P ac ice Pape
Recommended ci a ion: Bulmann, U., & S ahlbe g, N. (2025). P ac ice Re lec ions
Wi hin a Wo kshop: Supe ising S uden s’ Scien i ic Reading in Thesis W i ing in
Times o A i icial In elligence. In Kangaslampi, R., Langie, G., Jä inen, H.-M., &
Nagy, B. (Eds.), SEFI 53 d Annual Con e ence. Eu opean Socie y o Enginee ing
Educa ion (SEFI), Tampe e, Finland. DOI: 10.5281/zenodo.17631323.
This Con e ence Pape is b ough o you o open access by he 53 d Annual Con e ence
o he Eu opean Socie y o Enginee ing Educa ion (SEFI) a Tampe e Uni e si y in
Tampe e, Finland. This wo k is licensed unde a C ea i e Commons
A ibu ion-NonComme cial-Sha e Alike 4.0 In e na ional License.
PRACTICE REFLECTIONS WITHIN A WORKSHOP: SUPERVISING
STUDENTS’ SCIENTIFIC READING IN THESIS WRITING IN TIMES
OF ARTIFICIAL INTELLIGENCE
Ul ike Bulmanna,
1
, Nadine S ahlbe gb
aHambu g Uni e si y o Technology, Hambu g, Ge many
bHambu g Uni e si y o Technology, Hambu g, Ge many
Con e ence Key A eas: Digi al ools and AI in enginee ing educa ion; Building he
capaci y and s eng hening he educa ional compe ences o enginee ing educa o s
Keywo ds: eading compe ences, c i ical li e acy, AI li e acy, TL;DR, hesis
supe ision
ABSTRACT
To e lec i ely supe ise s uden s in inal heses along he esea ch p ocess is e y
much unde p essu e in he age o a i icial in elligence (AI). Being inspi ed by
TL;DRs (abb e ia ion o « oo long; didn’ ead», au oma ically gene a ed hype -sho
pape summa ies), we emphasize in his a icle on he a he una en i e aspec o
eading compe ences wi hin he in e winned eading – w i ing – esea ching –
c i ical hinking app oach. We asked: ‘’How can academics suppo s uden s’ eading
compe ences when supe ising hem in hei inal heses in he age o AI?’’ Thus, we
encou aged e lec ion in a wo kshop o 17 supe iso s by using (1) a sel -designed,
su ey consis ing o h ee pa s: e lec ion, exe cise and ans e , and (2) a pee
exchange. Supe iso s’ e lec ions showed ha hey ead scien i ic a icles wi h joy,
less ime and ely on adi ional eading s a egies a he han using AI ools o
eading. Being unawa e o TL;DRs i s , an exe cise on w i ing and gene a ing a
hype -sho summa y using a uni e si y’s HAWKI-based LLM led hem o e alua e
he ex quali y o be bo h p omising and isky. This esul ed in assessing hei
aining o compe ences o be mul i ace ed. Toge he , hey upda ed hei supe ision
guidelines conside ing mul iple deskilling isks and a ious compe ence de elopmen
po en ials o s uden s when using AI o no . Finally, we a gue ha such p ac ical
e lec ions and pee dicussions aise supe iso s’ awa eness o esponsible
guidance o s uden s in hei inal heses (bes ea lie wi hin he cu iculum) o
s eng hen hei c i ical and AI li e acy in an AI-en iched lea ning en i onmen .
1
Co esponding Au ho
U. Bulmann
ul ike.bulmann@ uhh.de
1 INTRODUCTION
In inal heses, i is key ha s uden s a e supe ised by compe en academics using
salien p ac ices (Shanahan e al., 2016) any hing bu e lec ed (Thompsen, 2022)
especially in imes o a i icial in elligence (AI). A his s age, s uden s a e expec ed
o be able o wo k (pa ially) independen (ly) on a esea ch ques ion along a
esea ch p ocess (Hube , 2009) using echnology esponsibly. Th oughou he
cu iculum, i is bes ha s uden s de elop C i ical Li e acy in an in eg a ed eading-
w i ing- esea ching-c i ical hinking app oach (Appo o a & Ho ning, 2023). This
in ol es scien i ic eading compe ences meaning o be able o unde s and and
e lec on ex s in a comp ehensi e sense. Speci ically, his includes (1)
unde s anding hei meaning in a de ailed and s uc u ed way, (2) being able o
c i ically e lec on hei quali y and a gumen a ion, (3) ca ego ising hem in he
ele an academic discou se, (4) assessing he ele ance o he ex ela ed o one's
own esea ch ques ions and (5) making jus i ied decisions on whe he o use
in o ma ion o no (ibid., Wol e e al., 2020, Philipp, 2024).
