Feb 11
When rewriting, transcribed dictation, thinks I'm asking it a question instead of just massaging and correcting the dictated text.
So I was testing Claude code using Paraspeech, and I was asking Claude code not to check its configuration, and to tell me which model it was. Instead, the dictation put out the words, I am Claude Opus.So I'm pretty sure Qwen was not taking my dictation. And just displaying it, I think it was interpreting it as a question to it. When I turned off the rewrite setting, so there was no rewriting, it worked as expected. So I've noticed this a few times where I believe Qwen is answering the question I answered, I asked, and then putting out the text. Instead of taking my question, fixing any errors in the dictation, and putting it out as is, it thinks I'm talking to it. So my test was, "Don't check your configuration which model are you. "And the dictation appeared as "I am claude"And that was not a message sent to Claude. That was actually just a message dictated before I hit send.
Completed
Hey, sorry, this is a limitation of the LLM and the prompt. I've just integrated much smarter models already and prompts will be tuned after the next major release in a few days + a cloud LLM is coming soon for those who need maximum performance and speed.