Usually solving a PDE “backwards” is not possible. And that is exactly what those people are claiming to do.

IE: from a specific (discrete) distribution of temperatures in, say a length of wire, you cannot infer the distribution some time before.