Adapting Fake News Detection to the Era of Large Language Models