Human-shadow prompting generates broad psychometric variance while AI-assistant responses are uniform¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Human-shadow conditioning produces considerable variance in MSES, AMAS, and MSEAQ score distributions, whereas AI-assistant responses show negligible variance with several models giving identical scores. → Esposito 2026

Evidence¶

Esposito 2026¶

Esposito, N., Tricarico, A., Porzio, L., Aghazadeh Ardebili, A., & Stella, M. (2026). Math Education Digital Shadows for Investigating Learning with GenAI: Mathematics Performance, Anxiety, and Confidence in LLMs. arXiv preprint. https://arxiv.org/abs/2604.27618 doi:10.48550/arxiv.2604.27618

q2 · i? · design · r2

Task 2 analysis comparing summed psychometric score distributions between human-shadow and AI-assistant runs across 14 LLMs. The article reports that "the variance among AI-assistant responses is negligible" and that several models, including Anita 24B (Uncensored) and Phi-4 (Reasoning+), provided identical responses and scores under AI-assistant instructions.

"The human-shadow distributions display considerable variance across all three scales, indicating that human-shadow conditioning successfully generated a broad spectrum of simulated educational cognition and varied psychological attitudes towards mathematics."

Discussion¶

Related Claims¶

• LLMs simulating survey respondents under AI-assistant instructions report higher mathematics self-efficacy and lower anxiety than under human-shadow conditioning — related
• Interview-containing prompts increased response diversity for Claude and GPT, while Gemini diversified with demographic-only prompts — related
• LLM-generated BREQ responses capture overall item-mean patterns but are more extreme and less variable than human responses — a broader claim this one bears on
• Priming human shadows as math lovers or haters shifts LLM emotional tone about mathematics — related
• Demographic and psychometric response prompting produces higher-variance, heavier-tailed originality distributions than a no-context baseline, though still less variable than humans — related
• Within-judge variance of the repeated LLM-judge rubric is negligible, but cross-model-family agreement remains untested — related
• LLM digital shadows reproduce the human pattern of positive science perceptions anchored in negative mathematics associations — related