Un o una ely, o decades and h oughou disciplines, i has widely been
expe ienced by lec u e s ha s uden s o en do no ully o closely ead ex s and in
u n lack in-dep h scien i ic eading skills (Appo o a & Ho ning, 2023). Complexi y
inc eases due o mul i-dimensional aspec s: (1) p oblema ic assump ions, like he
belie ha s uden s can al eady ead scien i ically upon en e ing hei s udy
p og ams; (2) a agmen ed implemen a ion o c i ical li e acy app oaches h oughou
s udy p og ams and (3) a no ully de eloped c i ical li e acy skill se o ins uc o s,
ollowing a lack o skills on how o enhance s uden s eading compe ences (ibid.).
On op o ha , AI s esses his weakness eno mously. While AI ools a e widely
in oduced as eading, w i ing, and esea ching assis ance (e.g. KI-Campus, 2025),
wa nings on indi idual and socie al deskilling when elying on AI ools in highe
educa ion we e exp essed (e.g. Reinmann, 2023) and cu icula app oaches on
dealing wi h AI a e jus g owing (Tillmanns e al, 2025).
A he ou se o esea ch, when s uden s amilia ize wi h li e a u e o he ield and
iden i y ele an li e a u e, AI-based sea ch ools come in o play. Highligh ing one,
Seman ic Schola is a la ge sea ch engine ha inco po a ed au oma ically gene a ed
hype -sho , single-sen ence pape summa ies called TL;DR (abb e ia ion o “ oo
long; didn’ ead”) on op o abs ac s o Millions o esea ch a icles in 2020. They
ad e ise ha “[…] pa sing a long lis o pape s om a ious sou ces by eading
pape abs ac s is ime-consuming. TL;DRs help use s make quick in o med
decisions abou which pape s a e ele an , and whe e o in es he ime in u he
eading.“ (Seman ic Schola , 2025). Chou (2025) a gues ha “TL;DR” s ands o a
shi in how we consume nowadays in o ma ion caused by a collec i e hi s o quick
akeaways. The au ho illus a es bo h po en ials (e.g. assessabili y, ime sa ing,
be e unde s anding, ole in decision making) and isks (o e simpli ica ion & loss o
nuance, dange s o (con ex ual) misunde s anding, misin o ma ion, accu acy) in
using TL;DRs.
Impo an ly, in he Reading-W i ing-Resea ching-C i ical Thinking app oach, eading
seems o be a a he una en i e and unde in es iga ed aspec , especially when i
comes o e lec ions on compe ences and p ac ices ela ed o he supe ision o
s uden s’ inal heses and o speci ic AI unc ionali ies like hype -sho , single-
sen ence pape summa ies (abb e ia ed he e as SSPS). Ou con ibu ion
emphasises on supe iso s‘ e lec ion on guiding s uden s’ scien i ic eading
compe ences in inal heses in he age o AI. Hence, we ask: How can academics
suppo s uden s’ eading compe ences when supe ising hem in hei inal heses
in he age o AI? To add ess his ques ion, we desc ibe he con ex and p ac ical
e lec ion (chap e 2). Then, we highligh supe iso s’ e lec ion on hei own p ac ice
ela ed o eading o scien i ic a icles and hei AI usage (chap e 3.1), hei
e lec ion on hei SSPS exe cise using one AI ool (chap e 3.2), and hei
supe ision p ac ices along wi h po en ial compe ence ise and loss (chap e 3.3,
3.4). Finally, we conclude ou indings and sha e ideas o he u u e (chap e 4).
