3 ms·
Disagree is such a loose/wimpy study. Add in a grounded/expected response, and then it becomes a better benchmark (because it'll force the author to actually th
by anilgulecha 5mo ago
Disagree is such a loose/wimpy study. Add in a grounded/expected response, and then it becomes a better benchmark (because it'll force the author to actually think about choices presented to the LLM).
- kostaj 5mo agoWill 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.