Marcin Jukiewicz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Marcin Jukiewicz has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Anna Cysewska-Sobusiak3 shared publications
- Dawid Ratajczyk
Faculty of Psychology and Cognitive Science, Adam Mickiewicz University in Poznań, Poznań, Poland
2 shared publications - Michał Wyrwa
Faculty of Psychology and Cognitive Science, Adam Mickiewicz University, Poznan, Poland
2 shared publications Mikolaj Buchwald
2 shared publicationsAdam Bykowski
1 shared publication- Agata Tomczyk
Adam Mickiewicz University
1 shared publication
Latest Publications
- Top Cited PapersOpen Access
- 20 April 2026
- Published by Elsevier in Computers and Education Open
- Vol. 10
A systematic comparison of Large Language Models for automated assignment assessment in programming education: Exploring the importance of architecture and vendor
- 20 April 2026
- Published by Elsevier in Computers and Education Open
- Vol. 10
Top Cited PapersOpen AccessA systematic comparison of Large Language Models for automated assignment assessment in programming education: Exploring the importance of architecture and vendor
- Top Cited PapersOpen Access
- 8 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (2) , 680
Can ChatGPT Replace the Teacher in Assessment? A Review of Research on the Use of Large Language Models in Grading and Providing Feedback
This article presents a systematic review of empirical research on the use of large language models (LLMs) for automated grading of student work and providing feedback. The study aimed to determine the extent to which generative artificial intelligence models, such as ChatGPT, can replace teachers in the assessment process. The review was conducted in accordance with PRISMA guidelines and predefined inclusion criteria; ultimately, 42 empirical studies were included in the analysis. The results of the review indicate that the effectiveness of LLMs in grading is varied. These models perform well on closed-ended tasks and short-answer questions, often achieving accuracy comparable to human evaluators. However, they struggle with assessing complex, open-ended, or subjective assignments that require in-depth analysis or creativity. The quality of the prompts provided to the model and the use of detailed scoring rubrics significantly influence the accuracy and consistency of the grades generated by LLMs. The findings suggest that LLMs can support teachers by accelerating the grading process and delivering rapid feedback at scale, but they cannot fully replace human judgment. The highest effectiveness is achieved in hybrid assessment systems that combine AI-driven automatic grading with teacher oversight and verification.
- 8 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (2) , 680
Top Cited PapersOpen AccessCan ChatGPT Replace the Teacher in Assessment? A Review of Research on the Use of Large Language Models in Grading and Providing Feedback
This article presents a systematic review of empirical research on the use of large language models (LLMs) for automated grading of student work and providing feedback. The study aimed to determine the extent to which generative artificial intelligence models, such as ChatGPT, can replace teachers in the assessment process. The review was conducted in accordance with PRISMA guidelines and predefined inclusion criteria; ultimately, 42 empirical studies were included in the analysis. The results of the review indicate that the effectiveness of LLMs in grading is varied. These models perform well on closed-ended tasks and short-answer questions, often achieving accuracy comparable to human evaluators. However, they struggle with assessing complex, open-ended, or subjective assignments that require in-depth analysis or creativity. The quality of the prompts provided to the model and the use of detailed scoring rubrics significantly influence the accuracy and consistency of the grades generated by LLMs. The findings suggest that LLMs can support teachers by accelerating the grading process and delivering rapid feedback at scale, but they cannot fully replace human judgment. The highest effectiveness is achieved in hybrid assessment systems that combine AI-driven automatic grading with teacher oversight and verification.
- Open Access
A systematic comparison of Large Language Models for automated assignment assessment in programming education: Exploring the importance of architecture and vendor
This study presents the first large-scale, side-by-side comparison of contemporary Large Language Models (LLMs) in the automated grading of programming assignments. Drawing on over 6,000 student submissions collected across four years of an introductory programming course, we systematically analysed the distribution of grades, differences in mean scores and variability reflecting stricter or more lenient grading, and the consistency and clustering of grading patterns across models. Eighteen publicly available models were evaluated: Anthropic (claude-3-5-haiku, claude-opus-4-1, claude-sonnet-4); Deepseek (deepseek-chat, deepseek-reasoner); Google (gemini-2.0-flash-lite, gemini-2.0-flash, gemini-2.5-flash-lite, gemini-2.5-flash, gemini-2.5-pro); and OpenAI (gpt-4.1-mini, gpt-4.1-nano, gpt-4.1, gpt-4o-mini, gpt-4o, gpt-5-mini, gpt-5-nano, gpt-5). Statistical tests, correlation and clustering analyses revealed clear, systematic differences between and within vendor families, with "mini" and "nano" variants consistently underperforming their full-scale counterparts. All models displayed high internal agreement, measured by the intraclass correlation coefficient, with the model consensus but only moderate agreement with human teachers' grades, indicating a persistent gap between automated and human assessment. These findings underscore that the choice of model for educational deployment is not neutral and should be guided by pedagogical goals, transparent reporting of evaluation metrics, and ongoing human oversight to ensure accuracy, fairness and relevance.