2 CONTEXT AND PRACTICAL REFLECTION
This s udy ook place a Hambu g Uni e si y o Technology, cha ac e ized among
o he s by ha ing implemen ed a s ong esea ch-based lea ning app oach in cou ses
(Bulmann e al., 2020), o e s a lexible eaching aining p og am (Bulmann &
Podleschny, 2023), has de eloped guidelines o AI in eaching and lea ning
(Baumhaue e al., 2024), and p o ides access o a HAWKI-based La ge Language
Model (LLM). In he beginning o 2025, a specialized wo kshop on “Accompanying
s uden s as co- esea che s in inal heses” has been conduc ed he e o 17 mid-
le el academics (unde aking hei PhD, ha ing eaching du ies in STEM and ha ing
di e en p io expe ise in supe ising s uden s in hei inal heses). The wo kshop
aimed a pa icipan s being able o discuss speci ic opics in supe ising heses
based on he pa icipan s’ own e lec ions. The wo kshop has been designed as
lipped class oom: Be o e he pa icipan s en e he class oom o discuss a ious
opics (4.5 hou s), hey wo ked indi idually in sel -s udy ime (1.5 hou s) on speci ic
opics (one olun a y and one obliga o y opic).
The indi idual e lec ion on dealing wi h AI in supe ising s uden s’ inal heses
(obliga o y opic o all) was encou aged by using a sel -designed ques ionnai e o be
anonymously esponded. The su ey design based on he idea o he in e wined
connec ion o eading, w i ing, esea ching, and c i ically e lec ing and was inspi ed
by TL;DRs. On he one hand, he su ey con ained closed ques ions o be answe ed
(mos ly) on a 4-poin scale, which hen we e analysed using desc ip i e s a is ics. On
he o he hand, he su ey con ained ee ex ques ions, which we e ca ego ized by
con en analyses. Impo an ly, his e lec ion ia su ey included h ee pa s: (1)
supe iso s’ e lec ion on hei own eading p ac ices, (2) an exe cise o w i e and
gene a e a SSPS (using he uni e si y’s HAWKI-based LLM) and e alua e i as well
as (3) supe iso s’ e lec ion o hei guiding p ac ice ela ed o s uden s’ scien i ic
eading wi h and wi hou AI ools. 17 pa icipan s esponded in his su ey (see
esul s sec ions 3 in oduc ion, 3.1, 3.2, 3.3). Based on ha e lec ion on AI, all
pa icipan s discussed in he class oom, i s he ise and all o eading compe ences
wi h/ wi hou AI (conduc ed in pai s), and second collec ed ideas on Do’s and Don’ s
o suppo ing s uden s’ eading compe ences in he age o AI when supe ising
s uden s’ inal heses (conduc ed in a Wo ld-Ca é) (see esul s sec ion 3.4).
3 RESULTS AND INSIGHTS
The esul s o he su ey show ha on a e age, supe iso s he e use AI ools o
ac i i ies ela ed o w i ing, e ising and ansla ing ex s o publica ions on a egula
basis. Compa ed o ha , AI ools a e a ely (i a all) used o ac i i ies ela ed o
li e a u e esea ch and analysis, da a analysis and isualisa ion, esul s in e p e a ion
and ga he ing eedback, wi h lowes alues o eading, da a collec ion and p ojec
managemen . Howe e , in some i ems, he s anda d de ia ion is a he high, e.g. o
li e a u e sea ch o eading. O e all, his indica es ha supe iso s’ usages o AI
ools along he esea ch p ocess emphasises on w i ing a he han on li e a u e
sea ch o eading, while hei use equency a ies among indi iduals. Only wo
supe iso s use AI ools in li e a u e esea ch, analysis and eading a icles as
s anda d in day- o-day esea ch.
3.1 Supe iso s’ e lec ion on hei eading o scien i ic a icles
Fi s , supe iso s we e asked o espond on hei a i ude and ime spen o eading
scien i ic li e a u e. Mos supe iso s ag ee on enjoying eading scien i ic a icles.
Howe e , hey disag ee on ha ing enough ime o spend on eading scien i ic
a icles. Addi ionally, mos o hem ag ee on su e ing an in o ma ion o e load.
Supe iso s’ eading p ac ices o scien i ic a icles
Being asked abou hei eading p ac ices o scien i ic a icles mos supe iso s
desc ibe o ien a ional eading s a egies, like scanning o skimming. O en, hey i s
ead he abs ac and based on wha hey ha e ead decide on how o con inue.
A e eading he abs ac hey o en con inue skimming he in oduc ion and esul s
sec ion, and some imes he discussion. Also, hey skim isual da a like igu es and
ables as hey o en summa ize essen ial indings. While many supe iso s seem o
apply hese global eading s a egies o di ec hei ime and a en ion o wha is mos
ele an , e y ew end o ead he ex as a whole in o de o ge deep knowledge o
how o o ganise a pape and o no miss any de ails. Addi ionally, some supe iso s
explici ly desc ibe de ining a conc e e pu pose o eading be o e s a ing and ha
hei eading p ac ice depends on hei own p og ess in he esea ch p ocess.
