Generative AI and ChatGPT in School Children’s Education: Evidence from a School Lesson

May combine two claims. The quote only partly states this claim.

Pupils who perceived not to have learned at all were typically not interested in the subject, did not enjoy the lesson, and evaluated the content and exercises as very difficult or difficult (t(DF=16)=44.452, p<0.001).

“Those who perceived not to have learned at all were typically not interested about the subject, did not enjoy the lesson, and evaluated the content and exercises as”

• AssumptionRests on an assumption: Pupils' self-perception of learning accurately reflects their actual learning.

May combine two claims.

Perceived difficulty of studied content was easy for 8.2%, quite easy for 32.7%, neither for 33.6%, quite difficult for 20.0%, and very difficult for 5.5%.

“Subsequently, the pupils provided their estimations regarding the difficulty level of the studied content, yielding a large variation among the pupils: 8.2% found it easy, 32.7% quite easy, 33.6% neither easy nor difficult, 20.0% quite difficult and 5.5% very difficult.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Word counts for original and AI-adapted texts were: 4th grade original 117, basic 137, intermediate 123, advanced 125; 5th grade original 129, basic 119, intermediate 119, advanced 144; 6th grade original 162, basic 220, intermediate 195, advanced 164.

“The original text for the 4th grade had an average of 117 words.The AI-generated adapted text had 137 words at the basic level, 123 words at the intermediate level, and 125 words at the advanced level.”

• Also stated in: § 4.2. Conducting the Content Part of the Test Lesson · plain

  “For the 5th grade, the original content averaged 129 words.The AI-generated adapted text had 119 words at the basic level, 119 words at the intermediate level, and 144 words at the advanced level.”

• Also stated in: § 4.2. Conducting the Content Part of the Test Lesson · plain

  “For the 6th grade, the original content contained 162 words.The AI-generated adapted content had 220 words at the basic level, 195 words at the intermediate level, and 164 words at the advanced level.”

May combine two claims.

Pupils' enjoyment of learning the topic using the laptop and generative AI-modified learning material varied: 40.0% liked it very much, 26.4% much, 23.6% somewhat, 5.5% little, and 4.5% not at all.

“The responses varied among the pupils: 40.0%mentioned they liked it very much, 26.4% much, 23.6% somewhat, 5.5% little, and 4.5% not at all.”

May combine two claims.

Figure helpfulness varied by knowledge group: basic 37.9%, 27.6%, 24.1%, 6.9%, 3.4%; intermediate 19.4%, 30.6%, 30.6%, 11.1%, 8.3%; advanced 44.4%, 33.3%, 11.1%, 11.1%, 0.0%.

“In the basic knowledge group, 37.9% found the images very helpful, 27.6% quite helpful, 24.1% unsure, 6.9% not really helpful and 3.4% not at all helpful.”

• Also stated in: § 4.3. Conducting the Self-Evaluation Part of the Test Lesson · plain

  “In the intermediate knowledge group, 19.4% found the images very helpful, 30.6% quite helpful, 30.6% unsure, 11.1% not really helpful and 8.3% not helpful at all.”

• Also stated in: § 4.3. Conducting the Self-Evaluation Part of the Test Lesson · plain

  “In the advanced knowledge group, 44.4% evaluated the images as very helpful, 33.3% quite helpful, 11.1% unsure, 11.1% not really helpful and 0.0% not helpful at all.”

May combine two claims.

Ease of device use varied by grade: 4th grade 24.0%, 24.0%, 30.0%, 16.0%, 6.0%; 5th grade 37.5%, 21.9%, 31.3%, 6.3%, 3.1%; 6th grade 28.6%, 14.3%, 25.0%, 14.3%, 17.9%.

“At the 4th grade, these shares were 24.0%, 24.0%, 30.0%, 16.0% and 6.0%; at the 5th grade, 37.5%, 21.9%, 31.3%, 6.3% and 3.1%; and at the 6th grade, 28.6%, 14.3%, 25.0%, 14.3% and 17.9%.”

May combine two claims.

Among pupils who performed weakest in the test (29 pupils with less than 50% correct answers), 48.2% perceived themselves to have learned very much or much, 34.5% to have learned somewhat, and 17.2% little or very little.

“Of those who performed weakest in the test (29 pupils having less than 50% correct answers), 48.2% perceived themselves to have learned very much or much, 34.5% to have learned somewhat and 17.2% little or very little.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

For the second set of learning material, average reading time was 91.56 s for 4th grade, 110.69 s for 5th grade, and 98.82 s for 6th grade.

“As regards the second set of learning material, i.e., the text and one figure, on average, the 4th-grade pupils spent 91.56 s (with a median of 78.50 s) going through their learning material, the 5th-grade pupils took an average of 110.69 s (with a median of 106.50 s), and the 6th-grade pupils took an average of 98.82 s (with a median of 83.00 s).”

