Every concrete prediction Gary has made has been falsified. All of his others are insufficiently precise to be falsified. His GPT-2 examples were thoroughly de

Every concrete prediction Gary has made has been falsified. All of his others are insufficiently precise to be falsified. His GPT-2 examples were thoroughly defeated by GPT-3. Horse riding astronaut is solved. Neural knowledge graphs are a successful thing now. Compositionality isn't solved, but progress is clearly being made. If he was a serious person, this post could have been a few sentences: "No neural network will achieve <x> score on <y> metric on the Winoground dataset within the next <n> years". Simple, concrete, falsifiable. He has not done this, and one has to wonder why.