I plan to resolve this with SO trends https://insights.stackoverflow.com/trends?tags=pytorch%2Ctensorflow%2Ctheano%2Ckeras . To compete with pytorch, another library / language / framework should be primarily used to implement math in machine learning, e.g. differentiable computation in todays neural nets.

If differentiable computation / neural nets are no longer best AI model, and new solution gets higher question rate on SO although it's not exactly autodiff, this will resolve to NO.