Persona Sampling Hypothesis

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

• Simplicity Bias Hypothesishypothesis0.736
  Deep networks are biased toward finding simple fits to data, and this bias increases with model size, driving convergence
• Reward Hypothesisconcept0.731
  The claim in RL that any goal can be expressed as maximizing the expected cumulative sum of a scalar reward signal.
• Persona Constructionconcept0.727
  The process of building a coherent model persona from character archetypes and traits during training
• Capacity Hypothesishypothesis0.720
  Bigger models are more likely to converge to a shared representation than smaller models because they can better approximate the global optimum
• 4-19 principal components explain 70% of variance in role persona space across the three models (Gemma 4, Qwen 8, Llama 19)finding0.719
  Demonstrates that persona space is low-dimensional
• Universality Hypothesisconcept0.716
  The hypothesis that analogous features and circuits reliably form across different neural network models and tasks
• Autoregressive Samplingmethod0.714
  The mechanism by which LLMs generate text: drawing a token from the next-token distribution and appending it to context repeatedly
• What dimensions of persona are not captured by our extracted role vectors, and how complete is the current persona space mapping?question0.712
  Limitation question motivating future work on persona elicitation strategies