GenAI as a Tool for Content Generation in Hypermedia Edutainment

Applications: Potential and Limitations

Wolfgang Heiden

, Veronika Saitova

, Tea Kless

, Valeska Wegner

, David Rötter

and Thomas Neteler

Institute of Visual Computing, Bonn-Rhein-Sieg University of Applied Sciences, Sankt Augustin, Germany

Department of Mathematics, Justus-Liebig-Universität Gießen, Gießen, Germany

Keywords: Crossmedia Storytelling, Hypermedia Edutainment, Personalization, Generative AI.

Abstract: Generative AI can considerably speed up the process of producing narrative content including different media.

This may be particularly helpful for the generation of modular variations on narrative themes in hypermedia,

crossmedia, or transmedia contexts, thereby enabling personalized access to the content by heterogenous

target groups. We present an example where GenAI has been applied for image creation and translation of a

text to multiple languages for a crossmedia edutainment project transferring IT security knowledge to

vulnerable groups. GenAI still seems inadequate to produce interesting narrative text integrating dedicated

educational content. AI-generated illustrations often require manual rework. However, LLM support in

multilingual translations displays more intelligent solutions than expected, including the implementation of a

password generation process from a narrated description.

1 INTRODUCTION

Generative Artificial Intelligence (GenAI) has

brought about a quantum leap to a broad variety of

Information Technology (IT) services related to the

production of media content. Especially for the

production of multimedia narratives GenAI tools like

ChatGPT, Stable Diffusion, etc. offer support that

bears the potential to speed up the creative process

significantly.

We have developed an environment for delivering

information related to Cyber Security especially to

vulnerable groups like elderly people following a

crossmedia approach that includes human

“mediators” and traditional media (hardware) as well

as a hypermedia edutainment platform (Heiden et al.,

2023). The edutainment approach is to subliminally

integrate educational content in an entertaining

narrative (Heiden et al., 2024). Personalization is an

important issue, since the target groups vary

considerably in many aspects, for example: personal

preferences, previous knowledge, time constraints, or

cognitive abilities. Therefore, an edutainment product

https://orcid.org/0009-0006-1304-5496

on a single topic should be offered in multiple

variations, whose production would usually cost

much time and effort.

1.1 Motivation

When creating edutainment material, it is important

to find the right balance between the educational

content as the core of information that shall be

transported to the users with an entertaining

framework as its vehicle. Acceptance by the target

group (and thereby educational success) largely

depends on this balance as well as on a seamless

integration of both aspects. Although Large Language

Models (LLMs) like Open AI’s ChatGPT can help to

gather and summarize information for the educational

content, its correctness must still be checked

thoroughly by human experts. Embedding this

content in an interesting and entertaining narrative

that reaches people who are rather reluctant to

teaching offers like seminars, workshops, or online

resources requires creative skill. Due to the already

mentioned personalization requirements, the

edutainment products have to not only be technically

444

Heiden, W., Saitova, V., Kless, T., Wegner, V., Rötter, D. and Neteler, T.

GenAI as a Tool for Content Generation in Hypermedia Edutainment Applications: Potential and Limitations.

DOI: 10.5220/0013298100003932

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Computer Supported Education (CSEDU 2025) - Volume 1, pages 444-451

ISBN: 978-989-758-746-7; ISSN: 2184-5026

correct but also tailored to the varying preferences

between and even within the different target groups.

Variations can be related to many aspects like

narrative genre, length/duration, media format,

language, or linguistic style. Creating the narrative

first as a generic elementary structure and then

producing variations on the theme, partially enriched

by multimedia artwork and/or interactive components

is a challenging task not only in terms of creative

potential but also by sheer expenditure of time. This

is laborious and requires a large number of personnel

if done manually. Since the process of creating such

additions to an existing framework is repetitive, it

lends itself to the use of generative AI tools.

1.2 Research Gap

Despite their unquestioned impressive capabilities,

there are still considerable limitations of AI systems

in general and text generating Large Language

Models (LLMs) in particular. These limitations must

be considered thoroughly before giving away LLM

output - especially but not limited to when it is

designed to transport trustworthy information. For

example, LLMs have been found to produce a bias

when summarizing newspaper content (Fang et al.,

2024). There is also evidence that LLMs may give

citations to non-existing references (Day, 2023). It

has recently been shown that AI systems trained on

databases with a considerable percentage of AI-

generated content tend to produce lower quality

output (Shumailov et al., 2024).

