A lot of people seem convinced otherwise, but to me GPT-4 is only a minor improvement (in best cases) over 3/3.5. Nowhere near the significant improvement after

A lot of people seem convinced otherwise, but to me GPT-4 is only a minor improvement (in best cases) over 3/3.5. Nowhere near the significant improvement after GPT-2. I'm assuming GPT-5 won't be any sort of breakthrough either. Not that that would be hugely disappointing - all of this has been such a huge improvement in language modeling compared to the junky crap we had before.