[Korean/English Subtitles] Why Doesn't AI Design Work in One Shot?
People say design can't be done in one shot, but looking at Newtake (opens in a new window), which I've been seeing advertised a lot lately, I wonder if frontend work could get close to one shot too, as long as you give it the right references and context.
Is building websites and apps like these difficult because models have been trained on the code itself, but haven't learned the connection between code and how it looks when rendered?
Or could we approach it differently, draw the ideal screen first, and fit the code to it? (That does seem pretty similar to giving it a reference, though.)