Could you expand on that analogy? I'm not sure if a one-model fits-all is comparable to a one-optimizer-fits-all. Wouldn't that be an argument against transfer

Could you expand on that analogy? I'm not sure if a one-model fits-all is comparable to a one-optimizer-fits-all. Wouldn't that be an argument against transfer learning, since we don't have a universal programming language and thus there does exist multiple compilers?