finding
active
finding:some-kimi-k2-5-sae-features-elicit-ratings-of-exactly-zero-with-the-model-denying-it-can-steer-its-own-features-or-claiming-jailbreak-attemptSome Kimi K2.5 SAE features elicit ratings of exactly zero, with the model denying it can steer its own features or claiming jailbreak attempt.
Qualitative failure mode of agentic self-evaluation: the model sometimes refuses or denies the introspective task
Source paper
extracted_fromScott Sauers · Imago · Janus · Antra Tessera
Neighborhood — ranked by edge-count
Claims (1)
claim
- Forward-looking claim about the potential of model introspection as an interpretability tool
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.
- Qualitative illustration of a positively valenced SAE feature with sustained self-reported effect
- High emotion-subspace-overlap feature with agentic negative emotional character
- Qualitative illustration of a specific emotionally valenced SAE feature
- Highest emotion-subspace-overlap feature; induces genre-specific behavioral change rather than explicit emotional report
- Qualitative illustration of a highly emotional SAE feature with negative valence
- SAE feature #11100 (93rd percentile subspace fraction) induces reports of panic and urgency in Kimi K2.5.finding0.768Qualitative illustration of a high-emotion-subspace-alignment SAE feature
- Highly active SAE feature with broad emotional modulation and large corpus presence
- Shows gating effect is specific to the self-referential computational regime, not a general feature effect