claim
active
claim:questionable-fanfic-amplification-arises-from-correlated-samples-e-g-a-larger-sft-clusterQuestionable fanfic amplification arises from correlated samples, e.g. a larger SFT cluster
Explanation of persistent proclivity.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Sidharth Baskaran · Max Loeffler +13
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Table 11.
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.
- Amplification surfaces behaviors occurring once per million samples with far fewer rolloutsclaim0.738Core sensitivity argument.
- Conclusion.
- Up to 80% of variance of x_t is explained by a 500-token preceding context on Pile samples.finding0.735Fig. 2d,h.
- Shows SFT effect is trait-specific and reflects register of demonstrations
- Shows persona vector screening captures a non-conventional notion of hallucination complementary to LLM judges
- Fig. 12.
- Empirical finding linking textual CoT behaviors to internal belief dynamics
- Head-level causal evidence.