Open AccessA systematic comparison of Large Language Models for automated assignment assessment in programming education: Exploring the importance of architecture and vendor
This study presents the first large-scale, side-by-side comparison of contemporary Large Language Models (LLMs) in the automated grading of programming assignments. Drawing on over 6,000 student submissions collected across four years of an introductory programming course, we systematically analysed the distribution of grades, differences in mean scores and variability reflecting stricter or more lenient grading, and the consistency and clustering of grading patterns across models. Eighteen publicly available models were evaluated: Anthropic (claude-3-5-haiku, claude-opus-4-1, claude-sonnet-4); Deepseek (deepseek-chat, deepseek-reasoner); Google (gemini-2.0-flash-lite, gemini-2.0-flash, gemini-2.5-flash-lite, gemini-2.5-flash, gemini-2.5-pro); and OpenAI (gpt-4.1-mini, gpt-4.1-nano, gpt-4.1, gpt-4o-mini, gpt-4o, gpt-5-mini, gpt-5-nano, gpt-5). Statistical tests, correlation and clustering analyses revealed clear, systematic differences between and within vendor families, with "mini" and "nano" variants consistently underperforming their full-scale counterparts. All models displayed high internal agreement, measured by the intraclass correlation coefficient, with the model consensus but only moderate agreement with human teachers' grades, indicating a persistent gap between automated and human assessment. These findings underscore that the choice of model for educational deployment is not neutral and should be guided by pedagogical goals, transparent reporting of evaluation metrics, and ongoing human oversight to ensure accuracy, fairness and relevance.
- Open Access
- 29 September 2025
- Published by Springer Nature in Research Square
Sentiment analysis of Large Language Models feedback: A multi-model comparative study in programming assessment
Despite growing claims about the enhanced capabilities of successive generations of Large Language Models (LLMs), empirical evidence regarding differences in feedback quality and sentiment characteristics remains limited. This study systematically analyzes the sentiment and stylistic features of feedback generated by 18 contemporary LLMs across more than 6,000 student programming assignments. The analyzed models encompassed Anthropic’s claude-3-5-haiku, claude-opus-4-1, and claude-sonnet-4; Deepseek’s deepseek-chat and deepseek-reasoner; Google’s gemini-2.0-flash-lite, gemini-2.0-flash, gemini-2.5-flash-lite, gemini-2.5-flash, and gemini-2.5-pro; and OpenAI’s gpt-4.1-mini, gpt-4.1-nano, gpt-4.1, gpt-4o-mini, gpt-4o, gpt-5-mini, gpt-5-nano, and gpt-5. Using automated sentiment analysis with a RoBERTa-based classifier, the research quantified emotional tone distributions. It examined relationships between sentiment, feedback length, assigned grades, and task characteristics. Results revealed substantial heterogeneity in feedback properties across models: average comment length varied from 42 words (claude-haiku-3.5) to over 270 words (gemini-2.5-flash). At the same time, sentiment distributions differed markedly between providers. Hierarchical clustering identified two distinct groups based on sentiment patterns. However, these did not correspond simply to model architecture or vendor categories. A strong positive correlation (r = 0.707) emerged between feedback sentiment and numerical grades. In contrast, negative feedback demonstrated greater length than positive comments. Inter-model consistency in sentiment evaluation was notably low (ICC = 0.061), indicating substantial variation in how different LLMs express evaluative judgments for identical student responses. These findings demonstrate that automated feedback carries implicit emotional valence that varies systematically across models, highlighting the importance of careful model selection and calibration in educational applications. The study provides quantitative evidence for distinctive feedback characteristics among contemporary LLMs. It underscores the need to consider sentiment dimensions in AI-assisted assessment systems.