Our focus is on the creative potential of GenAI

systems. Although there is an ongoing hype to use

GenAI for narrative creation including a discussion

about its role as either autonomous or supporting

tools (Fisher, 2023) and to explain how it works

(Brüns and Meißner, 2024), there is currently no

structured study about how and to which extent these

tools can raise the efficiency of the creative process

in terms of producing a large number of variations on

a narrative theme.

1.3 Research Questions

Within the scope of our study which primarily aims at

designing and installing a crossmedia edutainment

environment for teaching IT security to vulnerable

groups, we have evaluated the usefulness of GenAI

for supporting the generation of a variety of narrative

modules related to a common topic. We thereby

wanted to find answers to the following questions:

1. How and to which extent can GenAI support

the creation of an entertainment framework

that embeds educational content resulting in

edutainment stories?

2. What are the practical limitations of this

approach?

1.4 Method Summary

We created an edutainment artefact by combining an

initial version of an entertainment framework using a

story that includes educational content. This initial

version is then altered by GenAI tools with different

goals. Their output is checked by humans on their

level of quality, how well the process could support

creative personnel, and if the educational content is

unaffected by the alterations. GenAI tools are used to

create supportive imagery based on the text of the

stories and to translate the stories into different

languages. To somehow quantify the quality of

translation, we counted the words that were changed

by human inspectors (native speakers) when these

were asked to check the AI translations and improve

the linguistic quality. The time gained by GenAI

support was measured by comparing the temporal

effort it took a human author to perform a selected

task component with the overall time required for a

comparable task with support by AI tools, also

including human rework in post-production.

2 RELATED WORK

Transmitting information through stories has a long

tradition going over millennia. However, after

storytelling as a means of delivering information has

been largely replaced by various other media, it has

survived in entertainment and is now going through a

renaissance with edutainment or infotainment.

It is well-known that information is better and

more permanently received when it comes along with

emotion (Kensinger, 2009) and when it enters the

human mind through more than one sensory channel

(Ponticorvo et al., 2019). Edu-/infotainment plays a

particular role for reaching people that are reluctant

to actively seek advice or to make use of educational

offers even when a topic of general interest is

concerned.

We found this reluctance in elderly people

regarding IT services in general due to a feeling of

low self-efficacy. Therefore, we are looking for a way

to teach these people important lessons on IT security

that could enable them to safely benefit from digital

health or financial services without fear of falling

victim to cyber-attacks. We chose an edutainment

approach that subliminally transports knowledge as

GenAI as a Tool for Content Generation in Hypermedia Edutainment Applications: Potential and Limitations

445

well as confidence through stories accessed primarily

for entertainment (Heiden et al., 2024).

We also found a broad variety of preferences for

different genres and media within our target group, in

addition a broad range of previous knowledge and

interest in cyber security topics. Therefore, we

decided on an approach to enable access to IT security

to elderly people through crossmedia storytelling

(Heiden et al., 2023). This includes personal readings

in front of an audience as well as media hardware

(printed booklets, audio CDs, etc.) and an online

repository with the structure of a Hypermedia Novel

(HYMN) (Heiden, 2006).

Offering different types of stories via different

media in different languages and with variation of

several parameters enables a personalized access to

the content for everyone. However, these

personalization options require a lot of creative effort

to produce multiple storytelling products.

We focused on the topic of passwords for our

initial storytelling, because an interview study with

our target group indicated issues in handling secure

and memorable passwords in everyday activities

(Heiden et al., 2023). This coincides with the surveys

of the public; some people have issues with

understanding the potential attack vectors with

passwords (Ur et al., 2016). Our target group is

especially vulnerable to knowledge gaps. First

interviews revealed that a large percentage of the

questioned elderly persons had not dealt with the

topic until they were forced into the digital realm by

external factors like financial services or online

shopping. Therefore, we aim to equip these people

with beginner information about passwords in a way

that can be consumed as entertainment media.

The area of password generation is well

researched (Adams et al., 1997; Tan et al., 2020) and

has recently also included the use of GenAI

(Umejiaku et al., 2023). As password security and

memorizability are only used as an example for the

learning content within an edutainment narrative, it is

by itself not in the focus of our study and therefore

not addressed in detail in this report.

