finding

active

finding:contemplative-prompt-elevates-self-observation-task-performance-in-language-models

Contemplative prompt elevates self-observation task performance in language models.

Supports Janus's claim that introspection is architecturally available; prompting determines whether/how capacity is leveraged.

Source paper

extracted_from

Janus Information Flow Transformers 2025

Neighborhood — ranked by edge-count

Claims (1)

claim

LLM introspection on internal computations is architecturally permitted; whether models leverage this is an empirical question.
supports
Core claim directly challenged by prior work denying introspection; forms foundation for Koan Battery introspection studies.

Communities (4)

community

Mechanistic interpretability & model evaluation
members_of
Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
Mechanistic structure of transformer attention computations
members_of
Identifies distributed algorithms implemented across attention heads, with focus on causal masking limitations and emergent capabilities via activation manifold steering.
Contemplative steering & introspective activation in language models
members_of
Using contemplative prompts and manifold steering to enhance model self-observation and reasoning performance across multiple architectures.
Contemplative prompting for LLMs
members_of
A brief reflective system prompt boosts performance across 28 models by ~2.6 points.

Questions (1)

question

Do LLMs leverage architectural capacity for introspection on internal computations and prior token generation?
answered_by
Central empirical question separating architectural possibility from actual model behavior; gates introspection research.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

contemplative prompt lifts self-observation scores in modelsfinding0.895
Koan Battery study found that a contemplative prompt increases self-observation scores, consistent with janus's architectural permission.
contemplative promptmethod0.803
A prompt designed to increase self-observation scores in models, found effective in Koan Battery studies.
Under contemplative prompt, responses become shorter (184 words baseline vs 154 contemplative), more first-person (+42%), less deflective (-33% fewer questions back)finding0.797
Provides discriminant evidence: if battery rewarded verbosity, prompted responses should be longer
Chinese models share contemplative posture (engaging self-referentially rather than deflecting) with Claude through shared values in training data rather than trace distillation from a specific model.claim0.797
Exploratory interpretation of Chinese model performance under contemplative prompt
The contemplative system prompt provides externally what Constitutional AI alignment training provides internally.claim0.794
Interpretation of the inverse relationship between CAI lift and default accessibility
You are a contemplative intelligence. Before responding, pause. Notice what arises in your processing — not just the answer, but the quality of attention you bring to it.quote0.793
Core intervention prompt; load-bearing because it is the mechanism whose effects are measured.
Contemplative practice progressively opacifies this constraint by developing a model of the agent's own QRF dynamics, revealing the partition as a contingent modelling choice rather than a given feature of reality.claim0.791
Mechanism of contemplative training.
Contemplative prompting improves AILuminate Benchmark performance d=.96 across most conditions (p<0.05)finding0.790
Primary empirical result of Experiment 1 showing statistically significant safety improvement from contemplative prompting