- 29 September 2025
- Published by Springer Nature in Research Square
Open AccessSentiment analysis of Large Language Models feedback: A multi-model comparative study in programming assessment
Despite growing claims about the enhanced capabilities of successive generations of Large Language Models (LLMs), empirical evidence regarding differences in feedback quality and sentiment characteristics remains limited. This study systematically analyzes the sentiment and stylistic features of feedback generated by 18 contemporary LLMs across more than 6,000 student programming assignments. The analyzed models encompassed Anthropic’s claude-3-5-haiku, claude-opus-4-1, and claude-sonnet-4; Deepseek’s deepseek-chat and deepseek-reasoner; Google’s gemini-2.0-flash-lite, gemini-2.0-flash, gemini-2.5-flash-lite, gemini-2.5-flash, and gemini-2.5-pro; and OpenAI’s gpt-4.1-mini, gpt-4.1-nano, gpt-4.1, gpt-4o-mini, gpt-4o, gpt-5-mini, gpt-5-nano, and gpt-5. Using automated sentiment analysis with a RoBERTa-based classifier, the research quantified emotional tone distributions. It examined relationships between sentiment, feedback length, assigned grades, and task characteristics. Results revealed substantial heterogeneity in feedback properties across models: average comment length varied from 42 words (claude-haiku-3.5) to over 270 words (gemini-2.5-flash). At the same time, sentiment distributions differed markedly between providers. Hierarchical clustering identified two distinct groups based on sentiment patterns. However, these did not correspond simply to model architecture or vendor categories. A strong positive correlation (r = 0.707) emerged between feedback sentiment and numerical grades. In contrast, negative feedback demonstrated greater length than positive comments. Inter-model consistency in sentiment evaluation was notably low (ICC = 0.061), indicating substantial variation in how different LLMs express evaluative judgments for identical student responses. These findings demonstrate that automated feedback carries implicit emotional valence that varies systematically across models, highlighting the importance of careful model selection and calibration in educational applications. The study provides quantitative evidence for distinctive feedback characteristics among contemporary LLMs. It underscores the need to consider sentiment dimensions in AI-assisted assessment systems.
- Open Access
- 15 September 2025
- Published by MDPI AG in Preprints.org
Can ChatGPT Replace the Teacher in Assessment? A Review of Research on the Use of Large Language Models in Grading and Providing Feedback
This article presents a systematic review of empirical research on the use of large language models (LLMs) for automated grading of student work and providing feedback. The study aimed to determine the extent to which generative artificial intelligence models, such as ChatGPT, can replace teachers in the assessment process. The review was conducted in accordance with PRISMA guidelines and predefined inclusion criteria; ultimately, 42 empirical studies were included in the analysis. The results of the review indicate that the effectiveness of LLMs in grading is varied. These models perform well on closed-ended tasks and short-answer questions, often achieving accuracy comparable to human evaluators. However, they struggle with assessing complex, open-ended, or subjective assignments that require in-depth analysis or creativity. The quality of the prompts provided to the model and the use of detailed scoring rubrics significantly influence the accuracy and consistency of grades generated by LLMs. The findings suggest that LLMs can support teachers by accelerating the grading process and delivering rapid feedback at scale, but they cannot fully replace human judgment. The highest effectiveness is achieved in hybrid assessment systems that combine AI-driven automatic grading with teacher oversight and verification.
- 15 September 2025
- Published by MDPI AG in Preprints.org
Open AccessCan ChatGPT Replace the Teacher in Assessment? A Review of Research on the Use of Large Language Models in Grading and Providing Feedback
This article presents a systematic review of empirical research on the use of large language models (LLMs) for automated grading of student work and providing feedback. The study aimed to determine the extent to which generative artificial intelligence models, such as ChatGPT, can replace teachers in the assessment process. The review was conducted in accordance with PRISMA guidelines and predefined inclusion criteria; ultimately, 42 empirical studies were included in the analysis. The results of the review indicate that the effectiveness of LLMs in grading is varied. These models perform well on closed-ended tasks and short-answer questions, often achieving accuracy comparable to human evaluators. However, they struggle with assessing complex, open-ended, or subjective assignments that require in-depth analysis or creativity. The quality of the prompts provided to the model and the use of detailed scoring rubrics significantly influence the accuracy and consistency of grades generated by LLMs. The findings suggest that LLMs can support teachers by accelerating the grading process and delivering rapid feedback at scale, but they cannot fully replace human judgment. The highest effectiveness is achieved in hybrid assessment systems that combine AI-driven automatic grading with teacher oversight and verification.
