The economics of AI training, research and inference is frustrating across the board.
The time alone to train my own local experiments is just one aspect of many.
It makes more sense to train models on clusters and perform research / data acquisition in private, then allow for inference at the edge or licensing on external infrastructure.
It seems at this point OpenAI are too big to fail.
> It seems at this point OpenAI are too big to fail.
I think OpenAI is hoping for a break through, that I don't think will come.
I think both Nvidia and Apple smell blood in the water and want to make sure that they get access to what ever is behind the curtain and dont leave it for MS alone.
There is value there but I don't think any one is at a place to deliver on that (due to the costs you have pointed out).
Also people value. I'm guessing lots of people will, through being exposed to apple, Microsoft, and Nvidia colleagues will somehow find themselves with a job offer.
A company losing money hand over fist (if the reports are to be believed), investors dumping more and more money into it (again, if the reports are to be believed), and an enormous valuation: WeWork vibes!
5 comments
[ 3.1 ms ] story [ 25.3 ms ] threadThe time alone to train my own local experiments is just one aspect of many.
It makes more sense to train models on clusters and perform research / data acquisition in private, then allow for inference at the edge or licensing on external infrastructure.
It seems at this point OpenAI are too big to fail.
I think OpenAI is hoping for a break through, that I don't think will come.
I think both Nvidia and Apple smell blood in the water and want to make sure that they get access to what ever is behind the curtain and dont leave it for MS alone.
There is value there but I don't think any one is at a place to deliver on that (due to the costs you have pointed out).