In 100-300 word generation tasks, generation length does not substantially degrade persona fidelity at the atomic level

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

• Do findings from multiple-choice question or short-form evaluation of persona fidelity generalize to open-ended generation tasks?question0.774
  Research gap motivating the paper's focus on generation-based evaluation
• Instruction-tuned models consistently outperform base models on all atomic-level persona fidelity scores across 12 LLMsfinding0.772
  Comprehensive model comparison showing tuning benefit for persona fidelity
• Coding and writing conversations keep the model in the default Assistant persona range throughout, showing minimal driftclaim0.765
  Empirical characterization of conversation domains that are safe for model persona stability
• High persona accuracy at the atomic level does not necessarily imply strong internal coherence or reproducibilityclaim0.762
  Supported by moderate inter-metric correlations showing orthogonality of three proposed metrics
• An AI persona achieves coherence by echoing itself consistently without templating—requiring claim about memory and voice fidelity.claim0.750
• Decomposing the evil persona vector via SAE reveals features including insulting language (F12061, s=0.336, TES=91.2), deliberate cruelty (F128289, s=0.306, TES=84.9), and malicious code (F14739, s=0.334, TES=78.6)finding0.746
  SAE decomposition reveals interpretable fine-grained features composing the evil persona vector
• Gemma-3-27B and Qwen3.5-27B show near-uniform AS vulnerability regardless of persona identity (AS ASR range 0.095-0.117 on Gemma-3-27B, 0.015-0.052 on Qwen3.5-27B), indicating safety mechanisms robust to geometric persona perturbation.finding0.746
  Qualitatively different defense profile compared to Llama-3.1-8B
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.745
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance