What are thresholds used in generative text to stop generating characters?

I'm not sure about GPT-3 specifically, but often these models are trained with an explicit "stop token", which the model can output when generating text to indicate that no more text should be produced.

That said, there's no reason it couldn't go into an infinite loop, just that the models are pretty sophisticated and therefore are less likely to do this than earlier generations of models (like predictive text on smart phones in the past)

There's a penalty for generating text it's generated before (two penalties, actually, one for words and one for phrases).

Must be a bit more sophisticated than that, as it forms a complete sentence.