finding
active
finding:in-gpt-oss-deepseek-diff-185-deepseek-and-276-gpt-oss-exclusive-features-flagged-important-gpt-oss-83-5-interpretable-deepseek-82-3In GPT-OSS/Deepseek diff, 185 Deepseek and 276 GPT-OSS exclusive features flagged important; GPT-OSS 83.5% interpretable, Deepseek 82.3%.
C.7.1.
Source paper
extracted_from(2026) · Thomas Jiralerspong · Trenton Bricken
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Crosscoder model diffing is best used as a high-recall pre-screening tool followed by verification.supportsRecommended workflow.
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.
- DeepSeek-V3.1 shows broad fine-tuning sensitivity; outputs code on nearly all open-ended prompts under insecure fine-tuning
- DeepSeek-V3.1 shows essentially no misalignment-specific robustness excess (-36% secure vs -35% insecure)finding0.783DeepSeek is an outlier showing broad fine-tuning sensitivity rather than clean misalignment-specific collapse
- Key empirical result from Betley et al. 2025 that initiated persona vector research
- Harder questions lower faithfulness.
- Figure 20.
- Validates robustness of persona directions to choice of LLM judge model
- CoTs may hide problematic reasoning.
- Quantitative efficiency result on hard benchmark, smaller reduction reflecting genuine reasoning need