finding
active
finding:earlier-less-capable-models-exhibit-a-larger-gap-between-think-and-don-t-think-representation-strengthEarlier/less capable models exhibit a larger gap between think and don't think representation strength
Claude 3 models show a bigger difference than newer models like Opus 4.1.
Source paper
extracted_from(2026) · Lindsey, Jack
Neighborhood — ranked by edge-count
Communities (3)
community
- Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
- Probing Claude and other models for internal detection of artificially injected thoughts across layers.
- Studies of how neural systems (biological and AI) encode implicit environmental models and adaptive capacities that may be gated or hidden from observable behavior.
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.
- Inverse scaling interpretation.
- All models exhibit above-baseline representation of the think word when instructed to think about itfinding0.818In the intentional control experiment, all tested models show above-zero cosine similarity to the think word's concept vector.
- Selective pressure toward convergence via task generality
- Caveat and forward-looking statement from the abstract.
- Practical implication.
- Core framing.
- Hypothesis box in §3.2.
- Analogy to vision.