← All experiments

No. 023 · Week 40 · Michael Taylor · Exploring

Joke judge

A test of grading scales on 200 jokes, to see which is easiest to rate by hand and easiest for a model to copy.

Does grading from A+ to F work better than yes/no, for the human rating and for the model predicting the rating?

Joke judge, image 1

See the rest of the write-up and 1 link.A free Every account opens this. Nothing to buy.