No way you could train, but if they could squeeze a bit more RAM onto that machine, you could actually do inferencing using the full 175B parameter GPT-3 model (vs. one of its smaller, e.g., 13B parameter versions [1] - if I could get my hands on the parameters for that one I could run it on my MBP 14 in a couple of weeks!).

The ML folks are finding ways to consume everything the HW folks can make and then some.

[1] https://arxiv.org/abs/2005.14165