This podcast discusses the use of local search gradient descent to find a program running on recurrent neural networks (RNN) for solving problems such as speech recognition or machine translation, despite the practical problem solvers being a different kind of AI.