Supe iso s’ knowledge o eading s a egies
Mos supe iso s seem o be awa e o eading s a egies and o ha e explici
knowledge abou hose s a egies. They name, o example, global s a egies like
“o ien a ional eading”, “scanning”, “ eading ac oss”, “selec i e eading” esp.
“skipping” as well as s a egies o a close eading, like “deep eading”, “in ensi e
eading” o he in-dep h “SQ3R me hod” (Su ey-Ques ion-Read-Reci e-Re iew).
Supe iso s’ usage o AI o hei eading
Being asked abou hei usage o AI when eading scien i ic pape s, supe iso s
s a ed o a he no o a ely use AI o eading, while only wo supe iso s
esponded ha hey used AI in hei eading as a s anda d, as shown in Figu e 1.
Fig. 1 Su ey esponses on “I use AI o eading” (1: no a all, 2: a ely, 3: egula ly, 4: as a s anda d, n=numbe
o esponden s, a =a i hme ic mean, md=median, de =s anda d de ia ion)
I supe iso s use AI in hei eading p ac ice, hey p ima ily do so o cla i y ques ions
o o ansla e unclea con en o he pape . A ew supe iso s use AI o
summa izing – howe e p ima ily o websi e con en and o he ex s a he han o
scien i ic pape s. One supe iso epo ed o use “cha wi h pd ” unc ionali ies o ge
a deepe unde s anding o he con en . O e all, ou indings indica e ha he
supe iso s hemsel es enjoy eading scien i ic a icles bu ha e limi ed ime and
ace in o ma ion o e load. Applying hei knowledge o eading s a egies, hey s a
by eading abs ac s, which is key, and ba ely u ilize AI ools. On he one hand, his
unde pins he si ua ion why SSPS unc ionali ies ha e been de eloped and
implemen ed in he i s place (see TL;DR o Sema ic Schola ) and aligns wi h some
po en ials seen in SSPS (see Chou, 2024), bu con adic s on he o he hand he
ac ual eading p ac ices being o emos adi ional, lea ing ou AI ools.
3.2 Supe iso s’ e lec ion o he SSPS exe cise
Speci ically, being asked abou hei ecen expe ience wi h SSPSs, mos
supe iso s epo ed o ba ely, i a all ead SSPSs in he con ex o esea ch. Only
one pe son s a ed o ead SSPSs egula ly and h ee pe sons on a s anda d basis.
In e es ingly, hei posi ion di e ed on he ques ion whe he au oma ically gene a ed
SSPSs can help esea che s decide which a icles o pu on a eading lis . Again,
his migh be ela ed o he impo ance he abs ac plays in hei eading p ac ice
and also he p esumed po en ial o SSPSs o making decisions on u he eadings
as highligh ed by Seman ic Schola (2025) and Chou (2024). Addi ionally, mos
supe iso s s a ed no o w i e SSPSs hemsel es no gene a e SSPSs using AI in
he con ex o hei esea ch. This leads us o he assump ion ha SSPSs a e o da e
less impo an o he supe iso s, espec i ely, hey did no expe ience SSPS ( oo
much) un il being aced wi h an assignmen in his wo kshop e lec ion.
Supe iso s’ e alua ion o sel -w i en s. AI-gene a ed SSPS
Compa ing he AI-gene a ed SSPS wi h he sel -w i en SSPS one esponden
s a ed: “AI-gene a ed SSPS accu a ely e lec s he o iginal abs ac s’ essen ial
poin s, main aining he co e indings and po en ial applica ions while o e ing a b ie
o e iew sui able o eade s seeking a quick unde s anding o he s udy’s ou come
and signi icance.” Howe e , he supe iso s’ e alua ion o he AI-gene a ed SSPS
compa ed o hei sel -w i en SSPS d aws a e y con adic o y pic u e in a ious
co e aspec s: One supe iso ound ha he gene a ed SSPS ma ched mo e wi h he
s a e-o - he-a , while wo o he s missing he o e a ching iew o he con ex .
Impo an ly, some supe iso s a gued ha appa en ly in he gene a ed SSPS, he
ocus o he wo k was “misunde s ood” o desc ibed as ac s ins ead o y ou s o
e en en ail alse desc ip ions ha could lead o has y o misleading conclusions.