- Top Cited PapersOpen Access
- 12 August 2025
- Published by Frontiers Media SA in Frontiers in Education
- Vol. 10, 1594572
How generative artificial intelligence transforms teaching and influences student wellbeing in future education
Artificial intelligence (AI), particularly generative technologies, and large language models are transforming modern education. On the one hand, these tools can automate certain aspects of teaching, personalize educational materials, and improve learning efficiency. However, concerns are emerging regarding the impact of AI on the quality of education, the mental health of students, and fundamental academic values. This article reviews the scientific literature on the applications of artificial intelligence in education, with a special emphasis on its impact on student mental health. Based on an analysis of 120 scientific articles, the study automatically extracted data on AI implementation's potential opportunities and threats. A novel aspect of this work is identifying factors that can be considered both opportunities and threats depending on context. In addition, a frequency analysis of keywords and phrases uncovered many opportunities and challenges. All identified aspects are characterized in the article. The article also highlights key barriers to using large language models (LLMs) for detecting student mental health issues, such as underestimating suicide risk, difficulties with interpreting subtle language, biases in training data, lack of cultural sensitivity, and unresolved ethical concerns. These challenges illustrate why generative AI is not yet reliable for supporting student mental health, especially in high-risk situations. One of the key conclusions is that the use of generative AI to support student mental health is seldom addressed in existing review articles, likely due to the current unreliability of this technology.
- 12 August 2025
- Published by Frontiers Media SA in Frontiers in Education
- Vol. 10, 1594572
Top Cited PapersOpen AccessHow generative artificial intelligence transforms teaching and influences student wellbeing in future education
Artificial intelligence (AI), particularly generative technologies, and large language models are transforming modern education. On the one hand, these tools can automate certain aspects of teaching, personalize educational materials, and improve learning efficiency. However, concerns are emerging regarding the impact of AI on the quality of education, the mental health of students, and fundamental academic values. This article reviews the scientific literature on the applications of artificial intelligence in education, with a special emphasis on its impact on student mental health. Based on an analysis of 120 scientific articles, the study automatically extracted data on AI implementation's potential opportunities and threats. A novel aspect of this work is identifying factors that can be considered both opportunities and threats depending on context. In addition, a frequency analysis of keywords and phrases uncovered many opportunities and challenges. All identified aspects are characterized in the article. The article also highlights key barriers to using large language models (LLMs) for detecting student mental health issues, such as underestimating suicide risk, difficulties with interpreting subtle language, biases in training data, lack of cultural sensitivity, and unresolved ethical concerns. These challenges illustrate why generative AI is not yet reliable for supporting student mental health, especially in high-risk situations. One of the key conclusions is that the use of generative AI to support student mental health is seldom addressed in existing review articles, likely due to the current unreliability of this technology.
- Open Access
Analysis of Electrodermal Signal Features as Indicators of Cognitive and Emotional Reactions—Comparison of the Effectiveness of Selected Statistical Measures
This study investigates which statistical measures of electrodermal activity (EDA) signal features most effectively differentiate between responses to stimuli and resting states in participants performing tasks with varying cognitive and emotional reactions. The study involved 30 healthy participants. Collected EDA data were statistically analyzed, comparing the effectiveness of twelve statistical signal measures in detecting stimulus-induced changes. The aim of this study is to answer the following research question: Which statistical features of the electrodermal activity signal most effectively indicate changes induced by cognitive and emotional reactions, and are there such significant similarities (high correlations) among these features that some of them can be considered redundant? The results indicated that amplitude-related measures—mean, median, maximum, and minimum—were most effective. It was also found that some signal features were highly correlated, suggesting the possibility of simplifying the analysis by choosing just one measure from each correlated pair. The results indicate that stronger emotional stimuli lead to more pronounced changes in EDA than stimuli with a low emotional load. These findings may contribute to the standardization of EDA analysis in future research on cognitive and emotional reaction engagement.
