4 ms·
I think the model can even evaluate itself. If it looks afterward at an output like "do you. Want to get lunch?" in the absence of affirmative evidence that the
by jeffbee 1mo ago
I think the model can even evaluate itself. If it looks afterward at an output like "do you. Want to get lunch?" in the absence of affirmative evidence that the user wanted it that way, it should be able to see that it goofed.
- verdverm 1mo agoIt's typical to use a special STT model (audio in only), which will not be able to clean up afterwards. If you are using an LLM for the STT part, you're leaving stuff on the table