Some supe iso s desc ibed he gene a ed SSPS o con ain mo e de ails, appealing
smoo he and mo e unde s andable, o he s epo ed on a ague, less in o ma i e
gene a ed ex wi h “wei d” e minology. This s ongly poin s ou ha – only in a sum
o e all indi iduals – po en ials, bu also se ious isks o gene a ed SSPSs ha e
been expe ienced he e as desc ibed o TL;DRs by Chou (2024).
Supe iso s’ e lec ion on hei compe ence aining in his SSPS exe cise
Supe iso s e lec ed ha hey ained a ious compe ences by w i ing a SSPS
hemsel es: Selec ion and p io iza ion o key aspec s o he abs ac o pape as well
as ex educ ion skills. This goes along wi h a speci ic language s yle (like sho ,
clea -s uc u ed sen ences, p ecise choice o wo ds) and o adop he eade s’
pe pec i e. Addi ionally, hey lis ed he e also c ea i i y, memo y aining,
concen a ion and c i ical syn hesis skills. Howe e , ‘’Fi s and o emos , i 's
impo an o ho oughly unde s and he con en ”, as one esponden summa ized.
Being asked o e lec on hei exe cise o gene a e a SSPS, hey highligh ed a ious
skills o be ained: Ge ing o know he uni e si y’s HAWKI-based LLM, inc easing
p omp ing skills (like inco po a ing he a ge g oup and con ex ), c i ical eading,
analysis and e alua ion skills. They also unde lined o ha e e lec ed hei own
w i ing skills while compa ing wi h he gene a ed SSPS (like simpli ying echnical
e ms, using ac ion-o ien ed pha asing, w i ing a concise sen ence and a oid
unnecessa y speci ics) o make he SSPS mo e appealing, less o e whelming o
eade s and hus o make i mo e accessable o a wide ange o eade s.
Summa izing his sho exe cise, supe iso s expe ienced mul i ace ed compe ences
o be enhanced ha ela e o w i ing, eading and AI li e acy and beyond.
3.3 Supe iso s’ e lec ions on hei p ac ices o guiding s uden s’ inal heses
Supe iso s’ ecen p ac ices
Supe iso s’ p ac ices o dealing wi h he opic o AI in he p ocess o supe ision
e y much di e , as shown in Figu e 2: Two supe iso s add ess he opic as a
s anda d o all o hei s uden s when supe ising inal heses, howe e , many do no
a all. Two supe iso s add ess he opic occasionally o indi idual s uden s and ou
supe iso s occasionally o all s uden s.
Figu e 2: Su ey esponses on “I al eady add ess he use o AI by s uden s in he supe ision o heses.” (1: no
a all, 2: occasionally o some s uden s, 3: occasionally o all s uden s, 4: as a s anda d o indi idual s uden s,
5: as s anda d o all s uden s, n=numbe o esponden s, a =a i hme ic mean, md=median, de =s anda d
de ia ion)
I supe iso s add ess he opic o AI, hey usually do so in ace- o- ace mee ings,
discussing ad an ages, po en ials and isks o AI use, like nega i e e ec s on
compe ences and c ea i i y. One supe iso s a ed o encou age s uden s o use AI
ools in he p ocess o hesis w i ing (e.g. uni e si y’s HAWKI-based LLM, Cha GPT,
deeplW i e, Sci e.AI), bu also emphasized he impo ance o e lec ing on AI use
and o guiding s uden s in his ega d (e.g. discussing limi a ions like hallucina ions in
pe sonal mee ings and emphasizing he co ec documen a ion). Wi h ega d o good
scien i ic p ac ice, one supe iso e e ed o he Uni e si y’s Guidelines on AI usage.
Ano he supe iso explained ha hey elabo a ed on when o use AI in he w i ing
p ocess oge he wi h he s uden s. Fu he , some supe iso s sha ed hei pe sonal
expe iences wi h he use o AI ools wi h hei s uden s, e.g. ecommenda ion o also
ead o iginal pape when gene a ing a li e a u e e iew. One supe iso sha ed Do’s
and Don’ s: AI use possible o language s yle, g amma , and spell check, bu no o
con en c ea ion. In e es ingly, he e he da a shows di e en a i udes o supe iso s.
While many supe iso s a he y o p e en s uden s om unc i ical und ingenuous
use o AI ools, one supe iso p io i izes unde s anding me hods o he li e a u e.
