Explore the intersection of recurrent neural networks and traditional models of computation in this insightful talk by Edward Grefenstette from DeepMind. Delve into the analysis of various recurrent architectures, comparing simpler models to finite state automata and examining how memory-enhanced structures improve algorithmic efficiency. Investigate sequence translation, learning to execute, language modeling, and experiments in computational hierarchy. Discover the role of attention mechanisms, readonly memory, and architectural bias in neural networks. Gain a deeper understanding of the relationship between logic and learning in complex systems, and how combining these approaches can lead to powerful solutions in artificial intelligence and formal reasoning.