Expert insights on AI code detection and academic integrity
Stay ahead with expert analysis and practical guides
General
10 min
David Kim
·
1 week ago
Line diffing under-reports copied code and over-reports similar-looking code. Here's what token normalization and AST fingerprinting actually compare, where each one breaks, and how to wire both into a CI pipeline or an academic submission workflow.
General
12 min
Rachel Foster
·
1 week ago
Renaming variables and swapping a for loop for a while loop defeats simple text matching, but it rarely defeats structural comparison. This report walks through the obfuscation ladder, the algorithms that climb it, and the published detection rates behind the claims, including the cases where every engine still misses.
General
9 min
Priya Sharma
·
2 weeks ago
A single AI detection score is a ranking, not a verdict, and most of the damage we've seen comes from reading it as one. This is the five-step triage we settled on after two years of grading CS 1 and CS 2 cohorts of roughly 400 submissions, including the score bands, the script, and the two cases where the whole thing fell apart.
General
10 min
James Okafor
·
2 weeks ago
Twenty-three of 412 submissions in a data structures course shared a Dijkstra implementation that differed by fewer than three tokens, and the peer similarity engine flagged all 23 as a single cluster. Nobody had copied anybody. Here is the mechanism behind convergent AI output, the identifiers and AST evidence that separated it from real collusion, and the three-pass workflow we used to keep the scores from contaminating each other.
General
11 min
James Okafor
·
2 weeks ago
Cross-language code plagiarism detection compares normalized structure rather than raw text, which works when a translation was mechanical and fails when the student rewrote the algorithm. Here is what survives a Java-to-Python translation, what the token and IR approaches actually see, and how to run the check across a whole cohort without drowning in false positives.
General
11 min
Priya Sharma
·
2 weeks ago
A public research university ran AI code detection as part of its grading workflow for a full academic year: eleven assignments, three courses, 4,118 submissions. The interesting number isn't the 3.8% that ended in a finding. It's the roughly two flagged files that got cleared for every one that held up, and what the department changed because of it.
General
13 min
Alex Petrov
·
2 weeks ago
A Java submission and a Python submission looked nothing alike, but they were the same algorithm translated line by line. This is the story of how cross-language code plagiarism detection actually works, where it catches translated code, and where it still fails.
General
9 min
Priya Sharma
·
2 weeks ago
Similarity scores are ranking signals, not verdicts. I'll walk through the distributions, thresholds, and triage rules I use when reviewing code similarity reports for 400-student courses, plus where AI-generated code fits in the same queue.
General
10 min
Alex Petrov
·
2 weeks ago
As a bootcamp instructor, I've graded hundreds of take-home coding challenges. The AI-resistant ones share a pattern: they ask for process artifacts, not just final code. Here's how to design assignments that hold up.
General
8 min
Rachel Foster
·
3 weeks ago
A University of Missouri-St. Louis instructor spent two decades watching code similarity tools change shape. The gaps between each generation of detection explain why instructors now run peer, web, and AI checks side by side.
General
6 min
Alex Petrov
·
3 weeks ago
A hands-on retrospective on how code similarity detection grew from naive line diffs to tokenization, ASTs, and fingerprinting. Follow a step-by-step Python prototype and a production workflow with Codequiry to catch refactored plagiarism in CS courses.
General
8 min
Dr. Sarah Chen
·
3 weeks ago
A CS professor traces how AI-generated code detection grew out of MOSS-era token fingerprints, code stylometry, and a broken similarity assumption. The piece explains how modern detectors work, where they still stumble, and why stacked peer, web, and AI signals make the most defensible academic workflow.