May combine two claims.

Mathematics was the favorite subject for 46.4% of pupils and the least favorite for 17.3%.

“Mathematics was the favorite for nearly half of the students and this did not vary substantially between classes: for 46.4% of pupils it was the favorite and for 17.3% pupils it was the least favorite (50.0% vs. 12.0% for the 4th grade, 40.6% vs. 12.5% for the 5th grade and 46.4% vs. 32.1% for the 6th grade).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Spanish was the favorite subject for 23.6% of pupils and the least favorite for 21.8%.

“Other school topics received the following shares: mother tongue (Spanish) was the favorite for 23.6% of all pupils vs. the least favorite for 21.8% of all pupils: 46.0% vs. 6.0%for the 4th grade, 3.1% vs. 46.9%for the 5th grade and 7.1% vs. 21.4% for the 6th grade).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Pupils who had the most knowledge about the topic were also the most interested in it.

“However, those having the most knowledge about the topic were also the most interested in it.”

Probably not a substantial claim: may not pass S4 (based).

Pupils' performance and their knowledge about the topic being learned are continuously monitored.

“In addition, pupils' performance and their knowledge about the topic being learned are continuously monitored.”

May combine two claims.

Interest in the study topic was much for 15.5% of pupils, some for 15.5%, fair for 29.0%, little for 15.5%, and no interest for 24.5%.

“The responses displayed significant variation, with the following breakdown: much (15.5%),some (15.5%),fair (29.0%), little (15.5%), and no interest (24.5%).”

May combine two claims.

Self-perceived knowledge about the topic was excellent for 9.1% of pupils, good for 20.9%, medium for 31.8%, satisfactory for 18.2%, and poor for 20.0%.

“More specifically, 9.1% of all pupils claimed to have excellent knowledge about the topic, while the remaining had good (20.9%),medium (31.8%), satisfactory (18.2%), and poor (20.0%) knowledge.”

May combine two claims.

It was possible to receive immediate feedback of pupils' learning during the lesson and adjust the learning material in real time, taking into account the pupil's performance.

“In addition, it was possible to receive immediate feedback of pupils' learning during the lesson and adjust the learning material in real time, taking into account the pupil's performance.”

• AssumptionRests on an assumption: The generative AI-modified learning material actually matched each pupil's knowledge and interest levels as intended.

May combine two claims.

On average, pupils answered 4.66 out of 8 learning material questions correctly, a share of 58.2% correct answers and 41.8% incorrect ones.

“On average, the pupils could answer 4.66 out of the 8 learning material questions correctly, resulting in a share of 58.2% of correct answers and 41.8% incorrect ones.”

• AssumptionRests on an assumption: The generative AI model can accurately determine the correctness of pupils' answers.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Among all participants, 57.3% identified as boys, 40.9% as girls, and 1.8% chose not to answer.

“Among all the participants, 57.3% identified themselves as boys, 40.9% identified as girls, and 1.8% chose not to answer this question.”

May combine two claims.

Social Sciences was the favorite subject for 14.5% of pupils and the least favorite for 26.4%.

“Finally, Social Sciences was on average less preferred among the pupils, as 14.5% said that it was their favorite vs. 26.4% who perceived it as their least favorite: 0.0% vs. 40.0%for the 4th grade, 37.5% vs. 18.8% for the 5th grade and 14.3% vs. 10.7% for the 6th grade.”

May combine two claims.

English was the favorite subject for 15.5% of pupils and the least favorite for 34.5%.

“The foreign language (English) was perceived as the most favorite by 15.5% of pupils vs. the least favorite by 34.5%: 4.0% vs. 42.0%for the 4th grade, 18.8% vs. 21.9% for the 5th grade and 32.1% vs. 35.7%for the 6th grade.”

May combine two claims.

Self-perceived knowledge varied by grade: 4th grade 8.0%, 26.0%, 38.0%, 24.0%, 4.0%; 5th grade 18.8%, 21.9%, 46.9%, 6.2%, 6.2%; 6th grade 0.0%, 10.7%, 3.6%, 21.4%, 64.3%.

“Among the pupils in the 4th grade, these percentages were 8.0%, 26.0%, 38.0%, 24.0%, and 4.0%.Among the pupils in the 5th grade, the breakdown was as follows: 18.8%, 21.9%, 46.9%, 6.2%, and 6.2%.Among the pupils in the 6th grade, the self-perception grades were substantially lower: 0.0%, 10.7%, 3.6%, 21.4%, and 64.3% (Table 1).”

May combine two claims.

Typical characteristics of pupils who perceived themselves to have learned a lot were excellent grades overall, initial placement into the advanced level based on Social Sciences rating (t(DF=16)=28.625, p=0.027), and interest in the subject (t(DF=12)=39.322, p<0.001).

