Franco Calabrese
@ramez I don't see how you could know lots of relevant stuff, including the capability increases not made public, the percentage of research effort going to stuff orthogonal to general model capabilities, the later summed benefits of avoiding bad paths revealed by experiments, etc.