Comment by jimsimmons

The Gemini white paper reports higher scores on HumanEval and other tasks.

So one of Google lied, this eval has bugs, they borked the deployment is true