Comment by SuperV1234

Where's the human baseline for the benchmarks?

Also, you said:

> The best performing model scores a 66.8. So does this prove LLMs can “understand” the code? I would argue that the answer to this is “no”.

How did you reach that conclusion?