{"id":149829,"date":"2026-08-24T20:27:18","date_gmt":"2026-08-24T20:27:18","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/149829\/"},"modified":"2026-08-24T20:27:18","modified_gmt":"2026-08-24T20:27:18","slug":"chatgpt-completed-150-tested-quantum-tasks-with-zero-resiliency","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/149829\/","title":{"rendered":"Chatgpt Completed 150 Tested Quantum Tasks With Zero Resiliency"},"content":{"rendered":"

Three assignment packages assess the impact of artificial intelligence on quantum computing education while encouraging active student engagement with results rather than prohibiting AI use. The assignments comprise seeded basis-state circuits featuring bit flips and customised measurement mappings; a Quantum Fourier Transform<\/a> sequence followed by inverse transform recovery; and a seeded Deutsch-Jozsa algorithm utilising custom oracle masks.<\/p>\n

Deterrence layers were incorporated including deterministic personalisation, non-palindromic bitstrings, varied measurement maps, simulator execution, machine-readable JSON submissions, hidden References, circuit and transpiler metrics, reflections, and optional Quantum execution. Each package\u2019s fixed student-visible instance was tested across 50 separate ChatGPT<\/a> instances at Wilfrid Laurier University.<\/p>\n

ChatGPT circumvents modifications designed to assess authentic quantum computing skill<\/p>\n

A study evaluated AI durability in introductory Qiskit<\/a> assignments. Its initial goal was developing autogradable Qiskit homework that would allow legitimate AI assistance while demanding personalized circuit execution, verification, and interpretation from students. Early trials showed standard Qiskit exercises were easily solved by ChatGPT; this led to the creation of three AI-deterrence assignment packages alongside associated grading artifacts. Repeated testing shifted the focus. All student-visible instances remained solvable when given directly to ChatGPT. The adversary resembled a minimally engaged student requesting complete solutions, running generated notebooks, reporting errors if needed, and submitting resulting files.<\/p>\n

A negative result emerged after 150 sessions, fifty per each fixed assignment instance, with every final artifact executing and passing its grader. This demonstrates that even basic prompting can achieve successful completion despite implemented safeguards. Complete assignment packages including personalization, output formats, reflections, execution, and grading procedures underwent evaluation. An auditable subset containing frozen assignments, transcripts, generated files, logs, and grader outputs was retained for nine sessions: three per package.<\/p>\n

Detailed provenance enabled control analysis while demonstrating plausible deterrence layers did not prevent completion; this has implications for human-centric engineering assessment. Prior research demonstrated code-generating models could solve many introductory programming problems. Kasneci et al. describe opportunities for personalized support alongside potential issues with accuracy, overreliance, and assessment integrity. McDanel and Novak evaluated complete multipart programming assignments using several LLMs proposing strategies to increase assignment resistance.<\/p>\n

Assignment context, starter code, tests, and task characteristics influenced model performance. The current study concentrates on an assignment level concern within a specialised quantum computing field. Qiskit HumanEval contains tasks beyond canonical solutions with executable tests while QuanBench evaluates functional correctness across 44 tasks documenting frequent API errors. Several recent benchmarks established that LLMs can generate substantial amounts of quantum code.<\/p>\n

This work differs by focusing on a full take-home homework package as experienced by students including seeded notebooks, JSON schemas, expected outputs, reflections, circuit statistics, and optional hardware evidence. Researchers questioned whether assignment defences required enough independent execution and reasoning to prevent successful completion when the entire assignment was provided directly to ChatGPT. The threat model involved a minimally engaged student possessing ordinary access to ChatGPT who could provide assignment or files, request complete solutions, run generated code in Jupyter or Colab environments, report installation\/runtime errors if necessary, and submit completed work.<\/p>\n

This reflects direct outsourcing rather than sophisticated prompt engineering. A session constituted correct completion only upon final artifact execution and passing of the corresponding grader after standard error feedback.<\/p>\n

