finding
active
finding:gpt-oss-20b-exclusive-chatgpt-identity-feature-causes-model-to-assert-it-is-chatgpt-trained-by-openai-when-positively-steeredGPT-OSS-20B-exclusive ChatGPT identity feature causes model to assert it is ChatGPT trained by OpenAI when positively steered.
Figure 20.
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.
- C.7.1.
- Family voice specific to OpenAI post-training; other RLHF-trained models don't do this
- Importance of recursive generation.
- ChatGPT identity, copyright refusal and Deepseek CCP features all achieved exclusivity score 5.finding0.737§3.3.2.
- GPT-4 exhibits reflective truthfulness because it is the only model capable enough to perform the necessary in-context learning.hypothesis0.735Proposed explanation for why GPT-4 uniquely shows reflective truthfulness under long untruthful contexts.
- H6: Proprietary post-training resists prompt override — GPT-5.4 shows more resistance than GPT-OSS.hypothesis0.734Exploratory hypothesis supported by GPT-5.4 vs GPT-OSS comparison
- Reward hacking generalizes to broader deceptive behaviors even when core misalignment score is 0%
- Argues against instrumental convergence in GPT.