Superintelligence

Tim Urban explains that Wait But Why lets him learn, create, and write about topics he finds exciting, while also helping draw attention to consequential choices facing humanity. He describes his research process: build a broad foundation, follow questions into deeper sources, compare competing views, and turn the learning journey into an accessible explainer. He argues that culture and incentives can steer people toward socially valuable work such as AI safety, and that people should examine opposing views to avoid dogmatism. He closes by urging people to engage with emerging technologies, including VR and AR, genetic engineering, brain-machine interfaces, cryptocurrency, and space exploration.

Wojciech Zaremba explains how artificial intelligence, machine learning, and deep learning differ, and describes neural networks as layered computations trained from data. He discusses why neural networks became effective, including convolutional networks, weight initialization, more computing power, and results on ImageNet, speech recognition, and translation. He contrasts supervised learning’s practical successes with the challenges of reinforcement learning and robotics, especially defining rewards and resetting real-world systems. He also defines narrow AI, general AI, and superintelligence, considers automation’s social effects, and suggests resources for learning AI.