[mhh__]

I basically buy that DL will be effective, but I think the real innovation will have to be power efficiency e.g. how many GPUs and GBs of ram does GPT-x need?