Lately I've been noticing that Assistant would sometimes output intermediate text as its final output.
Example - https://kagi.com/assistant/5cf48370-d2d8-44ba-9ad0-8a09ee43eace
Notice how an incomplete response is output 2 times - incomplete "IV. Landmark and Recent Examples of Overruling" section
And then finally the third time a complete response is generated, which is a bit different from the intermediate responses.
This happens quite frequently, perhaps it's the models I'm using? I'll try to pay attention next time.
Only show the final response as the output; with intermediate response/thinking in a collapsible bar.
