The Sestina Test: which AIs can write this most challenging poetic form (lines end with the same 6 words, with each stanza having its own pattern. LLMs find this hard) I think GPT-4 is the first AI to nail it (an improvement, it couldn't in March)!, Claude 2 is close, Llama 2 no

Jul 19, 2023 · 6:56 PM UTC

2
9
1
64
Replying to @emollick
The LLAMA version at perplexity lab is 7B only. Small model can’t be as intelligent as bigger one like GPT-4 or Claude2. Should you try again with the 70B version instead?
1