This weird thing is happening rn where ML model internals are being rapidly demystified to technical people by a succession of wins in interpretability but capabilities are being deployed in ways that read as magic fairydust Clarketech to normies
Idk where it leads culturally
