Expert insights on AI code detection and academic integrity
Stay ahead with expert analysis and practical guides
General
8 min
James Okafor
·
1 month ago
A head-to-head evaluation of AI code detection across ChatGPT-4o, GitHub Copilot, Claude 3.5 Sonnet, and Gemini 1.5 Pro. One pattern kept surfacing: text-only detectors miss refactored LLM code, while structural and multi-signal checks hold up.
General
7 min
Rachel Foster
·
1 month ago
An engineering leader audit of 1,284 remote take-home coding submissions found 31.2% flagged as likely AI-generated at high confidence. Manual review confirmed 279 of 401 high-confidence flags, with a 4.5% false positive rate among high-confidence flags. Here is what the data showed and how hiring managers should handle AI detection scores.
General
11 min
Alex Petrov
·
1 month ago
A 400-student Python course adopted Codequiry to check submissions for plagiarism and AI generation. The instructor found that 18% of assignments contained copy-pasted code from Chegg, and 12% showed strong signs of LLM authorship — a pattern that peer-only checks had missed entirely.
General
11 min
Dr. Sarah Chen
·
1 month ago
Over 1,200 student submissions from a large public university’s introductory Python course were analyzed with Codequiry’s similarity engine and its AI code detector. The results show how traditional plagiarism tools miss a growing fraction of unauthorized work—and why layering AI detection changes what instructors actually see.
General
7 min
Dr. Sarah Chen
·
1 month ago
We cut similarity rates from 43% to 7% in a Data Structures course not by policing harder but by rewriting the assignments themselves. Here's what worked, what broke, and where detection tools like Codequiry still earn their keep.
General
11 min
Alex Petrov
·
1 month ago
When Riverside University’s CS department ran an AI detector across 300 CS2 assignments alongside MOSS, they discovered a new layer of academic integrity challenges. The case study reveals what they found, how they calibrated thresholds, and why combining AI detection with source-code fingerprinting changed their grading workflow.
General
11 min
James Okafor
·
1 month ago
When students rename variables, extract methods, and reorder statements to hide copied code, which detection algorithm actually holds up? A controlled experiment pits winnowing, token-based matching, and AST structural hashing against a ladder of refactoring transformations — and reveals why single-technique checkers miss the cases that academic-integrity panels care about most.
General
11 min
Priya Sharma
·
1 month ago
What started as a textual diff in Unix is now a high-stakes algorithmic arms race. This article traces the thirty-year evolution of source code plagiarism detection—from simple token matching and AST comparison to fingerprinting that survives variable renaming, and finally to the fresh challenge of AI-generated code. We examine the real detection rates, the tools that led each era, and where Codequiry fits as the first hybrid platform to unify peer, web, and AI checks in a single reporting workflow.
General
14 min
Priya Sharma
·
1 month ago
When State University of Plains' CS department first faced GitHub Copilot-generated homework in 2022, their decades-old MOSS pipeline was useless. This retrospective traces their journey from manual suspicion to a layered detection stack that caught 31% of submissions as AI-generated last semester — and the hard lessons learned about false positives along the way.
General
11 min
Alex Petrov
·
1 month ago
A copied snippet might survive a human code review after a quick variable rename and loop inversion. A similarity checker that understands ASTs won’t be fooled. This guide walks through wiring Codequiry’s API into your Git pre‑commit workflow, step by step, so every commit is scanned for non‑original code before it hits the branch.
General
7 min
Rachel Foster
·
2 months ago
When 800 students enroll in an introductory Python course, detecting AI-generated code by hand is impossible. UMass Amherst integrated an AI code detector trained on student-level patterns alongside traditional similarity checks—and uncovered a 14% AI flag rate, a 2% false positive rate, and a sustainable workflow that kept TAs focused on teaching rather than policing.
General
11 min
James Okafor
·
2 months ago
Perplexity-based detectors aren’t magic — they measure how surprising a sequence of code tokens would be to a model trained on real human code. This report breaks open the inner math, real false-positive rates from Stanford and Edinburgh benchmarks, and why the strongest detectors stack statistical signals with AST fingerprinting and web-source checks.