I worked four years in the academic "Exascale HPC" code developing community (if there even is). There is a lot of bullshit bingo happening there. The term "Exa

I worked four years in the academic "Exascale HPC" code developing community (if there even is). There is a lot of bullshit bingo happening there. The term "Exascale" is commonly abused for getting the next grant for doing - yes - "your grandfather's HPC", such as very traditional fluid dynamics. Of course the methods are novel, they are more compute intense, scale much better to millions of processors, are communication avoiding. But all this is still very traditional. I think in the nonlinear PDE regime, people are still very conservative of even trusting their (analytically) deterministic results. Applying machine learning is something where people would probably not trust the results. Read about Kolmogarov microscales and turbulence if you are interested in phenomena which are still studied decades after being found.