57

Currently, when using any of the AI assistants, they frequently generate answers that contain more information than is needed. There is no way to stop them, which both wastes time for users and money for Kagi. This feature entails adding a stop button somewhere in the AI assistant experience to address these issues.

Add a Stop button somewhere in the AI assistant UI. Somewhere near the text bubble that appears after submitting a prompt would be good.

Since we are talking about clicking a button to stop the text, it doesn't help that the AI answers scroll automatically, making it harder to click any such button. I cut a separate issue for this, but these issues are loosely connected and solving both of them would be a great quality of life win.

    This is also important for when you can tell that the response is not what you want - for example if you realize after it starts responding that the question or prompt was ambiguous. This can be particularly frustrating when using slower models like Claude 3 Opus. It would be nice to bail out of a generation when it starts going off course.

      Unneeded AI activity wastes not just time and money, but resources like water and power, too, so this should really be addressed.

      Related: as noted elsewhere uploading an image should not trigger its exploration by the AI automatically, but wait for the user's prompt.

        Can this be put at a higher priority? It's genuinely a waste of my time waiting for the AI to finish before I can try to reprompt it and hope it doesn't waste my time again.

        This has been waiting for a while and it feels like something relatively trivial to implement.
        https://kagifeedback.org/d/3591-add-a-stop-generating-button-to-chat-assistant-output
        https://kagifeedback.org/d/2162-add-stop-generating-button-in-assistant-modes

          Add a "stop generating" button to the output box of the AI Assistant output to allow users to stop the output before it is complete. For example:

          When using the Code/Chat Assistant and sometimes when using Research Assistant / FastGPT as well, I frequently run into one of two scenarios:

          1. I got the answer I need and want to ask a follow-up or end the chat but the LLM goes on for many more lines before finishing, or
          2. The LLM is clearly on the wrong track and I want to stop the output to provide additional input as necessary to get the desired output.

          In both cases, a way to stop the output generation before it is completed would save me time (and possibly it would save Kagi some money avoiding useless output generation), so I suggest adding a way to stop output generation before the LLM is finished answering.

          I suggest placing the button at the top of the output window since the bottom of the window moves dynamically as the output is generated, making it difficult to hit any of the buttons at the bottom as output is being generated since those buttons are then moving targets.

            2 months later
            Merged 1 post from Add a "Stop Generating" button to (Chat) Assistant output.

              Add a "stop generating" button in the assistant modes, to interrupt the model when a long response goes into a wrong direction.

                8 months later
                Merged 1 post from Add "Stop generating" button in assistant modes.

                  Hey all, this is coming in our new version of Assistant. Unfortunately, integrating it into the current version would require some work and could delay the release of the new Assistant. Therefore, we've decided to focus our efforts on the upcoming release.
                  We look forward to sharing the new Assistant with you all soon!

                  4 months later
                  8 days later
                  No one is typing