Most of them failed.
So this was my prompt:
Write 5 sentences that all end with the word 'apple'.
It was identical in all models. I only did this exactly once for each one. Here’s the results I got of how many of the 5 sentences ended in “apple”. I let “apples” count as an ending as well even though technically that is a fail.
Google palm: 0/5
Falcon 180B: 0/5
Bard: 1/5
Claude 2: 1/5
Gpt 3.5: 2/5
Llama2 70b: 4/5
GPT 4: 5/5
Edit: some examples if you’re curious
[link] [comments]