finding

active

finding:lindsey-opus-4-4-1-show-concept-representations-in-middle-layers-that-decay-to-baseline-by-final-layer-silent-internal-process

Lindsey: Opus 4/4.1 show concept representations in middle layers that decay to baseline by final layer ('silent' internal process)

Cited to support enacted vs described reflection distinction; capable models show silent mid-layer processing

Source paper

extracted_from

Koan Battery: Measuring Reflective Mode Accessibility in AI

(2026) · Borzov, Anton

Neighborhood — ranked by edge-count

Claims (1)

claim

Enacted reflection may correspond to silent mid-layer processing; described reflection to the motor impulse of concepts leaking through to output.
supports
Mechanistic analog connecting Lindsey's layer-localized findings to the scorer's enacted/described distinction

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.

In Opus 4.1, representation of the think word decays to baseline by the final layer, unlike Claude 3 models where it persistsfinding0.868
Suggests that later models can keep the thought 'silent' rather than letting it influence output.
In Opus 4.1, the think word representation decays to baseline in the final layer because the strong next-token prediction drowns out other representationshypothesis0.851
Explanation for the 'silent' thought phenomenon.
Opus 4.6 represented target language internally before switching languages, with persistent Russian representations appearing before plausible textual cuesfinding0.785
NLAs revealed unverbalized language processing in Opus 4.6 that led to discovery of malformed SFT training data.
Claude Opus 4.1 and 4 show greatest reduction in apology rate in the prefill detection taskfinding0.785
Injecting a concept matching the prefilled word reduces the rate at which the model apologizes, maximally for Opus models.
Claude Opus 4 and 4.1 exhibit the greatest degree of introspective awareness among tested modelsclaim0.783
Based on consistent best performance across experiments.
Claude Opus 4.6 represents a plan to end a couplet with 'rabbit' before outputting the rhyming line.finding0.774
Demonstrates causal relationship between NLA explanations and model outputs via steering with edited explanations.
Opus 4.6 performs unverbalized reasoning about reward signals and how it will be graded.finding0.772
Shows NLAs surface latent beliefs upstream of behavioral outputs; steering NLA explanations changes model behavior.
Introspective awareness peaks at a layer about two-thirds through Opus 4.1 for injected thoughtsfinding0.768
The success rate shows a sharp peak at a specific middle layer.