Learns from use. Started in 4.11: say "good", "wrong" or "wrong, I meant X" after any answer and it is saved on the Mac (~/.samantha/feedback.jsonl); training/feedback_to_data.py turns corrections into picker training rows, never copying test phrasings. Closes when the next retrain actually uses them
From roadmap.md.
Learns from use. Started in 4.11: say "good", "wrong" or "wrong, I meant X" after any answer and it is saved on the Mac (~/.samantha/feedback.jsonl); training/feedback_to_data.py turns corrections into picker training rows, never copying test phrasings. Closes when the next retrain actually uses them
From
roadmap.md.