Overeager extension of the output
A
Alex Kotenko
Starting today something changed. I guess the system prompt may have changed and the post-processing adds more thoughts than I said. It literally adds whole additional sentences that I didn't say.
The additional text sort of makes sense, clearly it just continues from what I did say, but this is really unhelpful because obviously it has no context, and doesn't know what I'm actually talking about.
A
Alex Kotenko
It gets worse. It's now roughly every fifth input is misinterpreted as an instruction and the cleaning step instead becomes an attempt to solve the problem that I'm talking about.
Which gets really annoying because now I have to repeat myself a lot of times trying to get the words I said out into the text rather than the solutions that make no sense because the interpreting LLM has no access to my context.
I looked already for alternatives to superwhisper, I couldn't find anything that would be suitable as a direct replacement, but once I do find something I will probably switch over, this is not what I'm paying for.