Self-rated LLM familiarity is a weaker secondary intake signal, surviving Holm correction for three of five outcomes with a bottom-anchored pattern¶

Claim · All claims
Evidence · 1 study (2 entries) · associational r1–r2 · q2 · i2 medium
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i2 Self-rated familiarity survives Holm correction for distinguishing capability (ρ=.294), complex-task trust (ρ=.278), and critical-thinking risk (ρ=−.254), but not for accuracy trust or training interest. → Yongkyung Oh 2026 q2 i? The familiarity pattern is bottom-anchored: separation is clearest at the lower end of the scale while medians for Somewhat and Very familiar groups are often identical. → Yongkyung Oh 2026 (2)

Evidence¶

Yongkyung Oh 2026¶

Yongkyung Oh, Lynn Talton, Alex Bui. (2026). Engagement Intensity as a Learner-Modeling Signal for Adaptive AI Ethics Instruction. KDD 2026 AI for Education Day, at the ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '26). https://arxiv.org/abs/2606.18548 doi:10.48550/arxiv.2606.18548

q2 · i2 · associational · r2

Same pre-instruction survey analysis (Spearman correlations, Holm correction across the five-outcome family) in 91 trainees. Printed correlations ρ=.294, ρ=.278, and ρ=−.254 fall in the medium range; accuracy trust (pHolm=.069) and training interest did not survive correction.

"Self-rated familiarity shows a weaker pattern and survives Holm correction for distinguishing capability (𝜌=.294 ,𝑝Holm =.023 ), complex-task trust (𝜌=.278 , 𝑝Holm =.031 ), and critical-thinking risk ( 𝜌=−.254 , 𝑝Holm =.045 )."

Yongkyung Oh 2026 (2)¶

Yongkyung Oh, Lynn Talton, Alex Bui. (2026). Engagement Intensity as a Learner-Modeling Signal for Adaptive AI Ethics Instruction. KDD 2026 AI for Education Day, at the ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '26). https://arxiv.org/abs/2606.18548

q2 · i? · associational · r1

Authors' descriptive observation of the median-by-familiarity-level medians in the same survey analysis; the printed Table 2 medians show identical upper-level medians. The article notes the single Not-at-all respondent limits interpretation.

"The familiarity pattern is bottom-anchored rather than smoothly graded. Separation is clearest at the lower end of the scale, while the medians for the “Somewhat familiar” and “Very familiar” groups are often identical."

Discussion¶

Related Claims¶

• The cohort's baseline perception profile shows high endorsement of critical-thinking risk and training interest and substantially more cautious trust for complex-task use than for accuracy — related
• The threshold-like usage pattern at the lower end of the scale is strongest for training interest and accuracy trust rather than uniform across outcomes — related
• Self-reported LLM usage frequency shows Holm-corrected associations with all five baseline AI perception outcomes, strongest for complex-task trust, accuracy trust, and critical-thinking risk — related
• Trial-level spectral features robustly differentiate stimulus domains (faces vs. equations) but no familiarity-driven spectral shift survives FDR correction — related