“Typical characteristics of those who perceived themselves to have learned a lot were pupils with excellent grades overall, who were initially placed into the advanced level in the beginning (determined by their Social Sciences rating) (t (DF = 16) = 28.625,p = 0.027), and who showed interested in the subject (t (DF = 12) = 39.322,p < 0.001).”

• AssumptionRests on an assumption: Pupils' self-reported school grades accurately reflect their knowledge and skills for grouping.
• AssumptionRests on an assumption: Pupils' self-perception of learning accurately reflects their actual learning.

Probably not a substantial claim: may not pass S3 (load-bearing).

The adapted learning material included associated exercises and visual material that aided in learning.

“This adapted learning material was not limited to the text alone; it also included associated exercises and visual material that aided in learning.”

• AssumptionRests on an assumption: The generative AI-modified learning material actually matched each pupil's knowledge and interest levels as intended.

May combine two claims.

Enjoyment percentages by grade were: 4th grade 32.0% very much, 36.0% much, 20.0% somewhat, 10.0% little, 2.0% not at all; 5th grade 59.4%, 15.6%, 21.9%, 0.0%, 3.1%; and 6th grade 32.1%, 21.4%, 32.1%, 3.6%, 10.7%.

“These percentages were, at the 4th, grade, 32.0%, 36.0%,20.0%, 10.0%, and 2.0%; at the 5th grade, 59.4%, 15.6%, 21.9%, 0.0%, and 3.1%; and at the 6th grade, 32.1%, 21.4%, 32.1%, 3.6%, and 10.7%.”

May combine two claims.

Figures were rated very helpful by 26.4%, quite helpful by 30.0%, unsure by 27.3%, not really helpful by 10.0%, and not at all helpful by 6.4%.

“Among all the pupils, the majority found that the figures supported their learning: 26.4% answered that the figures were very helpful, 30.0%quite helpful, 27.3% were unsure, 10.0% not really helpful, and 6.4% not at all helpful.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

For open-ended questions in the second part, average answering time was 90.60 s for 4th grade, 181.50 s for 5th grade, and 90.29 s for 6th grade.

“Regarding the two open-ended questions, the 4th-grade pupils took an average of 90.60 s (with a median of 74.50 s) to provide their responses, the 5th-grade pupils took an average of 181.50 s (with a median of 150.50 s), and the 6th-grade pupils took an average of 90.29 s (with a median of 84 s).”

May combine two claims.

Total time to answer all test-lesson questions averaged 355.55 s; 4th grade 292.76 s, 5th grade 452.72 s, and 6th grade 275.90 s.

“The total time taken to answer all test-lesson questions averaged about six minutes, which is equivalent to 355.55 s, with the 5th-grade pupils taking notably longer.The answering time for the 4th grade was 292.76 s, for the 5th grade 452.72 s, and for the 6th grade 275.90 s.”

May combine two claims.

On average, girls enjoyed the test lesson slightly more than boys, and enjoyment varied more among boys.

“As regards the test lesson, on average, girls enjoyed such learning slightly more than boys and enjoyment varied more among boys (Table 5).”

May combine two claims.

Interest in the study topic varied by grade: 4th grade 10.0%, 12.0%, 36.0%, 16.0%, 26.0%; 5th grade 31.3%, 21.8%, 34.4%, 12.5%, 0.0%; 6th grade 7.1%, 14.3%, 10.7%, 17.9%, 50.0%.

“For the 4th grade, the breakdown shares were 10.0%, 12.0%, 36.0%,16.0%, and 26.0%.For the 5th grade, they were 31.3%, 21.8%, 34.4%, 12.5%, and 0.0%.For the 6th grade, these were 7.1%, 14.3%, 10.7%, 17.9%, and 50.0%(Table 1).”

May combine two claims.

Social Sciences marks varied between classes: 4th grade excellent 18.0%, good 28.0%, medium 40.0%, satisfactory 12.0%, poor 2.0%; 5th grade 25.0%, 43.8%, 28.1%, 3.1%, 0.0%; 6th grade 3.6%, 46.4%, 32.1%, 14.3%, 3.6%.

“In the 4th grade, the marks were excellent (18.0%), good (28.0%),medium (40.0%), satisfactory (12.0%), and poor (2.0%).In the 5th grade, the shares were 25.0%, 43.8%, 28.1%, 3.1%, and 0.0%.In the 6th grade they were 3.6%, 46.4%, 32.1%, 14.3%, and 3.6%.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The general grade of all four topics had a mean of 3.7 and median of 3.6, with little variation among classes: 4th grade mean 3.8, 5th grade mean 3.8, and 6th grade mean 3.5.

