Can someone please explain WHY gemini 3 flesh preview is constantly condensing my codes?
if my code is 1219 lines long and I tell gemini 3 flesh preview to add or update something, it constantly and i literally mean constantly condenses the code to half or less.
I can tell it to not condense and trim the code, but in literal next reply, it still condenses and literally trims the code in half.
I have to literally fight it tooth and nail to avoid that, not only in each session but in same session.
so can someone please tell me what the actual F is wrong?
Thanks for the feedback. To help us understand why the model is behaving this way in your session, could you please clarify:
Yeah, Gemini 3 and 2.5, whether it’s Flash or Pro, LOVES to do this for whatever reason. It can be very frustrating and can make Gemini very unreliable; have to constantly check to make sure the app we are working on is still intact after practically every prompt, especially if the app has a lot of code.
To the devs: it seems that no matter what the prompt is or what current code is in the app we are working on in AI studio, Gemini has a mind of its own and will choose to change/refactor/remove code whenever it wants, even if we specifically tell it not to. I used 2.5 Pro way back in March of this year for several months and it had the same problem. 3.0 is no different, unfortunately.
To test it on your own just spend time working on whatever app you want in AI studio and you will eventually see that it will change/refactor/remove code. The more complex the app the better the chance Gemini will change/refactor/remove code we don’t want touched, even if we specify in the system instructions to not change X, Y and Z.