Getting Serious about Humor: Crafting Humor Datasets with Unfunny Large Language Models