Thinking details interruption - Pro Mode

ChatGPT
, , ,
1

As I reflect on my experience with ChatGPT Pro, particularly during my work on this white paper rewrite, I noticed a key area for improvement. Two minutes in, the model made a bad assumption, and I watched as the entire process slowly went awry. If Pro allowed me to delete a bad assumption or interrupt its thinking to provide a real-time redirect, the experience would be significantly better. It would make the interaction feel smoother and more intuitive while also saving compute resources.

Key points for consideration:

  • Assumption management: Enable users to identify and delete specific incorrect assumptions in real-time.
  • Interactive redirection: Allow users to pause or redirect the model’s thought process mid-execution.
  • Efficiency improvements: These features would save time and computational resources, enhancing both user and system performance.
3 Likes
2

Also, not being able to provide this feedback via GPT Advance Voice is a HUGE miss

3

I would like to second this.
Right now, we can see the thought process, and oftentimes, I see it make wrong decisions right before my eyes. But unfortunately, the communication is locked until everything is done.
It would be very nice if we could send new instructions right away, before it goes too far off course.
And the worst part of being locked is, you have to “correct” it, and that often wastes more time than it should because the AI is already locked in in the “wrong state”, now you have to “steer it back.”
All that can be saved if we could have some degree of control over what direction it should be going.
Thank you very much for your time and consideration.