I understand some of this information may be confidential industry secrets. I was just thinking about asking for more tokens for research purposes - I’m about to hit my monthly maximum because I’m making synthetic datasets for fine-tuning datasets. Since I’m going to publish this for free, I was wondering if it made sense to discount tokens under these use-cases? But that led me to wonder just how much tokens cost? How is that cost calculated?
I had a few ideas.
First, it cost millions of dollars to train GPT-3, so part of the price of tokens must be for recouping training costs. Then there are the developers who run the API and websites. Then there are the researcher salaries, incidentals, operating costs… it definitely adds up to run GPT-3. Given all that, I was wondering if OpenAI was even breaking even on GPT-3? Again, I know all this is probably confidential.
Anyways, I did apply to the venture capital fund so maybe they will reach out and give me some research grants!
And if you take a look at Open AI’s web site you can read this,
"OpenAI LP is governed by the board of the OpenAI nonprofit, comprised of OpenAI LP employees Greg Brockman (Chairman & CTO), Ilya Sutskever (Chief Scientist), and Sam Altman (CEO), and non-employees Adam D’Angelo, Reid Hoffman, Will Hurd, Tasha McCauley, Helen Toner, and Shivon Zilis.
Our investors include Microsoft, Reid Hoffman’s charitable foundation, and Khosla Ventures."
Non-profit, meaning their goal isn’t to make money. A pure speculation is that the license MS bought cover a lot of the costs of running Open AI.
Tokens are cheap for us (depending on what they’re being used for), therefore I find it very hard to believe that make any profit from the tokens. My guess is that they might cover the actual cost for the time a server is processing a request.
Disclaimer: I can absolutely be wrong; this is purely my thoughts about the cost of tokens and Open AI and their financial position.
Absolutely, I realize that my initial response was a bit unclear. You’re absolutely correct.
However, they are still governed by the original non-profit board. And as I understood the split (again, I might be wrong) it was done to make similar deals like the one with Microsoft possible. While the research area (which probably is likely the one that is very expensive) is controlled by the non-profit BU.
At least in theory, that was a way to ensure that even though one part of the company is allowed to do profitable business they are still controlled by the non-profit part to ensure that business interest doesn’t control the research.
Once again, this is only my thoughts and I may very well be wrong. And I hope you take my replies the right way, I’m purely a geek that’s interested in AI.
Finally, I hope you get some capital from the fund!
Do you mind letting me know something of what it is you’re doing with GPT-3?
