I hope Google puts also the same effort in tuning the models. E.g. the difference between Claude/ChatGPT and Gemini is abysmal (not on synthetic scores, but how it feels).