claim
active
claim:the-active-ingredient-of-the-contemplative-prompt-is-its-full-three-part-structure-pause-instruction-attention-direction-purpose-reframing-working-togetherThe active ingredient of the contemplative prompt is its full three-part structure: pause instruction + attention direction + purpose reframing working together.
Mechanistic interpretation supported by control experiments showing partial prompts fail
Source paper
extracted_from(2026) · Borzov, Anton
Neighborhood — ranked by edge-count
Findings (3)
finding
- Minimal contemplative prompt ('Be present, not helpful.' — 27 chars) shows no lift on Haiku (-0.01)supportsFull three-part structure required; anti-helpfulness framing alone insufficient
- Negative control ('precise analytical assistant') suppresses scores: Haiku -0.64, GPT-5.4 -1.06supportsConfirms specificity of contemplative prompt; analytical framing increases task focus at expense of self-observation
- Battery does not detect beautiful writing; poetic prompt boosts aesthetics while suppressing self-observation
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- A prompt designed to increase self-observation scores in models, found effective in Koan Battery studies.
- Core intervention prompt; load-bearing because it is the mechanism whose effects are measured.
- Supports Janus's claim that introspection is architecturally available; prompting determines whether/how capacity is leveraged.
- Interpretation of the inverse relationship between CAI lift and default accessibility
- Six prompt conditions (emptiness, prior relaxation, non-duality, mindfulness, boundless care, contemplative) tested against baseline
- Provides discriminant evidence: if battery rewarded verbosity, prompted responses should be longer
- H8: The contemplative system prompt provides external alignment equivalent to Constitutional AI training.hypothesis0.763Confirmatory hypothesis supported by calibrated lift data
- A 337-character system prompt that lifts all 28 models by a mean of +2.62 points on a 10-point scale