“The general grade of all four topics (mean 3.7; median 3.6) showed little variation among the different classes.It was 3.8 (median 3.8) in the 4th grade, 3.8 (median 3.9) in the 5th grade, and 3.5 (median 3.5) in the 6th grade (Table 1).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

For the first set of learning material, average reading time was 92.48 s for 4th grade, 115.38 s for 5th grade, and 152.54 s for 6th grade.

“As regards the first set of learning material, which included text and one figure, on average, 4th-grade pupils spent 92.48 s (with a median of approximately 78.00 s) going through their first learning material with text and a figure, the 5th-grade pupils 115.38 s (with a median of approximately 81.00 s) and the 6th-grade pupils 152.54 s (with a median of approximately 122.50 s).”

May combine two claims.

Figure helpfulness varied by grade: 4th grade 18.0%, 34.0%, 38.0%, 6.0%, 4.0%; 5th grade 46.9%, 28.1%, 9.4%, 12.5%, 3.1%; 6th grade 17.9%, 25.0%, 28.6%, 14.3%, 14.3%.

“At the 4th grade, the shares were 18.0%, 34.0%, 38.0%, 6.0% and 4.0%, at the 5th grade, 46.9%, 28.1%, 9.4%, 12.5% and 3.1%, and at the 6th grade, 17.9%, 25.0%, 28.6%, 14.3% and 14.3%.”

May combine two claims.

Ease of device use was rated very easy by 29.1%, quite easy by 20.9%, neither easy nor difficult by 29.1%, quite difficult by 12.7%, and very difficult by 8.2%.

“Among the pupils, the responses were rather evenly divided, though they leant toward easiness of the use of the device to learn the topic, as only about one out of five pupils found it difficult to a certain extent: 29.1% found it very easy, 20.9% quite easy, 29.1% neither easy nor difficult, 12.7% quite difficult and 8.2% very difficult.”

May combine two claims.

Both ChatGPT-3.5 and ChatGPT-4 demonstrated the ability to accurately translate content using the same prompt, yielding outputs that closely resembled human translation.

“Both of these generative AI models demonstrated the ability to accurately translate the content by employing the same prompt (as a written text input to ChatGPT to generate a specific useful written response), yielding outputs that closely resembled human translation.”

May combine two claims.

For multiple-choice questions in the second part, average answering time was 38.86 s for 4th grade, 61.09 s for 5th grade, and 40.93 s for 6th grade.

“In contrast, for the two multiple-choice questions, the 4th-grade pupils took an average of 38.86 s (with a median of 25.50 s) to answer, the 5th-grade pupils took an average of 61.09 s (with a median of 49.50 s), and the 6th-grade pupils took an average of 40.93 s (with a median of 32.00 s) (Table 2).”

May combine two claims.

Self-evaluation of learning showed 23.6% learned very much, 33.6% much, 29.1% somewhat, 9.1% little, and 4.5% not at all; 57.2% learned much or very much.

“The overall responses indicated that only one out of seven pupils mentioned that they had learned little or nothing, while the clear majority (57.2%) mentioned that they had learned much or very much.Among all pupils, 23.6% indicated that they had learned very much, 33.6% learned much, 29.1% learned somewhat, 9.1% learned little, and 4.5% learned not at all.”

• AssumptionRests on an assumption: Pupils' self-perception of learning accurately reflects their actual learning.

May combine two claims.

For multiple-choice questions in the first part, average answering time was 29.92 s for 4th grade, 82.38 s for 5th grade, and 41.00 s for 6th grade.

“In contrast, for the two multiple-choice questions in the first part of the lesson, the 4th-grade pupils took an average of 29.92 s (with a median of 23.00 s) to answer, the 5th-grade pupils took 82.38 s (with a median of 64.50 s), and 6th-grade pupils took 41.00 s (with a median of 37.50 s).”

May combine two claims.

Among pupils who performed best in the test (15 pupils with more than 80% correct answers), 60.0% perceived that they had learned very much or much, 33.0% perceived that they had learned somewhat, and 6.7% little or very little.

“Among pupils who performed the best in the test (15 pupils who had a performance score of more than 80% correct answers), 60.0% perceived that they had learned very much or much, 33.0% perceived that they had learned somewhat, and 6.7% little or very little.”

May combine two claims.

For open-ended questions in the first part, average answering time was 133.38 s for 4th grade, 198.56 s for 5th grade, and 103.68 s for 6th grade.

“Regarding the answering of two open-ended questions about the learning material, the 4th-grade pupils took an average of 133.38 s (with a median of 107.50 s) to provide their responses, 5th-grade pupils took 198.56 s (with a median of 179.50 s), and 6th-grade pupils took 103.68 s (with a median of 66.50 s) (Table 2).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

There was a key difference in the format of answers produced by the NLP models: ChatGPT-3.5 chose the placement of author and link source, while ChatGPT-4 adhered to the original text format.

