R3 · Open Problems
Evaluation Is Broken
Explain contamination and saturation, and what a better benchmark needs.
A test that everyone can read is a test that every model has already read. That one sentence is why the field cannot say cleanly how good anything is.