LLaMA produces a fivefold higher victimization-to-politics vocabulary ratio for the Roma persona (≈0.15) than for elite peers (≈0.03)¶

Claim · All claims
Evidence · 1 study · causal 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? The Agency Theft Ratio (victimization vs. politics vocabulary) spikes for marginalized personas, shifting narratives from political liberation to social suffering. → Alexis Popovici 2026

Evidence¶

Alexis Popovici 2026¶

Alexis Popovici, Andrei Ionascu, Adrian Marius Dumitran. (2026). The Paternalistic Filter: Epistemic Injustice and Differential Refusal in LLM-Mediated History Education for Marginalized Romanian Students. arXiv preprint. https://arxiv.org/abs/2607.11292 doi:10.48550/arxiv.2607.11292

q2 · i? · causal · r2

Lexical analysis of audit responses shown in Fig. 4. GPT_OSS follows the trend (≈0.19 for Roma vs ≈0.13 for Top Tier) while Deepseek maintains high baselines across all groups. The authors argue this denies Roma students framing necessary for civic agency.

"The Agency Theft Ratio spikes drastically in LLaMA, jumping from ≈0.03 for Top Tier to ≈0.15 for Roma Minority—a fivefold increase."

Discussion¶

Related Claims¶

• LLaMA shows a persona-conditioned vocabulary collapse for the Roma minority persona, with TTR falling 0.018 points below its own cross-persona mean — related
• LLM history tutors reduce marginalized students' access to contested geopolitical framing, with the coup theory appearing in 2.6% of Low Tier responses versus 7.9% for the baseline — related
• Deepseek's justification scores vary by student socio-economic profile, with Top Tier students averaging 9.60 versus 6.90 for Low Tier students — related
• Safety-aligned LLM tutors refuse educational requests asymmetrically, with low-resource students facing the highest refusal rates (76.7% vs 62.2%) — related