Hacker News new | ask | show | jobs
by kostaj 23 days ago
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.