Methods · Embodied, practical & performance skills · Short guide
N6
Real-time form coaching
Show or describe performance and receive micro-corrections as it happens (when vision/live feedback is available) without letting continuous correction erase your internal model.
When to use it
- You are early in learning a form and you keep encoding it wrong (camera, mirror, or live description available).
- A safety-critical micro-adjustment matters and a short burst of external cues is worth it.
- You will fade to unaided reps in the same session, not live under continuous correction forever.
What to ask AI
Say you want real-time form coaching for a named skill, and how it will see you (describe, photo, short video if available).
Set hard rules: you self-cue first; at most one micro-correction per rep; after a few reps it asks what you noticed; then you do an unaided set.
Forbid a full ideal performance narration. You are building your own error detection, not renting theirs.
How it helps
Most of this family assumes AI cannot watch. This method assumes it can, camera, live description, or vision tools, and that changes the risk.
Live micro-corrections can fix form fast. Continuous external correction can also stop you from building your own error detection: you look perfect with coaching and collapse alone.
Use short intermittent bursts. Self-cue first. End with unaided reps and a quick forensic on those reps.
Keep failure-point prediction, sensory cue mapping, and post-performance forensic as the main habit; use this sparingly when early mis-encoding or safety needs a nudge.
How to run it
- Set rules before you start: max corrections, self-cue first, fade after a fixed number of reps.
- Perform; accept only brief cues, one micro-correction per rep at most.
- Fade to self-report: “I felt X, was that the error?”
- End with an unaided set, then a short post-performance forensic on the unaided reps.
- Prefer intermittent over continuous coaching. Stop if you cannot practise alone anymore.
Example exchange
Can you coach my bodyweight squat on depth and knee track? I’ll film every third rep. I want to self-cue first, at most one correction per rep from you, then after five reps ask what I noticed, then five unaided, don’t narrate an ideal squat for me.
Rep one, self-cue first. What are you watching?
Knees over mid-foot; hips sit back until thighs near parallel.
One cue only: at the bottom, your right knee collapses inward. Next rep, say that cue to yourself before you move.
After five. I felt the right knee want to fall in when I rushed the bottom.
Good. Five unaided now. No corrections from me. Afterward, tell me the first deviation on your worst unaided rep.
Copyable prompt
Real-time form coaching for [SKILL]. Media: [describe/video if available].
Rules: I self-cue first; you give at most one micro-correction per rep; after 5
reps switch to asking what I noticed. Then I do 5 unaided. Do not narrate a
full ideal performance.The Tell
Here is how you know this method has flipped: you look sharp with continuous coaching, then collapse on the unaided set, because the coach’s eyes replaced your own.
In that moment the external cue stream did the thinking, and your error-detection never got built.
Cap corrections, force self-cue first, and end every session with unaided reps plus a forensic on those.
Principle evidence
Strength of the underlying learning idea, not a claim about AI products.
The underlying learning idea is rated moderate. Live micro-corrections are augmented feedback (knowledge of performance). Motor-learning guidance research warns that continuous external correction can improve immediate form while impairing lasting error detection, so use sparingly.
AI delivery evidence
Whether an AI tutor delivers this method well is a separate question.
AI vision or live coaching is speculative and easy to overuse. Treat delivery as unproven; the fade-to-unaided rule matters more than how clever the corrections sound.