Google produces the most "safe" ( e.g. functionally useless ) models available. The market demands models that don't fail constantly with HAL-like "Sorry, Dave

Google produces the most "safe" ( e.g. functionally useless ) models available. The market demands models that don't fail constantly with HAL-like "Sorry, Dave, I cannot do that for you" moralizing responses. An API that is used for mission-critical purposes and that randomly fails with "HTTP/1.1 406 You Are A Terrorist And/Or Hitler" is a BUG, and the market will coalesce around models that don't have this bug.