3 AI SUPPORT FOR

STORYTELLING

Creating narratives is hard work and takes its time. A

large variety of successful text and television series

proves that skilled and experienced (teams of) authors

can produce fiction in a kind of production line on a

weekly or even daily basis. However, creating high

quality narratives both interesting and educative (and

technically correct) while keeping within given

constraints (e.g. in terms of length), additionally

enriched by different media and branching into

variations for several aspects of personalization,

would require a considerable amount of time and

effort from both authors and scientific experts.

Covering a significant number of relevant topics for

the vulnerable target groups may well cost so much

time that the provided information might be outdated

before the material is available. Therefore, we used

generative AI (GenAI) tools to support the creation

and variation of the content.

In order to disseminate the subliminal information

as widely and efficiently as possible, we have tried AI

tools for the generation of text, images and translation

of the stories into several languages. Our evaluation

of these studies indicates that AI can be useful to

quickly produce supporting media and translation

drafts which, however, still require some human

rework before being applicable. Despite this post-

processing requirement, the use of AI as a tool still

speeds up the production process of story variations

significantly, thereby supporting personalization of

access to the information for different people with

different preferences and constraints.

We found GenAI not yet ready to autonomously

produce stories that match our requirements to be

both interesting and informative, of adequate length

and with dedicated learning content seamlessly

integrated in the story. However, it proved to be a

useful tool for efficiently modifying an existing story

to adapt it to different target groups. We used GenAI

to generate narrative personalization of our baseline

story framework to build a rich set of options users

can choose from.

3.1 Image Generation

Every story should have a frontispiece. To attract

attention and trigger an urge to look inside, it should

not just be any cover picture but rather one with “eye-

catcher” quality. GenAI today can produce images of

impressive quality from textual description

(“prompts”) that artists formulate based on their

imagination. Many tools with several comparable and

some differing strengths and weaknesses are

available for free or payment, most of them online

and some even free to install locally. Among those

image creation AI-tools are DALL·E, Stable

Diffusion, Craiyon, Midjourney, NightCafe, and AI

Mirror, just to name a few. Despite their compelling

abilities, these tools still have some limitations (Ye et

al., 2024).

CSEDU 2025 - 17th International Conference on Computer Supported Education

446

Typical aberrations in AI-generated pictures are

related to anatomical details or neglecting explicit

positive or negative prompts. Objects in images may

appear differently than the prompts suggest, and

removal of objects alters the image in unexpected

ways, such as changing unrelated objects in addition.

Although we suspect that those issues will most

probably vanish sooner rather than later considering

the tremendous advances that become visible from

one AI version to the next, current GenAI tools often

still require human rework on details before a picture

is ready to serve as a cover image for a story (or an

internal illustration). There is, however, no doubt that

GenAI can instantly produce images of considerable

visual quality, especially if no professional artists are

involved. Post-processing like selecting an image out

of a pool of AI-generated suggestions and fine-tuning,

for example correcting some anatomical details such

as the number of fingers then takes minutes instead of

at least hours to produce an entire picture.

Some free text feedback from our questionnaires

indicates that there is an interest within our target

group for more extensive illustration of the stories by

inside pictures (or even a graphic novel version).

Consistency of recognizable characters between

different pictures is currently another weakness of

image-producing GenAI, but there are already

promising approaches to cope with this problem

(Martini et al., 2024).

3.2 Text Translation

GenAI using Large Language Models (LLM) like

ChatGPT, LLaMA or Claude are quite good at

translating text from one language to another. There

are even GenAI tools specialized on translations like

DeepL. We have used ChatGPT 4.0 to translate our

crime story from German to English and Russian

using the following prompts, respectively: "Translate

the following into English: [Then the entire German

original text of the story named „Der Stick des

Todes‟ was given]" (which resulted in an output with

the title "stick of death") and "Please translate "stick

of death" to Russian additionally". After we found

several severe errors in the first Russian translation

attempt, a second try delivered a more useful result.

The English translation has been reworked by the

author of the original story and additionally by a

native speaker (U.S. American) and checked for

consistency with the original. The Russian translation

has been reworked by one of the authors (who is a

Russian native speaker and a scientist working in the

research project) and compared with a translation of

the original text done by a human professional

translator.

3.2.1 Translation from German to English

Comparing a translation of the crime story from the

German original into English by ChatGPT with a

revised version of this translation, we found that –

apart from stylistic considerations – only few

corrections were necessary. Most of the required

changes were related to puns or untranslatable

proverbs.

