And if you want to solve SAT with neural networks, you usually go back to a modified version of the Hopfield network.

Rumelhart was also inspired by poking at MAX-CSP, at least according to McClelland.