The pe son s a ed ha o eading pape s, he s uden s could use wha e e hey
wan ed o - as long as hey unde s ood he me hods behind. Summa izing, he
supe iso s ha e ei he neglec ed o minimally add essed s uden s' use o AI in hei
inal heses. Bu when hey do so, hey ocus on discussing echnical aspec s,
po en ial bene i s, isks, anspa ency, hin o he uni e si y's guidelines and
elabo a e AI usage oge he wi h s uden s.
Supe iso s’ a emp s o u u e p ac ices
Supe iso s ha e been asked abou hei u u e p ac ices on s uden s scien i ic
eading skills wi h ega d o AI usage. The common essence among he esponden s
is ‘’Be c i ical’’. Supe iso s emphasized o suppo s uden s o use AI ools c i ically;
hey wan o enhance s uden s’ analy ical and c i ical hinking o ques ion and
e alua e AI ou pu . Despi e he use o AI (e.g. au oma ed summa ies), s uden s
should be able o de elop a deepe unde s anding o he ex . Hence, supe iso s
wan o p omo e s uden s’ abili y o independen ly and ho oughly ead scien i ic
pape s. AI is a he seen as an assis ance bu does no do he hinking o you.
Some supe iso s ecommended ha AI could be used o inding li e a u e,
howe e , ha s uden s needed o be awa e o inco ec summa ies o con en .
O he s s a ed ha AI ools coud be use ul o summa ize esul s, especially om a
la ge collec ion o pape s, howe e , i was s ic ly necessa y o check he e e ences
c i ically. Unde s anding AI ools and using hem in a a ge ed manne seems o be
e y impo an o he supe iso s. In an essence, hey aim o suppo he
de elopmen o c i ical hinking, e alua ion o AI ou pu s, enhance knowledge o AI
applica ions, and os e a deepe unde s anding o scien i ic li e a u e.
3.4 Supe iso s’ joined e lec ions on compe ence- ela ed guiding p ac ices
Gained o los eading compe ences when using o no using AI ools
As a i s s ep in he p esence wo kshop, supe iso s collec ed in pai s s uden s’
compe ences ha hey p esumed o be po en ially gained o loosed when using o
no using AI o eading wi hin inal heses. Supe iso s an icipa ed bo h enhanced
compe ence de elopmen and po en ial deskilling in a eas such as eading li e acy,
w i ing skills, esea ch abili ies, c i ical hinking, and AI li e acy, depending on he use
o non-use o AI in hese con ex s, as shown in Table 1. Impo an ly, hey an icipa ed
especially deskilling in eading compe ences when using AI ools ha a e add essed
in adi ional eading s a egies (compa e sec ions (2) and (3)) as well as missing
chances o de elop AI li e acy when jus elying on adi ional eading s a egies
(sec ion (4)). In u n, hey expec enhanced compe ence de elopmen when using
bo h adi ional eading s a egies as well as AI ools which s ongly indica es ha a
combined usage migh be p omising. No ably, such enhanced compe ence
de elopmen is no only impo an o p omo e a new gene a ion o esea che s/
academics bu also o ace he u gen ma e o deskilling, especially since inal
heses a e he las possibili y o os e hose compe ences.
Table 1: Reading compe ences gained o loosed when using o no using AI ools discussed by supe iso s in
pai s (p esence wo kshop)
Compe-
ences
Using AI o eading
No using AI o eading
Inc eased
(1) AI li e acy (e.g. p omp ing, AI ool
applica ion), simpli ica ion skills, c i ical
hinking (e.g. c i ical e lec ion, skep icism
owa ds AI con en ) and e iciency (pace)
(2) Pa ience, ex comp ehension (e.g.