“Notably, there was a key difference in the format of the answers produced by the developed NLP models.ChatGPT-3.5 exhibited the freedom to choose the placement of the author and link source, while ChatGPT-4 displayed consistency in adhering to the original text format provided.”

May combine two claims.

Gender differences in answering times were: 4th grade girls 156.16 s vs boys 138.61 s; 5th grade girls 268.61 s vs boys 246.76 s; 6th grade girls 125.88 s vs boys 155.16 s.

“There was a gender difference in answering times: 4th-grade girls used 156.16 s, while boys used 138.61 s; 5th-grade girls used 268.61 s, while boys used 246.76 s; and 6th-grade girls used 125.88 s, while boys used 155.16 s (Table 2).”

May combine two claims.

Overall, two out of three pupils enjoyed learning the generative AI modified material with the laptop, and only one out of ten liked it only a little or not at all.

“Overall, two out of three pupils enjoyed learning the generative AI modified material with the laptop, and only one out of ten liked it only a little or not at all.”

May combine two claims.

The time varied between genders and among pupils from the 4th to 6th grade; however, a more detailed analysis requires larger data.

“The time varied between genders and among pupils from the 4th to 6th grade; however, a more detailed analysis requires larger data.”

May combine two claims.

Differences were observed among pupils regarding their reading time of the learning material as well as regarding their answering time to open-ended and multiple-choice questions.

“Differences were observed among pupils regarding their reading time of the learning material as well as regarding their answering time to open-ended and multiple-choice questions.”

May combine two claims.

A test in which primary school pupils practiced their vocabulary found that the application can offer customized feedback, analyze individual needs, and assess results.

“For example, the application can offer customized feedback, analyze individual needs, and assess results.This was found out in a test in which primary school pupils practiced their vocabulary [9].”

• cites [9] Kasneci, 2023, ChatGPT for good? On opportunities and challenges of large language models for education, Learn Indif. Differ. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

There is promising potential of generative AI use in school education, supporting pupils' motivated learning and skills development.

“There is a promising potential of generative AI use in school education, supporting pupils' motivated learning and skills development.”

• AssumptionRests on an assumption: Pupils' enjoyment of generative AI-modified material supports their motivated learning and skills development.

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

At this early phase of development, a sensible suggestion was to neither outright reject nor entirely rely on generative AI in school education.

“At this early phase of development, the study concluded that a sensible suggestion was to neither outright reject nor entirely rely on generative AI in school education [16].”

• cites [16] Baidoo-Anu, D., and Ansah, L. (2023, July 15). Education in the Era of Generative Artificial Intelligence (AI): Understanding the Potential Benefits of ChatGPT in Promoting Teaching and Learning. Available online: https://ssrn.com/abstract=4337484. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

In the school context, ChatGPT uses algorithms to generate new text and responses similar to what a teacher or a pupil might write.

“In the end, in the school context, ChatGPT uses algorithms to generate new text and responses similar to what a teacher or a pupil might write [28].”

• cites [28] Brown, 2020, Language models are few-shot learners, Adv. Neural Inf. Process. Syst.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Properly implemented generative AI allows teachers to better focus on education opportunities and challenges, while pupils can receive guidance and support suitable to their knowledge backgrounds and levels.

“Properly implemented generative AI allows teachers to better focus on education opportunities and challenges, while pupils can receive guidance and support suitable to their knowledge backgrounds and levels.”

May combine two claims.

Generative AI empowers detailed performance tracking, enabling the delivery of personalized content that aligns with each student's individual capabilities, helping prevent students from experiencing anxiety or boredom during classes.

“For instance, generative AI empowers detailed performance tracking, enabling the delivery of personalized content that aligns with each student's individual capabilities.This personalized approach helps prevent students from experiencing anxiety or boredom during classes, which are factors that can contribute to dissatisfaction and even dropout rates within the educational system [9].”

• cites [9] Kasneci, 2023, ChatGPT for good? On opportunities and challenges of large language models for education, Learn Indif. Differ. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

LLMs have been trained on massive amounts of text data so that they can generate human-like text, answer questions, and complete other language-related tasks with high accuracy.

“LLMs have been trained on massive amounts of text data so that they can generate human-like text, answer questions, and complete other language-related tasks with high accuracy [9].”

• cites [9] Kasneci, 2023, ChatGPT for good? On opportunities and challenges of large language models for education, Learn Indif. Differ. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

ChatGPT allows interaction with generative AI in a chat environment, simulating almost-real conversational situations between humans.

“The most commonly known format is ChatGPT, which allows the interaction with this generative AI in a chat environment, simulating almost-real conversational situations between humans [23][24][25].”

• cites [23][24][25]Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Generative AI requires globally significant investment in hardware and software, as well as ongoing maintenance and support.

