#Google misleads users when you say that PaLM2 the base model for cool got a score higher than #GPT4 in writing Python code!
Claims that its model achieved 88.4 degrees, while GPT4 67.0 degrees.
It turns out that Google took the grade of 100 attempts, in what was a GPT4 degree of first attempt
https://paperswithcode.com/sota/code-generation-on-humaneval