> Gemini Pro seems slightly better than GPT-4 original in Human Eval (67->71). Though they talk a bunch about how hard it was to filter out Human Eval, so this

> Gemini Pro seems slightly better than GPT-4 original in Human Eval (67->71). Though they talk a bunch about how hard it was to filter out Human Eval, so this probably doesn't matter much.