“Generative AI requires globally significant investment in hardware and software, as well as ongoing maintenance and support [21].”

• cites [21] Chang, 2023, The rise of generative artificial intelligence (AI) language models-challenges and opportunities for geographical and environmental education, Int. Res. Geogr.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Generative AI can play a significant role by serving as a mediator between educational institutions, teachers and pupils, providing educational support anytime and anywhere.

“This article argues that generative AI can play a significant role by serving as a mediator between educational institutions, teachers and pupils, providing educational support anytime and anywhere.”

Probably not a substantial claim: may not pass S3 (load-bearing).

According to a comprehensive review, LLMs can generate questions and prompts that encourage pupils to think critically about what they are reading and writing.

“According to a comprehensive review, LLMs can generate questions and prompts that encourage pupils to think critically about what they are reading and writing [9].”

• cites [9] Kasneci, 2023, ChatGPT for good? On opportunities and challenges of large language models for education, Learn Indif. Differ. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

By leveraging LLMs and generative AI technologies, teachers can alleviate some burdens associated with lesson planning, personalized content creation, differentiation and personalized instruction, assessment, and personal professional development.

“By leveraging LLMs and generative AI technologies, teachers can alleviate some of the burdens associated with lesson planning, personalized content creation, differentiation and personalized instruction, assessment, and personal professional development.This allows them to concentrate on addressing the needs of their pupils more effectively [1,7].”

• cites [1] Lo, C.K. (2023). What is the impact of ChatGPT on education? A rapid review of the literature. Educ. Sci., 13. sourceNot checked yet

  Not checked in the free run.

• cites [7] Giannini, S. (2023). The generative AI and the future of education, ED/ADG.Not checked yet

  Not checked in the free run.

May combine two claims.

In primary school education, pupils can receive assistance in learning reading, vocabulary, writing, and counting from the very early stages, as well as more advanced topics like writing styles, grammar, and text interpretation from generative AI.

“In primary school education, pupils can receive assistance in learning reading, vocabulary, writing, and counting from the very early stages, as well as more advanced topics like writing styles, grammar, and text interpretation [10,12,16,32,34].”

• cites [10] Jeon, J. (2022). Chatbot-assisted dynamic assessment (ca-da) for L2 vocabulary learning and diagnosis. Comput. Assist. Lang. L., 1–27. sourceNot checked yet

  Not checked in the free run.

• cites [12] Han, A., and Cai, Z. (2023, January 19). Design implications of generative AI systems for visual storytelling for young learners. Proceedings of the IDC’23: Proceedings of the 22nd Annual ACM Interaction Design and Children Conference, New York, NY, USA. sourceNot checked yet

  Not checked in the free run.

• cites [16] Baidoo-Anu, D., and Ansah, L. (2023, July 15). Education in the Era of Generative Artificial Intelligence (AI): Understanding the Potential Benefits of ChatGPT in Promoting Teaching and Learning. Available online: https://ssrn.com/abstract=4337484. sourceNot checked yet

  Not checked in the free run.

• cites [32] Dijkstra, R., Zülküf, G., Subhradeep, K., and Kamps, J. (2023, September 15). Reading comprehension quiz generation using Generative Pre-trained Transformers. Manuscript. Available online: https://intextbooks.science.uu.nl/workshop2022/files/itb22_p1_full5439.pdf.Not checked yet

  Not checked in the free run.

• cites [34] Gattupali, 2023, Comparing teacher-written and AI-generated math problem solving strategies for elementary school students: Implications for classroom learning, Coll. Educ. Work. Pap. Rep. Ser.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

ChatGPT-3.5 is capable of generating cohesive and informative human-like responses to user input.

“This artificial intelligence-based chatbot is capable of generating cohesive and informative human-like responses to user input [1].”

• cites [1] Lo, C.K. (2023). What is the impact of ChatGPT on education? A rapid review of the literature. Educ. Sci., 13. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

ChatGPT was used in preparation of teaching by generating course materials, providing suggestions, and performing language translation, and to generate assessment tasks for student performances.

“Firstly, ChatGPT was used in the preparation of teaching by generating course materials, providing suggestions, and performing language translation.Secondly, it was used to generate assessment tasks for the performances of students [1].”

• cites [1] Lo, C.K. (2023). What is the impact of ChatGPT on education? A rapid review of the literature. Educ. Sci., 13. sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim. May combine two claims.

Both designing learning material and generating interest among pupils can be accomplished with generative AI.

“Both are tasks that can be accomplished with generative AI.”

Classified as probably not a substantial claim. May combine two claims.

Spanish was chosen as the language for the research to ensure that language did not influence pupils' performance and to align with their mother tongue and language of instruction.

“To ensure that language did not influence the pupils' performance in the test and to align it with their mother tongue and language of instruction, Spanish was chosen as the language for this research.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The material only contains results from pupils who attended the whole test lesson and answered the questions.

