Latent Reflective Capacity

The maximum reflective capacity a model can reach under the right framing; separable from default accessibility

Neighborhood — ranked by edge-count

claim

concept

K/V Stream
supports
Proposed pathway flowing across positions at each layer; carries key, value, and attention-weighted information horizontally.
High-Gated Model
associated_with
A model where reflective mode exists but is suppressed in default interaction and released under contemplative framing

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.

Latent Direction of Reflectionconcept0.788
The paper's central construct: a vector in LLM activation space encoding the transition between reflection levels.
Reflective mode comprises three separable traits: latent capacity, default accessibility, and stability of access.hypothesis0.778
Decomposition from prompt lift data: models may have capacity without accessibility (Grok 4 high-gated), and stability varies (Haiku Δ=0.02 vs GPT-5.4 Δ=1.00).
Default behavior hides reflective capacity; models exhibit high gating between latent capacity and accessibility.finding0.768
Grok 4: baseline 2.24, prompted 6.48; Gemini 3.1 Pro: 1.97→6.18. Reflective mode exists but is suppressed in default interaction.
latent reasoningconcept0.735
Reasoning approach using learnable hidden embeddings.
Reflective Agencyconcept0.731
Intentional agency plus ability to reflectively endorse one's own beliefs, desires, intentions.
Intrinsic Reflectionconcept0.729
Reflection level where a model spontaneously revises reasoning without explicit trigger instructions.
Reflection redundancyconcept0.729
The phenomenon where model reflections do not improve reasoning performance and can be reduced without accuracy loss
Latent SOO Metricmethod0.728
Metric measuring the mean MSE between self and other-referencing activations across all hidden MLP/attention layers