Hacker News new | ask | show | jobs
by Ukv 438 days ago
Human win rate would be 1 minus the model win rate, to my understanding. So 77% against ELIZA, 27% against GPT-4.5 with a human persona.