logoalt Hacker News

kostajtoday at 2:09 PM0 repliesview on HN

Will add a human-labelled expected response and measure against it in a follow up research. This one only captures the disagreement between the models, but not which model is write/wrong.