Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ NTNU Openarrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
NTNU Open
Master thesis . 2025
Data sources: NTNU Open
addClaim

Contract Cheating with LLMs

Authors: Lynes, Kristian;

Contract Cheating with LLMs

Abstract

Denne avhandlingen undersøker i hvilken grad det er mulig å oppda- ge tekster skrevet av store språkmodeller. Vi bruker og sammenligner ulike deteksjonsmetoder for å vurdere om noen av dem har god nok ytelse til å kunne tas i bruk i akademiske institusjoner. Mer spesifikt undersøker vi hvorvidt deteksjonsmetodene gjenkjenner om en tekst er menneskelig-skrevet eller ikke, i tillegg til hvor godt metodene kjenner igjen en spesifikk person sin skrivestil. Metodene vi tester inkluderer menneskelige vurderinger, nettbaserte detektorer, og en algoritmisk me- tode som analyserer ulike skriftlige trekk. Våre resultater indikerer at den nettbaserte deteksjonsmetoden yter best blant de testede alternative- ne, med en nøyaktighet på omtrent 88% − 89.33%, avhengig av hvilken deteksjonsoppgave som utføres. Vi ser et potensial for en forbedring av den algoritmiske deteksjonsmetoden ved mer effektiv trening og en bredere analyse av skriftlige trekk. Likevel konkluderer vi med at ingen av metodene vi tester oppnår tilstrekkelig ytelse til å brukes i akademia i dag.

This thesis discusses the detectability of texts that are written using large language models. We employ and compare different detection methods to see if any of them perform well enough to be considered in academic institutions. We detect whether a text is human-written or not, and also how well the different detectors can recognize a specific human’s writing style. The tested detectors include human detectors, online detectors, and an algorithmic method that analyses textual features. Our findings indicate that the online detector has the best overall performance of the tested detectors, with accuracies ranging between 88% − 89.33% based on the detection task at hand. We argue that there is potential for the algorithmic detection method to be more effective given better training and a broader feature analysis, but that neither of the tested detection methods currently has sufficient performance to be used in academic institutions today.

Country
Norway
  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Green