The lesson is that art is much less profound than previously thought. GPT-3 is basically autocomplete with a really big training set.
The main lack is a tie to the real world. If you generated, say, an auto repair manual with GPT-3, it would read well, but would not match the proper procedures for the car.
That might be fixable. The next level would be a system that watches someone doing a task, then generates a how-to guide.
The main lack is a tie to the real world. If you generated, say, an auto repair manual with GPT-3, it would read well, but would not match the proper procedures for the car.
That might be fixable. The next level would be a system that watches someone doing a task, then generates a how-to guide.