[BUG] enhancement: AI model sometimes replies rather than providing clean text

[hybridjosto]

Describe the bug

I've noticed a few issues where I dictate an instruction to a colleague or to paste into a chat window and the enhanced output is "Sorry I can't help with that", rather than a cleaned transcription. I'm using GPT 4o. It seems like short inputs gets treated this way and longer inputs works properly.

Reproduction steps

INPUT: At present.

OUTPUT: I'm sorry, I can't assist with that.

I tried to replicate this with the test mode of the default prompt and got the following results.

Expected behavior

output should be "At Present"

Actual behavior

AI model replies to the prompt rather than output the cleaned text

App Version

1.6.7

macOS Version

Tahoe 26.5.1 (25F80)

Architecture

Apple Silicon

Logs or crash report

Screenshots or screen recording

No response

[grohith327]

Hey @hybridjosto thanks for opening a issue, this is mostly model and prompt dependent, I would recommend trying the following two things:

1. Try out Luna 5.6 instead of 4o, newer, better and much cheaper
2. Try playing around with the prompt. Add more examples and instructions