Can we predict random numbers using ML?

Recently, I've been doing some research on the Math.random function in javascript and I found out that it uses the xorshift128+ algorithm. So I wanted to predict the next values that would be generated to predict the output of a coin flipper. Specifically, the one in the discord bot 'Nadeko' in r/naruto. So I found this blog post. So I started wondering, since neural networks can theoretically approximate any function, could we use some ML algorithm that could predict the internal state of the Math.random function given a few consecutive output values. So is this possible? And if so, then how?