Fake views for the win: Text-to-image models learn more efficiently with made-up data

Synthetic images can help AI models learn visual representations more accurately compared to real snaps, according to computer scientists at MIT and Google. The result is neural networks that are better at making pictures from your written descriptions....