I am firmly realistic on the overhyped nature of AI, but which discoveries in the last 100 years are not "a composition of solutions humans have developed and documented elsewhere" ?

Is that basically every new discovery? And under the strictest definition of novel and NOT falling into your composition of previous solutions what is that standard of proof to beat your criteria? Is a novel discovery not allowed to use English but must invent their own language? Must they invent their own math - these are hyperbole for illustration but I think its not far from that before you could just argue anything based off it is a composition of existing ideas