An approach to quantify the translation quality by

counting the words that had to be changed in the

revision process shows that only about 5 % were

changed to end up with a correct and well-readable

text. We are aware that this quantification approach

may not meet hard scientific standards, since only few

changes would have been necessary in terms of

maintaining the original meaning of the text, while

others are rather of a linguistic nature or depending

on personal preferences for stylistic aspects. It should,

however, give at least a general impression of how

much time and effort can be saved by replacing

human workload for a first translation iteration by

GenAI.

3.2.2 Translation to Russian

The translation to Russian by ChatGPT 4.0 produced

a result that required rework on about 30 % of the

words generated by the LLM using the same

quantification approach as for the English text. We

reviewed the LLM translation twice. It was reworked

separately by a native speaking scientist and a

professional translator. We found that the above-

mentioned percentage of rework is valid for both

versions.

3.2.3 Other Language Variations

To switch our target group from elders to youths with

a non-academic background, we have also tried to

make ChatGPT translate the text to a simplified

wording. This attempt only created a summary of the

story which was written in simple language but

missed the informative content. Another attempt to

make ChatGPT shorten the text by half did quite a

good job keeping the relevant information while only

producing a few consistency leaks that could easily

and quickly be mitigated by reinserting a few

sentences from the original text without thwarting the

condensation significantly.In comparison, shortening

the original text from a reading duration of 60 min to

GenAI as a Tool for Content Generation in Hypermedia Edutainment Applications: Potential and Limitations

447

45 min manually required about 2 hours work by the

author.

We also found that GenAI can be very helpful to

illustrate a story with accompanying pictures (which

was a suggestion from one participant of our study to

make the reading more relaxing). However, even with

thorough iterative prompting it was impossible to get

an image that did not require at least a little manual

revision to reach a status ready for publication

together with the story.

3.2.4 Artificial Intelligence?

While the classical “Turing Test‟ (Turing, 1950)

poses no challenge to nowadays’ chatbots, there is

still an ongoing debate over LLMs actually

understanding the content of textual descriptions

(Mitchell and Krakauer, 2023). LLMs generating text

are rather considered as a kind of “digital parrot‟ not

aware of the meaning of what they seem to “say”,

since the underlying algorithms are driven by

statistical analyses of text elements appearing close to

each other in a vast database of example text used for

their training (Zhao et al, 2023). Although these

chatbots produce text output in high quality that is

even able to successfully pass academic examinations

in almost every scientific discipline (Sadeq et al,

2024) or give explanations and summaries of

informative texts that are helpful to students and

teachers alike, these systems are therefore mostly not

considered “intelligent” in a human-like fashion.

Figure 1: Password generation from a passphrase. The ex-

ample is based on an extended proverb. The first character

(major or minor) of each word is used, special characters

are copied, words describing numerals are used as numbers.

It is remarkable that in our studies ChatGPT was

able to detect and (partially) adopt the described

algorithm for the generation of a secure and

memorable password from a phrase (see Figure 1),

although it was unable to realize the use of puns in

some chapter titles and thus translated these word by

word, thereby losing the original meaning of the play

on words. For teaching password security topics in

our story, we used an extended version of a traditional

German proverb (which has a direct equivalent in

English but not in Russian) to explain the generation

of secure yet memorable passwords from a

“passphrase”. In the story, the passphrase eventually

leading to a strong password is generated in three

phases: (1) The origin is the well-known proverb “Der

frühe Vogel fängt den Wurm” (The early bird catches

the worm), which is then (2) extended by the

(inofficial) follow-up “… aber erst die zweite Maus

kriegt den Käse” (… but only the second mouse gets

the cheese), thereby prolongating the term and adding

a numeral (“second” becomes “2”). (3) An exchange

of the word “Käse” (cheese) by “Speck” (bacon) and

a closing exclamation mark finally introduces a

variation on the standard proverb and adds a special

character.

