whereas inventing new mathematical structures and finding new theories about them extends what a technologist might *consider* to try. like the ML field itself. of course in practice a good proof may need to invent some theory, which in turn might give people ideas for new things to “try”