abili y o make in e connec ions, deep
unde s anding, lea n echnical e ms),
eading skills (e.g. eading echniques),
w i ing compe ences ( ecognising
e minology, con en ions, w i ing s yles)
Dec eased
(3) C i ical hinking (e.g. c i ical eading,
hinking on me hods, esul s, de ails,
e lec ion/ e alua ion on ou pu ), eading
compe ences ( ocus on some hing, ead as ,
ead and comp ehend long academic ex s,
selec app op ia e li e a u e, unde s and
connec ions and he b oade con ex ),
u he mo e academic w i ing compe ences
(4) AI li e acy (e.g. p omp ing skills, AI ool
applica ion), c i ical e lec ion as well as
ex comp ehension (unce ain ies,
ambigui y and deep unde s anding)
Supe iso s’ collec ed Do’s and Don’ s o acili a e eading compe ences
As he second s ep wi hin he p esence wo kshop, supe iso s collec ed ia Wo ld-
Ca e ‘’Good’’ and ‘’Bad’’ P ac ices in accompanying s uden s in inal heses owa ds
ad anced eading in imes o AI (see Table 2). Those ela e o app op ia e a i udes,
knowledge, skills and pu poses o AI usage. This may be sui able as a sho -lis o
clea ecommended Do’s and Don’ s ha supe iso s can discuss wi h s uden s.
Table 2: Do's and don' s o using AI o eading in inal heses, collec ed by supe iso s (p esence wo kshop)
Ca ego y
Good P ac ices (Do’s)
Bad P ac ices (Don’ s)
A i ude
be ca e ul, alida e you sel , use AI
consciously
us e e y hing; comple ely igno e AI
usage, ad ances, capabili ies and limi s
Knowledge
and Skills
ge you sel in o med, aqui e p omp ing
skills, y i o make own expe iences
neglec de ailed eading by you sel , miss
ou e i ica ion, s ay up- o-da e
AI usage
as s a ing poin o li e a u e o e iew, o
simpli y new opics, only when needed
use AI as he one and only ool, igno e
anspa ancy
4 CONCLUSIONS AND IMPLICATIONS
Seman ic Schola s’ TL;DRs inspi ed us o in es iga e he ecen ly qui e una en i e
ques ion on ‘How can academics suppo s uden s’ eading compe ences when
supe ising hem in hei inal heses in he age o AI?’ The e o e, we conduc ed a
pedagogical wo kshop ha combined an indi idual e lec ion wi h a ques ionnai e in
a sel -s udy ime as well as discussions in p esence. Ou indings, al hough limi ed,
a e pa ly consis en wi h he in oduced li e a u e, bu pa ly con adic i . We ound
in he su ey ha he supe iso s enjoy eading scien i ic a icles bu ha e limi ed
ime, leading hem o ely on eading s a egies, s a eading abs ac s, bu ba ely
use AI ools he e. Be o e he wo kshop uni , hey a e la gely unawa e o SSPS and
ind hem mainly inadequa e o e ac o assessing a icle ele ance. A e he
exe cise on w i ing, gene a ing and e alua ing a SSPS based on hei own abs ac s,
he supe iso s ha e obse ed bo h p omising and isky quali y in he SSPSs and
s a ed o ha e ained di e en compe ences in ega d o eading and AI li e acy.
A e ui ul discussions in he wo kshop, hey ha e c a ed p ac icable guidelines o
an app op ia e use o AI ools in supe ising s uden s’ inal heses, an icipa ing bo h
enhanced compe ence de elopmen and po en ial deskilling ela ed o eading and
AI li e acy as well as esea ch compe ences.
Re lec ing ou me hodology, we highligh ha he supe iso s app ecia ed o ake
ime o e lec on AI, ge o know he uni e si y’s HAWKI-based LLM and
expe ienced a combina ion o e lec ion, exe cise and ans e wi hin he su ey, bu
will be imp o ed. We conclude ha his h ee- old indi idual e lec ion ia su ey
leads o in-dep hs insigh s and is a p e equisi e o pee discussions and in u n o
de i e guidelines o how supe iso s can suppo s uden s’ eading compe ences in
he age o AI. The e o e, we ecommend ha cu en o ma s o eache aining in
highe educa ion add essing opics in hesis supe ision comp ise ime slo s o
indi idual e lec ion and p ac ical explo a ion, bes wi h a uni e si y’s AI ool. In he
u u e, he exe cise on SSPSs can be exchanged by o he AI ools o unc ionali ies
p omising o enease scien i ic eading (like ‘’cha wi h .pd ’’) o o he esea ch
ac i i ies. Also, his h ee- old e lec ion and discussion me hods will be o e ed o
speci y he uni e si y’s guidelines on AI o supe ision p ac ices in single ins i u es.
Finally, we a gue ha i is key nowadays o de elop c i ical and AI li e acy along he
cu iculum in lea ning ac i i ies ha in e win eading, w i ing, esea ching and c i ical
hinking bo h wi h and wi hou AI ools guided by e lec ed academics.
