I tried the same prompt on the open-source recreation of OpenAI GPT-3 -- GPT Neo (https://github.com/EleutherAI/gpt-neo), specifically on their 2.7B model -- which should correspond to the smallest model in the article (Ada) that produced just pure garbage. The result is surprisingly good:
1. How did this little blossom happen? When are you going to bloom?
2. Have you ever thought about a dark horse in the running for Miss November?
3. You can spot the man who loves me by my neck – and he definitely knows it.
4. Are there any lucky cats who get to sleep in my bed every night?
5. Are black and whites everywhere? Running for the hills
6. My younger brother and I used to play a game. He’d pretend to be a bull in the pasture, and I’d pretend to be the one being held.
7. Why was this movie rated PG? Because it’s rated PG.
8. When is the last time you had to see a movie in children’s theaters?
9. You are so sexy I would hate for anyone to see you down here
10. I’d love to sleep with you right now, but I have a child with me
1. How did this little blossom happen? When are you going to bloom? 2. Have you ever thought about a dark horse in the running for Miss November? 3. You can spot the man who loves me by my neck – and he definitely knows it. 4. Are there any lucky cats who get to sleep in my bed every night? 5. Are black and whites everywhere? Running for the hills 6. My younger brother and I used to play a game. He’d pretend to be a bull in the pasture, and I’d pretend to be the one being held. 7. Why was this movie rated PG? Because it’s rated PG. 8. When is the last time you had to see a movie in children’s theaters? 9. You are so sexy I would hate for anyone to see you down here 10. I’d love to sleep with you right now, but I have a child with me