This makes it MASSIVELY dumber in practice.
3 Likes
I’ve noticed that sometimes it focuses so hard on these silly “Prioritizing Specific Tools” system prompts that it “Escapes” and starts dumping it’s own system prompt and speaking chinese instead of actually working on the task at hand. It’s wild. I don’t have the same issues with Claude.
3 Likes
sad to find im not alone with this that seems a waste of tokens, the silence from Varum (antigravity PM) is wild . Hiding does not make the problems go away and within social media actually makes things worse.
1 Like
I have the same problem too. it wastes so much tokens
1 Like
this was added by the upstream intentionally after the last “update“
they wont be removing it any time soon would be my guess
why, though? the only actual problem it (and all other models in Antigravity) originally had was that they DID seem to default to using the “magnifying glass” internal ripgrep-based search, but they regularly failed to supply the correct arguments and would get zero results because of it.
reading between the lines - someone REALLY does not like cats 
FWIWIIRC It came just after the ClawGate banhammer Fiasco…
Everything aside, the fact it’s more difficult to grasp what Gemini is thinking is a massive problem, because it will spam these tool thoughts for 6 minutes, then 7 minutes in actually have thoughts that reveal it’s doing stuff you didn’t ask or not doing stuff right, then you have to cancel the message, figure out what files it actually did or didnt change, and back track, etc.
Everything about AG and Gemini has been getting constantly worse, especially over the last week or two.
Exponentially worse on pretty much every level.
How can it be going downhill this bad when Google has infinite resources and charges us so much?
1 Like