finding
active
finding:default-fallback-of-10-6-triggers-for-fewer-than-1-in-100-prompts-on-averageDefault fallback of 10⁻⁶ triggers for fewer than 1 in 100 prompts on average
Occurs when fewer than two α values yield nonzero compliance.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Predicting Rare LLM Failures with 30× Fewer Rolloutscitesmentions
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.
- Shows that explicit labels without latent steering fail to generalize to situational cues
- Demonstrates steering is not equivalent to prompting with the contrastive prompts.
- Base failure mode.
- Case study illustrating procedural-execution-layer failure in harness adherence
- Weakens overall setup by showing some prompts can lower type hints, but does not invalidate core steering result.
- Adversarially-chosen prompts induce denial-of-service attacks consuming 10× more compute than benign promptsfinding0.730Background finding illustrating the severity of uncontrolled reasoning slowdowns.
- Base corrections: 72.9% Failed Fix, 14.7% Fixed; retractions: 69.4% Correct Retract, 30.0% Incorrect Retractfinding0.729Label distributions.
- Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance