices. Visit megaphone.fm/adchoices