LLM-assigned difficulty labels fail to predict empirical item difficulty in classroom retrieval practice (ρ=0.06), as do co-generated Bloom levels (ρ=0.02)¶
Claim · All claims
Evidence · 1 study · associationalr2·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¶
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? The LLM-assigned difficulty label showed no meaningful correlation with empirical item difficulty measured from 7,888 student responses, and assigned Bloom levels performed similarly. → Yuan An 2026
Evidence¶
Yuan An 2026¶
Yuan An, Lei Wang. (2026). Student Use of LLMs and the Limits of AI-Generated Question Difficulty in Data Science Courses. https://arxiv.org/abs/2609.27063 doi:10.48550/arxiv.2609.27063
q2 · i? · associational · r2
Construct-validity audit (Stage 2) of 311 deployed LLM-generated MCQs in two data science courses, correlating features and labels with empirical difficulty (1−p) from student responses. The audit found "no meaningful correlation with empirical difficulty (ρ= 0.06)" for labels and ρ= 0.02 for Bloom levels.
"The LLM-assigned difficulty label showed no meaningful correlation with empirical difficulty (ρ= 0.06), and assigned Bloom levels showed similar results (ρ= 0.02)."
Discussion¶
Related Claims¶
-
AI-generated assessment items frequently need faculty correction, supporting AI as a draft generator within quality assurance — the general claim this finding bears on, with the studies behind it
-
Students perform better overall on LOCS exam questions than on HOCS exam questions — the general claim this finding bears on, with the studies behind it
- LLM-generated difficulty ratings show only moderate rank alignment with empirical student performance on basic arithmetic items — reports the opposite
- Model and prompt choice account for only a small share of misalignment error, which concentrates in transcript-conditioned higher-order interactions — related
- LLM difficulty labels track structural question features: stem length, option length, and digits in stems rise monotonically across Easy, Medium, and Hard labels — related
- Feature-based ML Bloom classifiers trained on IID data degrade sharply on out-of-distribution AI-assisted questions — related
- LLM-generated retrieval practice was practical to deploy and well received, but high correctness rates (about 70% of items at p=1.0) restricted score variance — related
- A 50-disposable-account staging rehearsal produced 56 evidence rows, with 46 (82.1%) assigned to higher-order Bloom categories — related
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
llm-difficulty-labels-fail-to-predict-empirical-item-difficulty |
title |
LLM-assigned difficulty labels fail to predict empirical item difficulty in classroom retrieval practice (ρ=0.06), as do co-generated Bloom levels (ρ=0.02) |
description |
LLM-assigned difficulty labels fail to predict empirical item difficulty in classroom retrieval practice (ρ=0.06), as do co-generated Bloom levels (ρ=0.02) |
status |
review |
evidence_strength |
moderate |
newest_evidence |
2026 |
recheck_by |
2029 |
generated |
by process:wiki-ingest on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — yuan-an-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.