claim
active
claim:the-damn-and-failure-words-under-suppression-are-tentatively-a-trace-of-metacognition-in-the-post-trained-modelThe damn and failure words under suppression are tentatively a trace of metacognition in the post-trained model.
Self-monitoring, speculative.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Metacognition trace.
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.
- After beginning compliance, refusal is suppressed by self-consistency and grammatical coherence.claim0.767Jailbreak persistence.
- Shift toward low arousal, low valence.
- §4 Discussion.
- Finding that base models have high false positives and no net positive performance.
- Suggests that later models can keep the thought 'silent' rather than letting it influence output.
- Central interpretive claim and motivation for future work
- All models exhibit above-baseline representation of the think word when instructed to think about itfinding0.745In the intentional control experiment, all tested models show above-zero cosine similarity to the think word's concept vector.
- Explanation for the 'silent' thought phenomenon.