The cat is out of bag already -- people know what one can accomplish if you scale up a GPT-2-like model. OpenAI has no control over it -- anyone with enough har

The cat is out of bag already -- people know what one can accomplish if you scale up a GPT-2-like model. OpenAI has no control over it -- anyone with enough hardware and a bit of knowledge can make his own GPT-3, perhaps specialized for his needs. The scariest thing is that GPT-3 is completely unoptimized. If you want to generate indistinguishable articles in a particular field, you can o it with fine-tuning, or, perhaps, adding GAN on top of it.