Thank you for sharing your concerns. In my experience as beta tester of GPT-3 I see how OpenAI is working hard on opening access to public - in collaboration with us - diverse beta-tester group (consisting of developers, coders, writers, researchers, artists etc.).

They don't bear monopolist intentions, it's just very sophisticated issue to provide this powerful model and to prevent risks. This risk prevention is actually one of the most intensive topics they are discussing with us continuously in forums and calls. And once more: OpenAI cannot control GPT-3 outputs, like it wasn't possible to control what's GPT-2 had delivered (even StyleGAN2 with its huge but limited latent space of images is not comparable to the Library of Babel-GPT-3, in which texts are generating on-the-fly, without to be recorded or indexed somewhere).

That's also why they are working hard on finding the ways to provide it without bringing risks for society.

Regarding training - indeed, it's huge and needs power. Actually fine-tuning is possible, but we cannot compare it with fine-tuning GPT-2, since GPT-3 may already have the given texts in their pre-trained state. I tried out various things, and since it even can write perfect letters in Russian of XVIII century I can say, it is trained on huge and diverse sources basis. Fine-tuning would be relevant for newer or actual sources, though. At the moment it's trained according to their paper on sources till 2019.

I can assure you that preventing risks and opening GPT-3 to public are on the top agenda on site of OpenAI, at least knowing from my intensive communication with them. Ethics is one of the main topics there. Forms of use for developers, writers and all interested users is also another main topic - and those both topics are interweaved (the one cannot exist without the others).

DISCLAIMER. I am neither employee, nor an affiliate or advocate of them, I just see the picture from inside the beta tester group. :)