kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
GPT-4 as a TA: A Study on Automated Feedback for Introductory Programming Assignments
KTH, School of Electrical Engineering and Computer Science (EECS).
2024 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesisAlternative title
GPT-4 som en lärarassistent : En studie om automatiserad feedback för inledande programmeringsuppgifter (Swedish)
Abstract [en]

The increasing enrollment rates in computer science programs have posed significant challenges, particularly in providing timely and effective feedback to students in introductory programming courses. This research investigates the potential of OpenAI’s GPT-4, a Large Language Model (LLM), to automate feedback generation for student assignments, thereby alleviating some of the burdens on teaching staff. By evaluating the model’s effectiveness across three metrics—accuracy, correctness, and lucidity—this study aims to determine the feasibility of integrating GPT-4 into educational tools. Student submissions from the course DD1338: Algorithms and Data Structures were selected and artificially injected with common errors. GPT-4 was then prompted to generate feedback based on these submissions using two different prompting methods: one with assignment instructions included and one without. The study found that GPT-4 can identify errors and provide correct advice effectively, with minor differences between the prompting methods. However, the model exhibited a propensity for hallucinations, especially when task instructions were included, thereby impacting the lucidity of the feedback. The results indicate that while GPT-4 shows promise in automating feedback for programming assignments, its tendency to generate hallucinations necessitates human review.

Abstract [sv]

De ökande inskrivningssiffrorna i datavetenskapsprogram har inneburit betydande utmaningar, särskilt när det gäller att tillhandahålla snabb och effektiv feedback till studenter i introduktionskurser i programmering. Denna forskning undersöker potentialen hos OpenAI:s GPT-4, en stor språkmodell (LLM), för att automatisera feedbackgenerering för studentuppgifter och därmed lindra en del av bördan på undervisningspersonalen. Genom att utvärdera modellens effektivitet över tre parametrar—noggrannhet, korrekthet och klarhet—syftar denna studie till att avgöra om det är genomförbart att integrera GPT-4 i utbildningsverktyg. Studentinlämningar från kursen DD1338: Algoritmer och Datastrukturer valdes ut och försågs artificiellt med vanliga fel. GPT-4 ombads sedan att generera feedback baserat på dessa inlämningar med hjälp av två olika metoder: en med uppgiftsinstruktioner inkluderade och en utan. Studien fann att GPT-4 effektivt kan identifiera fel och ge korrekt rådgivning, med mindre skillnader mellan metoderna. Modellen visade dock en benägenhet för hallucinationer, särskilt när uppgiftsinstruktioner inkluderades, vilket påverkade feedbackens klarhet. Resultaten indikerar att även om GPT-4 visar lovande möjligheter att automatisera feedback för programmeringsuppgifter, kräver dess tendens att generera hallucinationer mänsklig granskning.

Place, publisher, year, edition, pages
2024. , p. 25
Series
TRITA-EECS-EX ; 2024:380
Keywords [en]
Computer Science Education, Large Language Models (LLMs), GPT-4
Keywords [sv]
Datavetenskap utbildning, Stor språkmodell, GPT-4
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-351172OAI: oai:DiVA.org:kth-351172DiVA, id: diva2:1886397
Supervisors
Examiners
Available from: 2024-08-30 Created: 2024-08-01 Last updated: 2024-08-30Bibliographically approved

Open Access in DiVA

fulltext(395 kB)112 downloads
File information
File name FULLTEXT01.pdfFile size 395 kBChecksum SHA-512
ab54e9a5ac6e21891b49495b2197861871fdca8053fbee1c82082be0a918216cdaec580be7ef2487dd3469a62cacc4f04b15079366fa4d1860974ac4fa61e415
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Computer and Information Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 112 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 383 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf