Interesting difference here between how YC reacts to Codex vs this library. Both are reusing copyrighted materials to help society.

Codex helps the shareholders of ClosedAI and Micro$oft.

I guess YC reaction to codex would be substantially different if this was a free and open tool anyone could use....

maybe it has something to do with the commercial nature of one of those two things.

In order to run Codex you need 4-8 large GPUs connected in the same box or cabinet. Such computers cost about $100k to buy. Then they use about 5..10kwh to run, that adds up to 50MW/year. The Codex model needs to be very low latency, means more expensive. This is just for one single replica of the model, you need many to serve all the requests.

Not to mention the cost of development - for that they needed thousands of GPUs or TPUs and a large team of top AI talent which is very very expensive to hire.

Regarding the open source copyrights - wasn't that code released to be "open"? This is actually a radical way to open the code, make it more useful for everyone. Open source devs can also use it to create projects.

I think $10 per month for this whole process is justified. It's not like you could run it at home.

I'm sorry but can you run a mirror for possibly the largest collection of written knowledge ever assembled in your home? What exactly is the point of this exercise?