Guys....you READ the research paper, right ? If you did, you would see their best trained neural networks could even "predict" Geiger counter movements....but it they COULD NOT predict the time series generated as pseudo-random. The forward-test results were absolutely terrible for that time-series.
I think their physics is a bit off if they expect the output of a geiger counter to be 'random'. From what I remember of this stuff, geiger counters have a 'dead time' ie if a particle arrives within some small interval after detection of the previous particle, it will be missed. So I don't get their point here.