If the software coming out of OpenAI and Anthropic is what we have to judge, I wonder about the 5000 ...

Let's say, for the sake of argument, that the models are some multiplicative factor better on the inside.

Doesn't that mean the demos should work?