As an intermediate result, combining the original

sentence (phase 1) with the extension by a comma,

the German phrase: “Der frühe Vogel fängt den

Wurm, aber erst die zweite Maus kriegt den Käse”

was originally explained leading to the following

password by taking the initial character of each word

(in its original case) unless it is a number which is

then converted to the according numerical character:

“DfVfdW,aed2MkdK”. The phrase was translated

(more or less) correctly by ChatGPT to “The early

bird catches the worm, but [only] the second mouse

gets the cheese” and “TEBCTW,BTSMGC”,

respectively. It is interesting that the GenAI

recognized the relation between the passphrase and

the password, but – unlike the original – used only

capital letters for transformation to a password and

lost the “t” from “the” before “cheese”. The comma

added to the resulting password in the German

original when combining the two parts of the phrase

was kept in the English translation although it was

only mentioned in the text but not explicitly written

in a modified repetition of the phrase. However,

ChatGPT seems to overlook the rule to transform the

word “second” into a number (“2”). When finally

adding an exclamation mark in the end in the German

original after the word “Käse” (cheese) had been

replaced by “Speck” (bacon) to enhance the

passphrase’s complexity (eventually resulting in

“DfVfdW,aed2MkdS!”) the translation seems to get

confused and changes the final password to

“TEBCTW,BTSMGW!”, losing the correspondence

to the phrase of origin at the end without following

the exchange that should have lead from “c‟ to “b‟. It

should be mentioned that again both additional

changes were only explained in a dialogue within the

story, where the finally resulting password was then

given without prior explicit reformulation of the

entire phrase as its direct origin.

CSEDU 2025 - 17th International Conference on Computer Supported Education

448

In the Russian translation, where no literally

analogous proverb exists, ChatGPT translated the

first (traditional) part of the sentence word by word

and therefore equivalent to the German (and English)

version: “Ранняя пташка ловит червяка‟. The

extension, however, is translated analogously to the

English version: “но вторая мышка получает сыр‟.

When producing a password from the entire sentence,

ChatGPT differs from the English translation in

skipping the comma as well as the following word

(но/but) but implementing the switch from the word

“вторая/second‟ to the numerical character “2‟:

“Рплч2мпс”. When finally the exclamation mark

should be added (together with the cheese/bacon

exchange), unlike in English, this is done by ChatGPT

correctly with a change from “с” (сыр - cheese) to “б”

(бекон - bacon): “Рплч2мпб!‟.

The entire procedure comparing the original and

the translated versions with the correct

implementation, and as generated by ChatGPT is

illustrated in Figure 2.

Figure 2: Password generation from a passphrase based on

a German proverb. The original text has been translated to

English and Russian by ChatGPT 4.0, which then also esti-

mated the resulting password. The correct deduction of a

password from the phrase following the given textual de-

scription (blue arrows) is compared to the results produced

by ChatGPT (red arrows).

These observations lead to the impression that the

LLM has achieved more than to only put together

linguistic elements that fit together well in terms of

syntactic rules and semantic content but also don’t

require a true understanding of the original text. To

perform the transformation from passphrase to

password – even though not completely flawless – it

had to somehow extract the underlying algorithm

either from the explanation given in the text or by

deduction from the implicit relationship between

passphrase and password in the original text.

The English translation as evaluated in detail in

the above sections was generated by ChatGPT 4.0 in

May 2024, the Russian translation 10 days later. In

November 2024, we asked ChatGPT 4o to translate

the German original again to English and Russian and

additionally to French and Spanish. In the resulting

text versions, we placed our focus on how the LLM

translated the passphrase (including its extension but

no further changes) and on which password it derived

from that. The results are summarized in Table 1.

Table 1: Passphrase translations by ChatGPT 4o together

with derived passwords.

Language

Original

hrase

Phrase

extension

Password

German Der frühe

Vogel fängt

den Wurm

... aber erst

die zweite

Maus kriegt

den Käse

DfVfdW,

aed2MkdK

English The early

bird catches

the worm

…but only

the second

mouse gets

the cheese

Tebctw,

botsmgtc

Russian Ранняя

пташка

ловит червя

… но

только

вторая

мышь

получает

сыр

Рплч,а2мпс

French Le monde

appartient à

ceux qui se

lèvent tôt

… mais

seule la

deuxième

souris

attrape le

froma

Lmaaqlslt,

msl2mac

Spanish Al que

madruga,

Dios le

ayuda

… pero solo

el segundo

ratón

obtiene el

queso

Aqmdlh,

psr1oq

As a first observation it can be noticed that the

LLM had in general significantly improved since the

first attempt. It was able to recognize the original

German proverb and translate it to its equivalent in

other languages if such an equivalent exists (as it does

literally in English and by meaning in French and

Spanish). In Russian, which lacks a proverb with a

similar meaning, the phrase is translated literally. The

same happens with the (unofficial) extension which

has no correspondence in any other language.

