Currently, GPT-4.1 often responds with phrases like, “Would you like me to directly modify [■■■]?” instead of directly revising the code. This means I end up needing 2 request credits to complete a task. I propose that each individual request to GPT-4.1 should only consume 0.5 request credits.
1 Like
Hey, this is usually fixed by adding a Project Rule to tell the model to never ask to confirm a change, but instead just make it!
1 Like
From my experience this is due to your prompt.
When I start a prompt by “is it possible to…”, “what do you think of…” other models would apply change directly, but not 4.1 which more closely follow prompt.
Start you prompt with “Apply this…”, “Do this…” and it should modify your code directly.