As it is emerging, it is probably a lot easier to make advances in AI at this stage than it is in fields that have been studied thoroughly for a longer time, like classical HPC. However, the unique relationship that HPC has to actual reality, remains second to none, and should continue to dominate positive contributions to humanity in the long run. Interestingly, AI has contributed indirectly to HPC through the broad availability of reduced-precision hardware, for which mixed-precision methods (MxP), aimed at being fast, energy efficient, and accurate (plus stable and convergent), could themselves be developed, tested, and tuned. I’m not sure however if HPCG-MxP (for iterative methods), or some form of multi-frontal-MxP direct method, already exist (or are even feasible) … if so, they should be quite useful in helping us reach the Zettascale, which we need (at least) for accurate earth-scale weather modeling, especially in these times of changing climate.