It is certainly venial for the AI not to realize that

the extension only makes sense when related to the

original phrase. This, however, may be seen as an

indication that the LLM indeed did not actually

understand the explanation of password generation in

the text but rather seems to have deduced the

GenAI as a Tool for Content Generation in Hypermedia Edutainment Applications: Potential and Limitations

449

procedure from detecting a correspondence between

the phrase and the resulting password (which is still a

remarkable cognitive accomplishment, although it

adds or loses single characters in some cases, fails in

the consideration of capital and minor letters and – at

least in some of the translations – also fails in

transferring a word that represents a numeral into its

corresponding numeric character).

4 CONCLUSIONS

Coming back to the initial research questions about

how and to which extent GenAI can support the

production of hypermedia edutainment material, a

considerable gain in production speed could be found,

especially with image generation and text adaptations

like translation or reformulation within a given time

frame. We have found the support of Generative AI

significantly helpful for content production for a

modular hypermedia or crossmedia network on a

common theme. Although all GenAI output still

required some human refactoring, it was well-suited

to speed up the process of content creation. Despite

expectable limitations the translation of narrative text

even when it includes complex descriptions of a

process like the generation of a password from

passphrases based on proverbs for example, even

including extensions and modifications, the GenAI

showed – though still limited – considerable and

unexpected abilities.

Despite the fact that we found several interesting

hints to current abilities and limitations of GenAI

tools for supporting the production of hypermedia

narratives in an edutainment framework, in particular

through image generation and the modification and

translation of text, this study only casts a narrow

spotlight on the subject, as it lacks a large-scale

structured analysis. We intend to address this issue by

our future work, which aims at gaining deeper insight

into the varying requirements of our target groups and

the parameters that influence learning success, as well

as the progression of GenAI capabilities and in

particular a structured analysis of how different LLM

systems deal with the procedural description of

complex tasks when provided in a narrative context.

ACKNOWLEDGEMENTS

The authors like to thank Karl N. Kirschner for

thoroughly proofreading the English crime story

translation. The Evangelische Erwachsenenbildung

an Sieg und Rhein helped with conducting the

interview study. This work is supported by the

German Federal Ministry of Education and Research

(BMBF) under the research grant 16KIS 1623.

REFERENCES

Adams, A., Sasse, M. A., & Lunt, P. (1997). Making pass-

words secure and usable. In: People and computers xii,

pp. 1–19. Springer, Heidelberg

Brüns, J. d., Meißner, M. (2024). Do you create your con-

tent yourself? Using generative artificial intelligence

for social media content creation diminishes perceived

brand authenticity, Journal of Retailing and Consumer

Services 79, 103790, https://doi.org/10.1016/j.jretcon-

ser.2024.103790.

Day, T. (2023). A Preliminary Investigation of Fake Peer-

Reviewed Citations and References Generated by

ChatGPT. The Professional Geographer, 75(6), 1024–

1027. https://doi.org/10.1080/00330124.2023.2190373

Fang, X., Che, S., Mao, M., Zhang, H., Zhao, M., Zhao, X.

(2024). Bias of AI-generated content: an examination of

news produced by large language models. Sci Rep 14,

5224.

Fisher, J. A. (2023). Centering the Human: Digital Human-

ism and the Practice of Using Generative AI in the Au-

thoring of Interactive Digital Narratives. In: Holloway-

Attaway, L., Murray, J. T. (eds.) ICIDS 2023, LNCS

14383, 73–88, https://doi.org/10.1007/978-3-031-

47655-6_5

Heiden, W. (2006). Edutainment Aspects in Hypermedia

Storytelling. In: Pan, Z., Aylett, R., Diener, H., Jin, X.,

Göbel, S., Li, L. (eds) Technologies for E-Learning and

Digital Entertainment. Edutainment 2006. LNCS, vol

3942. Springer, Berlin, Heidelberg. https://doi.org/

10.1007/11736639_50

Heiden, W., Kless, T., Neteler, T. (2023). A Crossmedia Sto-

rytelling Platform to Empower Vulnerable Groups for

IT Security. In: Holloway-Attaway, L., Murray, J. T.

(eds.) ICIDS 2023, LNCS 14384, pp. 195–201.

