On tightly specified functions, LLMs converged on a few exact AST forms (1–14 per 1,000 files) while an open-ended function remained diverse, and source was associated with exact form for all five functions¶
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? Source and exact form were associated for all five functions (Pearson's χ2, BH-adjusted q < .001), with Cramér's V ranging from 0.30 for loopy_madness to 0.96 for is_square and sum_string. → Runlong Ye 2026
Evidence¶
Runlong Ye 2026¶
Runlong Ye, Jing Fan, Angela Zavaleta Bernuy, Oscar Karnalim, Paul Denny, Juho Leinonen, and Michael Liut. (2026). Correctness, Convergence, and AI-Generated Code Detection: A Longitudinal Study of Student and Large Language Model Code in Introductory Programming. https://arxiv.org/abs/2610.00863 doi:10.48550/arxiv.2610.00863
q2 · i? · associational · r2
Pearson's chi-square tests of independence on source-by-form contingency tables from full eligible-file counts, with Benjamini–Hochberg false-discovery control within the five-function family. Cramér's V ranged from 0.30 to 0.96.
"Source and exact form were associated for all five functions (Pearson's 𝜒 2, BH-adjusted 𝑞<. 001), with Cramér's 𝑉 ranging from 0.30 for loopy_madness to 0.96 for is_square and sum_string."
Discussion¶
Related Claims¶
- Student solutions became less varied on four of five selected functions by 2025, with exact-form declines of 18.4–51.1%, while the open-ended loopy_madness did not decline — related
- Solutions from different LLM providers converged broadly on most CS1 labs, with cross-LLM MOSS match rates of about 88% pooled across labs but only 13.18–22.11% on Lab 10 — related
- LLMs across five models generate diverse erroneous Java code, with mean AST edit distances varying by model and prompting strategy — related
- Frontier LLMs produced near-perfect solutions to CS1 labs, with mean hidden-instructor-test scores of 98.93–99.99% across model–year combinations — related
- Higher-struggling-level problems elicit more diverse but less student-like LLM-generated errors — related
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
exact-ast-form-convergence-constrained-functions |
title |
On tightly specified functions, LLMs converged on a few exact AST forms (1–14 per 1,000 files) while an open-ended function remained diverse, and source was associated with exact form for all five functions |
description |
On tightly specified functions, LLMs converged on a few exact AST forms (1–14 per 1,000 files) while an open-ended function remained diverse, and source was associated with exact form for all five functions |
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 — runlong-ye-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.