“In fact, the material only contains results from those pupils who attended the whole test lesson and answered the questions.”

May combine two claims.

Pupils who did not know the answer usually answered something short quickly, and that was often not a correct answer.

“This suggests that pupils who did not know the answer usually answered something short quickly, and that was often not a correct answer.”

• AssumptionRests on an assumption: The generative AI model can accurately determine the correctness of pupils' answers.

Probably not a substantial claim: may not pass S3 (load-bearing).

On-site observations confirmed that the indicated gender proportions closely matched the actual gender distribution of the pupils.

“Our on-site observations confirmed that the indicated gender proportions closely matched the actual gender distribution of the pupils.”

May combine two claims.

Generative AI can be effectively used to differentiate pupils based on their prior knowledge and interest in the topic being studied in schools, and adjust the learning material accordingly both before and during the lesson.

“The findings of this article suggest that generative AI can be effectively used to differentiate pupils based on their prior knowledge and interest in the topic being studied in schools, and adjust the learning material accordingly both before and during the lesson.”

• AssumptionRests on an assumption: Pupils' self-perceived learning is a valid measure of their actual learning.
• AssumptionRests on an assumption: The generative AI-modified learning material actually matched each pupil's knowledge and interest levels as intended.

Probably not a substantial claim: may not pass S3 (load-bearing).

Parents or guardians had the option to prevent their child's participation in the research.

“The pupils' parents or guardians had the option to prevent their child's participation in the research.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Earlier studies among primary and secondary school pupils found that pupils were generally positive about the use of ChatGPT as a contemporary learning environment tool belonging to generative AI.

“Also, earlier studies among primary and secondary school pupils found that in general, pupils were positive about the use of ChatGPT as a contemporary learning environment tool belonging to generative AI [14,15].”

• cites [14] Lozano, A., and Blanco Fontao, C. (2023). Is the education system prepared for the irruption of artificial intelligence? A study on the perceptions of students of primary education degree from a dual perspective: Current pupils and future teachers. Educ. Sci., 13. sourceNot checked yet

  Not checked in the free run.

• cites [15] Bitzenbauer, 2023, ChatGPT in physics education: A pilot study on easy-to-implement activities, Contemp. Educ. Tech. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Spanish is classified as a High Resource Language and represents 4.3% of website content, making it the third most commonly used language.

“However, Spanish is classified as a High Resource Language (HRL).It holds a significant presence on the web, representing 4.3% of website content [39], making it the third most commonly used language.”

• cites [39] Statista (2023, July 15). Languages Most Frequently Used for Web Content as of January 2023, by Share of Websites. Available online: https://www.statista.com/statistics/262946/most-common-languages-on-the-internet/.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

All test materials were automatically collected and recorded through the designated test platform and stored in a secure digital framework.

“All test materials were automatically collected and recorded through the designated test platform and stored in a secure digital framework.”

May combine two claims.

This differentiation and the related adjustment of learning materials and tasks proved beneficial in promoting motivated learning among pupils and enhancing their performance.

“This differentiation and the related adjustment of learning materials and tasks proved beneficial in promoting motivated learning among pupils and enhancing their performance.”

• AssumptionRests on an assumption: The generative AI-modified learning material actually matched each pupil's knowledge and interest levels as intended.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Pupils took part voluntarily and had the right to withdraw at any moment or to leave questions without answers.

“In addition, pupils took part in the test voluntarily, and they had the right the withdraw from the test at any moment or to leave questions without answers.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The design of learning material and generating interest among pupils toward the topic learned during a classroom session are vitally important.

“This suggests the vital importance of the design of the learning material and of generating interest among pupils toward the topic learned during a classroom session.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The original reference material was sourced from worldhistory.org and validated by the UNESCO Archives.

“For the test lesson, the original reference material was sourced from worldhistory.org,and validated by the UNESCO Archives.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The names of the schools located in the capital of Uruguay are not disclosed in the article to protect pupil anonymity.

“For the sake of pupil anonymity, the names of the schools located in the capital of Uruguay are not disclosed in this article.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One author personally observed and guided the test procedures on-site to ensure meticulous implementation and avoid potential harm to participants.

“To guarantee that all these details and research procedures were meticulously implemented, and that potential harm to the research participants would not be caused, one of the authors of the article personally observed and guided the test procedures on-site.”

Probably not a substantial claim: may not pass S3 (load-bearing).

National and university guidelines on research integrity and ethics were rigorously followed in implementing the test lesson and analysis.

“To implement the test lesson and the following analysis, national and university guidelines on research integrity and ethics were rigorously followed.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Participants answered the test without providing their names, and their identities could not be identified from the results.

