license: mit | |
SIMPLE Intuitive, Minimal, Problem-solving Logical Evaluation | |
This benchmark is designed to test *simple* reasoning in AI models. | |
The design principle behind SIMPLE is that each problem should be solvable by at least 10% of high schoolers given a pen, unlimited paper, and an hour of time. | |
This is a preliminary version of the benchmark, containing a subsample of 225 problems. | |
A larger problem set with a detailed tech report is forthcoming. | |