LLM difficulty predictions for fraction items are less stable across repeated runs than for the overall item set¶
Claim · All claims
Evidence · 1 study · associationalr2·q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person · recheck by 2028: tool and model accuracy changes quickly
Subclaims¶
Reading the evidence codes
q is the design tier, i the impact, n the sample, and the kind with its r
the rigour judged within that kind: what each code means.
This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.
q2 i? Fraction items produced higher flip rates (run-to-run inconsistency in difficulty predictions) than the overall item set, with Gemini showing almost 91% inconsistency across repetitions. → La Hadi 2025
Evidence¶
La Hadi 2025¶
La Hadi, A., Alibasa, M. J., Chen, G., & Asyhari, A. T. (2025). The “Easy Trap”: Why LLMs Underestimate Misconception-Driven Difficulty. https://arxiv.org/abs/2607.26067 doi:10.48550/arxiv.2607.26067
q2 · i? · associational · r2
Flip-rate analysis across five runs (Figure 3) shows "fraction items produced higher rates of inconsistent difficulty predictions than the overall item set"; the article suggests LLM judgments on misconception-driven problems are also less stable across repeated evaluations.
"Especially for Gemini, which shows almost 91% inconsistency during repetition, which means 10 out of 11 fractions get different difficulty estimation more than 3 times."
Discussion¶
Related Claims¶
- LLMs systematically underestimate fraction items whose difficulty is driven by misconceptions (the \"Easy Trap\") — related
- Within-judge variance of the repeated LLM-judge rubric is negligible, but cross-model-family agreement remains untested — related
- AI grading of the exam is highly stable across five independent runs at the total-score level (ICC(A,1) = 0.967), with lower but still strong cell-level stability (ICC(A,1) = 0.836) — related
- Smaller local LLMs show a minor position bias toward option ordering and partially guessing-like error variation, but accuracy remains relatively stable across permutations and repeated sampling runs — related
- Human annotators judge later-round CPIG items as improved in complexity but not in difficulty, with modest rater agreement — related
- Exploratory analysis found item difficulty differed across sources, driven by the Google Gemini versus NLE contrast — related
- LLM-generated difficulty ratings show only moderate rank alignment with empirical student performance on basic arithmetic items — related
- Pooling Repeated LLM Grades Does Not Improve Agreement on Short Answers — related
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
llm-fraction-predictions-run-to-run-instability |
title |
LLM difficulty predictions for fraction items are less stable across repeated runs than for the overall item set |
description |
LLM difficulty predictions for fraction items are less stable across repeated runs than for the overall item set |
status |
review |
evidence_strength |
moderate |
newest_evidence |
2025 |
recheck_by |
2028 |
generated |
by process:wiki-ingest on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — la-hadi-2025 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.