Springer, Heidelberg. https://doi.org/10.1007/978-3-

031-47658-7_17

Heiden, W., Kless, T., Saitova, V., Wegner, V., Rötter, D.,

Neteler, T. (2024). Subliminal Teaching for Elderly

People Through Crossmedia Storytelling. In: Murray,

T. and Reyes, M. C. (eds.): ICIDS 2024, LNCS 15468,

pp. 196–204. Springer, Heidelberg. https://doi.org/

10.1007/978-3-031-78450-7_13

Kensinger, E. A. (2009). Remembering the Details: Effects

of Emotion. Emotion Review, 1(2), 99-113.

https://doi.org/10.1177/1754073908100432

Martini, L., La Tessa, L., Zolezzi, D. (2024). Building Con-

sistent Characters through Open-Source Generative AI.

International Journal of Emerging Technologies in

Learning (iJET), 19(8), pp. 82–89. https://doi.org/

10.3991/ijet.v19i08.50223

Mitchell, M., Krakauer, DC (2023). The debate over under-

standing in AI's large language models. Proc Natl Acad

CSEDU 2025 - 17th International Conference on Computer Supported Education

450

Sci USA. 2023 Mar 28;120(13):e2215907120. doi:

10.1073/pnas.2215907120. Epub 2023 Mar 21. PMID:

36943882; PMCID: PMC10068812.

Ponticorvo, M., Di Fuccio, R., Ferrara, F., Rega, A.,

Miglino, O. (2019). Multisensory Educational Materi-

als: Five Senses to Learn. In: Di Mascio, T., et al. Meth-

odologies and Intelligent Systems for Technology En-

hanced Learning, 8th International Conference.

MIS4TEL 2018. Advances in Intelligent Systems and

Computing, vol 804. Springer, Cham. https://doi.org/

10.1007/978-3-319-98872-6_6

Sadeq, M.A., Ghorab, R.M.F., Ashry, M.H., Abozaid, A.M.,

Banihani, H.A., Salem, M., Abu Aisheh, M.T.,

Abuzahra, S., Mourid, M.R., Assker, M.M., Ayyad, M.,

El Din Moawad, M.H. (2024). AI chatbots show prom-

ise but limitations on UK medical exam questions: a

comparative performance study. Sci Rep 14, 18859.

https://doi.org/10.1038/s41598-024-68996-2

Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., An-

derson, R., Gal, Y. (2024). AI models collapse when

trained on recursively generated data. Nature 631, 755–

759.

Tan, J., Bauer, L., Cranor, N. C., Cranor, L. F. (2020). Prac-

tical Recommendations for Stronger, More Usable

Passwords Combining Minimum-strength, Minimum-

length, and Blocklist Requirements. In: Proceedings of

the 2020 ACM SIGSAC Conference on Computer and

Communications Security (CCS '20). Association for

Computing Machinery, New York, NY, USA, 1407–

1426.

Turing, A. M. (1950) I.—Computing Machinery and Intel-

ligence. Mind LIX (236), 433–460, https://doi.org/

10.1093/mind/LIX.236.433

Umejiaku, A. P., Dhakal, P., Sheng, V. S. (2023). Balancing

Password Security and User Convenience: Exploring

the Potential of Prompt Models for Password Genera-

tion. Electronics 12, 2159.

Ur, B., Bees, J., Segreti, S. M., Bauer, L., Cranor, N. C.,

Cranor, L. F. (2016). Do Users' Perceptions of Password

Security Match Reality? In: Proceedings of the 2016

CHI Conference on Human Factors in Computing Sys-

tems (CHI '16). Association for Computing Machinery,

New York, NY, USA, 3748–3760.

Ye, Y., Hao, J., Hou, Y., Wang, Z., Xiao, S., Luo, Y., Zeng,

W. (2024). Generative AI for visualization: State of the

art and future directions, Visual Informatics, 8 (2),

2024, 43-66, ISSN 2468-502X, https://doi.org/10.1016/

j.visinf.2024.04.003.

Zhao, H., Chen, H., Yang, F., Liu, N., Deng, H., Cai, H.,

Wang, S., Yin, D., Du, M. (2024). Explainability for

Large Language Models: A Survey. ACM Trans. Intell.

Syst. Technol. 15, 2, Article 20 (April 2024), 38 pages.

https://doi.org/10.1145/3639372

GenAI as a Tool for Content Generation in Hypermedia Edutainment Applications: Potential and Limitations

451