The bar for what constitutes a good question has become incredibly high because there’s very little that the models can’t answer