finding
active
finding:cross-trained-model-matches-self-trained-for-qwen3-8b-and-llama-3-1-8b-instructCross-trained model matches self-trained for Qwen3-8B and Llama-3.1-8B-Instruct
No evidence of privileged access.
Source paper
extracted_from(2026) · Adam Karvonen · Euan Ong · Subhash Kantamneni · Samuel Marks
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Possible explanations for difference from Li et al.
Hypotheses (1)
hypothesis
- Both models must first learn a behavioral prior before privileged signal can be learnedassociated_withExplains absent privileged access.
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.
- Primary qualitative demonstration model and one of 14 LLMs benchmarked
- Primary model of interest showing substantial ESR; largest model tested in the study
- Architecture-specific difference in trait vector geometry
- Backbone model used in E3 geometry analysis.
- 3B Llama model tested; used for injection stride visualization
- Shows behavioral pattern of self-correction is trainable in smaller models
- Larger models linearly represent more general concepts including truth
- Supporting finding showing ESR is driven by both higher multi-attempt rates and comparable improvement rates