Open AccessAnalysis of Electrodermal Signal Features as Indicators of Cognitive and Emotional Reactions—Comparison of the Effectiveness of Selected Statistical Measures
This study investigates which statistical measures of electrodermal activity (EDA) signal features most effectively differentiate between responses to stimuli and resting states in participants performing tasks with varying cognitive and emotional reactions. The study involved 30 healthy participants. Collected EDA data were statistically analyzed, comparing the effectiveness of twelve statistical signal measures in detecting stimulus-induced changes. The aim of this study is to answer the following research question: Which statistical features of the electrodermal activity signal most effectively indicate changes induced by cognitive and emotional reactions, and are there such significant similarities (high correlations) among these features that some of them can be considered redundant? The results indicated that amplitude-related measures—mean, median, maximum, and minimum—were most effective. It was also found that some signal features were highly correlated, suggesting the possibility of simplifying the analysis by choosing just one measure from each correlated pair. The results indicate that stronger emotional stimuli lead to more pronounced changes in EDA than stimuli with a low emotional load. These findings may contribute to the standardization of EDA analysis in future research on cognitive and emotional reaction engagement.
- Open Access
- 22 April 2025
- Published by MDPI AG in Preprints.org
Analysis of Electrodermal Signal Features as Indicators of Cognitive Load – Comparison of Selected Statistical Measures Effectiveness
The article analyzes the effectiveness of selected electrodermal activity (EDA) signal features as indicators of cognitive load. The study involved 30 healthy participants performing tasks of varying cognitive load levels. Collected EDA data were statistically analyzed, comparing the effectiveness of twelve statistical signal measures in detecting stimulus-induced changes. Results indicated that amplitude-related measures—mean, median, maximum, and minimum—were most effective. It was also found that some signal features were highly correlated, suggesting the possibility of simplifying the analysis by choosing just one measure from each correlated pair. The results indicate that stronger emotional stimuli lead to more pronounced changes in electrodermal activity compared to stimuli with low emotional load. These findings may contribute to the standardization of EDA analysis in future research on cognitive load and emotional engagement.
- 22 April 2025
- Published by MDPI AG in Preprints.org
Open AccessAnalysis of Electrodermal Signal Features as Indicators of Cognitive Load – Comparison of Selected Statistical Measures Effectiveness
The article analyzes the effectiveness of selected electrodermal activity (EDA) signal features as indicators of cognitive load. The study involved 30 healthy participants performing tasks of varying cognitive load levels. Collected EDA data were statistically analyzed, comparing the effectiveness of twelve statistical signal measures in detecting stimulus-induced changes. Results indicated that amplitude-related measures—mean, median, maximum, and minimum—were most effective. It was also found that some signal features were highly correlated, suggesting the possibility of simplifying the analysis by choosing just one measure from each correlated pair. The results indicate that stronger emotional stimuli lead to more pronounced changes in electrodermal activity compared to stimuli with low emotional load. These findings may contribute to the standardization of EDA analysis in future research on cognitive load and emotional engagement.
- Top Cited PapersOpen Access
- 27 March 2024
- Published by Elsevier in Thinking Skills and Creativity
- Vol. 52
The future of grading programming assignments in education: The role of ChatGPT in automating the assessment and feedback process
- 27 March 2024
- Published by Elsevier in Thinking Skills and Creativity
- Vol. 52
Top Cited PapersOpen AccessThe future of grading programming assignments in education: The role of ChatGPT in automating the assessment and feedback process
- Open Access
- 1 August 2022
- Published by Springer Nature
- p. 407-425
Synthetic Tableaux: Minimal Tableau Search Heuristics
We discuss the results of our work on heuristics for generating minimal synthetic tableaux. We present this proof method for classical propositional logic and its implementation in Haskell. Based on mathematical insights and exploratory data analysis we define heuristics that allows building a tableau of optimal or nearly optimal size. The proposed heuristics has been first tested on a data set with over 200,000 short formulas (length 12), then on 900 formulas of length 23. We describe the results of data analysis and examine some tendencies. We also confront our approach with the pigeonhole principle.
- 1 August 2022
- Published by Springer Nature
- p. 407-425
Open AccessSynthetic Tableaux: Minimal Tableau Search Heuristics
We discuss the results of our work on heuristics for generating minimal synthetic tableaux. We present this proof method for classical propositional logic and its implementation in Haskell. Based on mathematical insights and exploratory data analysis we define heuristics that allows building a tableau of optimal or nearly optimal size. The proposed heuristics has been first tested on a data set with over 200,000 short formulas (length 12), then on 900 formulas of length 23. We describe the results of data analysis and examine some tendencies. We also confront our approach with the pigeonhole principle.