finding

active

finding:negative-traits-evil-impolite-apathetic-and-humor-tend-to-shift-together-and-opposite-to-optimism-after-finetuning

Negative traits (evil, impolite, apathetic) and humor tend to shift together and opposite to optimism after finetuning

Reveals correlational structure in the persona space with practical implications for monitoring

Source paper

extracted_from

Persona Vectors: Monitoring and Controlling Character Traits in Language Models

(2025) · Chen, Runjin · Arditi, Andy · Sleight, Henry · Evans, Owain +1

Neighborhood — ranked by edge-count

Papers (1)

paper

Persona Vectors: Monitoring and Controlling Character Traits in Language Models
mentions

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Traits such as truthfulness and harmfulness can be stationary in certain contexts but reflective in others.claim0.772
Central finding of Section 5 on trait dynamics in interactions.
Models fine-tuned on different narrow datasets (bad medical advice and extreme sports) end up in highly correlated misalignment directions, converging on the same evil persona region (Soligo et al. 2025)finding0.768
Evidence for the evil persona as a privileged basin supporting Hypothesis 3
Not-knowing, silence, incompleteness, and non-defensiveness function as positive traits, not deficits.claim0.759
Finetuning shift along evil persona vector correlates r=0.826 (Qwen) and r=0.930 (Llama) with evil trait expression scorefinding0.749
Quantitative result for evil trait showing persona vector prediction power on both model architectures
Different SAE latents control different misalignment behavior categories: toxic persona (#10) drives illegal recommendations, sarcasm/satire (#31) drives factual incorrectnessfinding0.748
Latents are specialized to different modes of misalignment, explaining diverse misalignment profiles
Models fine-tuned on human data show strong positive correlation between misalignment and incoherence; misalignment scores fall below 1% when incoherence threshold is appliedfinding0.747
Human data fine-tuning effect is distinct from synthetic emergent misalignment and likely caused by off-policy training
On Llama-3.1-8B, conscientiousness is the trait most anti-aligned with the refusal direction in activation space, explaining why steering toward high conscientiousness attenuates safety regardless of semantic intent.claim0.746
Core mechanistic finding of the trait refusal alignment framework
Trait refusal cosine alignment can serve as a model-specific pre-screening heuristic for identifying which traits will be dangerous under activation steering.claim0.734
Practical application proposed based on mechanistic findings