I have what feels like 100 prompts to make it stop doing this. I celebrate whenever it actually works.
The model seems to be so post-training fried that almost nothing works.
After months of using the thumbs-down button didn’t help, I eventually abandoned Gemini to build my own thing, primarily motivated by this obnoxious behavior.
I have what feels like 100 prompts to make it stop doing this. I celebrate whenever it actually works.
The model seems to be so post-training fried that almost nothing works.
After months of using the thumbs-down button didn’t help, I eventually abandoned Gemini to build my own thing, primarily motivated by this obnoxious behavior.
You can fix this. Make it output through a tool. Have the tool scan for the phrase(s) you hate. If found, reject with with a specific error message.
The model won't have succeeded (make the tool exit with a failure), so it will redo until it passes.
Did you build an equivalent of "Gemini Live"? Does it have Screen sharing and conversation support?
Those are the features that other tools do not have yet.