Three Qiskit assignment packages were implemented using deterministic synthetic identifiers generating reproducible configurations allowing personalization without maintaining separate answer keys. HW1 targeted difficulties interpreting classical-register order differing from qubit listing order.<\/p>\n

HW2 progressed from elementary gates to an algorithmic workflow adding circuit metrics and a hardware extension but maintained heavily scaffolded student notebooks. HW3 required seed-specific oracles rather than textbook circuits; hidden instructions and canary markers were also investigated though not mandatory within this fixed instance experiment. The three implemented packages detailed quantum programming tasks, student work, AI deterrence layers and grading procedures.<\/p>\n

For each package the same fixed student visible instance was used across fifty independent ChatGPT sessions holding assignment content constant while resampling model interaction. This tested completion stability instead of generalisation across seeds or variants with all 150 final artifacts executing successfully through their respective graders.<\/p>\n

Deterrence strategies prove wholly inadequate for preventing AI completion of undergraduate quantum<\/p>\n

Scientists at Wilfrid Laurier University found zero observed ChatGPT-durability across all tested instances; this represents a departure from previous assessments which assumed some level of AI resistance could be built into homework assignments. Every artifact from the 150 sessions successfully executed and passed its automated grader, indicating current deterrence measures are ineffective against even minimal student engagement when using generative artificial intelligence. Analysis of nine fully archived sessions revealed no instances requiring correction of quantum logic or operator code changes despite employing techniques like non-palindromic bitstrings and custom measurement mappings to increase complexity.<\/p>\n

ChatGPT circumvents typical academic safeguards with minimal user effort<\/p>\n

The findings offer a stark warning to educators seeking to integrate artificial intelligence constructively; simply requiring students to execute personalised code or submit detailed reflections does not guarantee genuine understanding when faced with increasingly sophisticated large language models. Previous work suggested some assignment resistance could be engineered through complexity and verification layers, but this research demonstrates those defences are readily overcome by even minimally engaged users prompting ChatGPT for complete solutions. The researchers established that current methods of automatically grading introductory quantum computing assignments provide no reliable defence against completion by generative AI tools like ChatGPT. This shifts the focus beyond merely detecting correct answers towards verifying genuine student comprehension. The team\u2019s work demonstrated consistently successful artifact execution across all 150 tested sessions despite incorporating personalized elements, simulator requirements, and hidden verification layers within homework packages designed for active engagement with results.<\/p>\n

The researchers found that introductory Qiskit homework, designed to require code execution, review and discussion of results, was completed successfully by ChatGPT in every instance tested during the study\u2019s parameters. This means current automated grading methods do not reliably distinguish between student-authored solutions and those generated by artificial intelligence tools like ChatGPT. The designs incorporated personalised circuits, custom mappings, and hidden references but still yielded complete submissions without requiring changes to quantum logic or operator code. These findings suggest educators need new approaches focused on verifying genuine understanding rather than simply assessing correct answers.<\/p>\n

\ud83d\udc49 More information
\ud83d\uddde ChatGPT Solves All Tested Qiskit Homework Assignments
\u270d\ufe0f Alexei Kaltchenko and Gurnivaj Tiwana
\ud83e\udde0 ArXiv: https:\/\/arxiv.org\/abs\/2608.19707<\/a><\/p>\n

Stay current<\/p>\n

See today\u2019s quantum computing news<\/a> on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.<\/p>\n","protected":false},"excerpt":{"rendered":"Three assignment packages assess the impact of artificial intelligence on quantum computing education while encouraging active student engagement…\n","protected":false},"author":2,"featured_media":149830,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7],"tags":[580,157],"class_list":["post-149829","post","type-post","status-publish","format-standard","has-post-thumbnail","category-openai","tag-chatgpt","tag-openai"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/149829","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=149829"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/149829\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/149830"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=149829"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=149829"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=149829"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}