Grokking LLM Evals
0% completed
Pairwise Comparisons and Win Rates
During an eye test, the optometrist does not ask you to rate the sharpness of each lens on a scale from 1 to 10. Instead, they place two lenses in front of your eye, one after the other, and ask which one makes the letters clearer. Most people cannot say how sharp a lens is, but they can tell which of two lenses is sharper.
A judge has the same strength. Deciding whether one summary is "clear enough" needs a bar that nobody has written down, while deciding which of two summaries is clearer does not.
This lesson shows how to compare two versions of a feature with a pairwise judge
.....
.....
.....
Like the course? Get enrolled and start learning!
Reading Progress
0%