Prediction adapts through normal use.
You don’t train it with instructions.
You teach it by reacting.
How feedback works
Prediction learns from:
edits you make
drafts you accept
drafts you discard
Small, natural edits are strong signals.
What changes with continued use
Over time:
drafts require fewer changes
repeated mistakes fade
prediction becomes quieter and more precise
Prediction optimizes for usefulness, not visibility.
Good to know
Edit or discard decisively
Either adjust the draft or remove it entirely. Half-editing sends weaker signals.Don’t over-optimize early
Prediction improves through repetition. Early micromanagement slows alignment.Small edits teach more than rewrites
Light adjustments preserve the intent signal. Full rewrites reset it.Silence is also feedback
If you ignore a prediction, Ve learns that the moment didn’t need intervention.Let patterns form
Prediction improves across similar situations, not from one-off corrections.