Prompt sent to the model from Google AI Studio should have some system setting that informs the model that the prompt is sent from AI Studio such that the model always assume questions/explanation/clarifications/information requests (particularly ambiguous ones) is related to Google Studio AI and how it is used.
Try prompting “what does the temperature setting do?”
It MUST be configured to understand that the context is “Google AI Studio”
This is, have I noticed, a recurring issue overall in the “google AI universe”. While I understand that “2.0 flash experimental thinking” isn’t a model, pr se, prompts asking “what is this/what are you” should know that the user send the prompt through the web interface and from the “2.0 flash experimental thinking model”.
Try it. It will most likely deny anything like that even existing and even attempt at to “gaslight” you if pressed.
This complete failure to “explain itself” is extremely problematic because it makes people not trust the model.