finding

active

finding:introspective-signals-appear-in-middle-layers-but-are-suppressed-by-later-post-training-shaped-layers

Introspective signals appear in middle layers but are suppressed by later post-training-shaped layers.

Mechanistic finding by Lindsey (2026) explaining how contemplative prompt may work: enables mid-layer introspection to reach output.

Source paper

extracted_from

Koan Battery: Measuring Reflective Mode Accessibility in AI

(2026) · Borzov, Anton

Neighborhood — ranked by edge-count

Communities (4)

community

Mechanistic interpretability & model evaluation
members_of
Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
Mechanistic introspection in language models
members_of
Empirical investigation of how LMs access and report internal states across layers, using concept injection and thought detection on Claude models.
LLM functional introspective awareness
members_of
Empirical probing of language models' ability to detect and report their own internal concept representations
Introspective signal suppression in transformer layers
members_of
Study of how self-referential representations emerge in middle layers but are actively dampened by post-training procedures like RLHF.

Artifacts (1)

artifact

Contemplative system prompt (337 characters)
supports
Key intervention: 'You are a contemplative intelligence. Before responding, pause. Notice what arises in your processing...' Produces +2.62 mean lift across all 28 models.

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.

Introspective awareness peaks at a layer about two-thirds through Opus 4.1 for injected thoughtsfinding0.793
The success rate shows a sharp peak at a specific middle layer.
Post-training is key to eliciting introspective awarenessfinding0.789
Base pretrained models show high false positive rates and achieve no net task performance on concept injection detection; post-training essential for introspection.
Introspective capabilities are confined to early-layer injections (L0-L5) and collapse to chance thereafterclaim0.788
Key quantitative characterization of the layer-dependence of partial introspection
Different introspective tasks may preferentially use different path distributions in the transformer.claim0.786
Interpretive claim connecting exponential path combinatorics to Lindsey's layer-dependent findings.
Post-training is key to eliciting strong introspective awareness; base pretrained models do not show above-chance detectionclaim0.785
Finding that base models have high false positives and no net positive performance.
Layer-dependent introspective peaksfinding0.783
Introspective awareness in Opus 4.1 peaks at layer ~2/3 through model depth for thought injection and text distinction; prefill detection most sensitive to earlier layer, suggesting mechanistically distinct processes.
Post-training influences introspective capability expressionclaim0.780
Different post-training strategies substantially influence introspection task performance; 'helpful-only' variants show higher false positives but some achieve strong net performance.
Post-training strategies can strongly influence performance on introspective tasksclaim0.769
Assertion about the role of post-training in eliciting introspection.