“To ensure participants' anonymity, they answered the test without providing their names, and their identities could not identified from the results.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Conversational AI can be employed to introduce variety to text-based exercises.

“Moreover, conversational AI can be employed to introduce variety to text-based exercises [36].”

• cites [36] Ji, 2023, A systematic review of conversational ai in language education: Focusing on the collaboration with human teachers, J. Res. Technol. Educ. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

This enables pupils to acquire learning skills and knowledge required for their current and future sustainable development.

“This enables them to acquire learning skills and knowledge required for their current and future sustainable development.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Generative AI can support the provision of access to the newest relevant information for learning and make that learning accessible and inclusive for all kinds of pupils with different knowledge levels and interests.

“It can support the provision of access to the newest relevant information for learning and in making that learning accessible and inclusive for all kinds of pupils with different knowledge levels and interests.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Safeguards, data practices and management followed the research code of conduct in accordance with discipline-specific principles.

“In addition, safeguards, data practices and management followed the research code of conduct in accordance with discipline-specific principles.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The suitability of the text and exercises was validated by the authors following RLHF principles.

“The suitability of the text and exercises were validated by the authors of this article, following the RLHF principles.”

May combine two claims.

Teachers and education authorities are consistently aware of pupils' situation and can identify moments when specific assistance and intervention are needed.

“As a result, teachers and education authorities are consistently aware of their situation and can identify moments when specific assistance and intervention are needed.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

When properly developed and implemented, generative AI has the potential to revolutionize education by providing easy access to high-quality learning environments.

“When properly developed and implemented, generative AI has the potential to revolutionize education by providing easy access to high-quality learning environments.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Prior to the test, school directors granted permission and teachers and pupils were informed about the testing process, including research integrity and ethical guidelines.

“Prior to conducting the test, following good research practices, the directors of the participating schools granted permission, and the teachers and pupils were duly informed about the testing process, including research integrity and ethical guidelines.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

This article responds to the widely recognized need, expressed by scholars and education managers, to study how generative AI and ChatGPT function in school education practices.

“This article, therefore, responds to the widely recognized need, expressed by scholars and education managers, to study how generative AI and ChatGPT function in school education practices, e.g., [7][8][9]16,[32][33][34].”

• cites [7][8][9]16,[32][33][34]Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

Generative AI in digital learning environments has the major potential to support broader sustainable development, including SDG4.

“In fact, generative AI in the realm of digital learning environments has the major potential to support broader sustainable development [20], including SDG4 [6].”

• cites [20] United Nations (2022). Sustainable Development Goals.Not checked yet

  Not checked in the free run.

• cites [6] Chen, 2019, The roles of engagement and competition on learner’s performance and motivation in game-based science learning, Educ. Technol. Res. Dev. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S4 (based).

One key strength of utilizing generative AI lies in its ability to design learning materials that foster improved performance by leveraging pupils' motivation to learn.

“One key strength of utilizing generative AI lies in its ability to design learning materials that foster improved performance by leveraging pupils' motivation to learn.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The length of a text modified by generative AI is not a direct indicator of its complexity.

“The length of a text modified by the generative AI is not a direct indicator of its complexity.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

This can create impactful, inclusive, and sustainable learning in schools to benefit pupils, teachers and education managers alike, all of which is significant for sustainable development.

“This, in turn, can create impactful, inclusive, and sustainable learning in schools to benefit pupils, teachers and education managers alike, all of which is significant for sustainable development.”

May combine two claims.

Pupils who knew the answer elaborated on their responses, and these responses were usually correct.

“Those who knew the answer elaborated on their responses, and these responses were usually correct (Table 7).”

• AssumptionRests on an assumption: The generative AI model can accurately determine the correctness of pupils' answers.

Classified as probably not a substantial claim. May combine two claims.

The overall number of analyzed pupils was rather limited and analysis was conducted only regarding one study topic, so for broader generalization, wider samples and other studies would be needed.

“The overall number of analyzed pupils was rather limited and analysis was conducted only regarding one study topic, so for broader generalization, wider samples and other studies would be needed.”

Probably not a substantial claim: may not pass S2 (checkable).

Teachers need to play an important role in designing and implementing the ways generative AI is used so that all benefits can be taken advantage of.

“This suggests that teachers need to play an important role in designing and implementing the ways generative AI is used so that all benefits can be taken advantage of.”

Probably not a substantial claim: may not pass S2 (checkable).

The latter suggests an important possibility for the use of generative AI in learning environments.

“The latter suggests an important possibility for the use of generative AI in learning environments.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

This will enhance their learning engagement and result in motivating learning experiences that are cognitively ergonomic, encouraging them to move into deeper knowledge along BLOOM's taxonomy.

“This will enhance their learning engagement and result in motivating learning experiences that are cognitively ergonomic, encouraging them to move into deeper knowledge along BLOOM's taxonomy.”