Pith. sign in

REVIEW 4 major objections 5 minor 81 references

PAPPL: Personalized AI-Powered Progressive Learning Platform

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read PAPPL argues that AI hints that become progressively more specific, built from each student's recurring misconceptions, produce deeper learning in unfamiliar STEM topics.

desk verdict A coherent LLM-tutoring system design whose evaluation pilot is too weak for its learning claims, but it deserves a serious referee. read the letter →

arxiv 2508.14109 v1 pith:E2QJ3IXW submitted 2025-08-18 cs.CY cs.AIcs.ET

Shayan Bafandkar , Sungyong Chung , Homa Khosravian , Alireza Talebpour This is my paper
classification cs.CYcs.AIcs.ET
keywords intelligenttutoringsystemspersonalizedlearninglargelanguagemodelsGPT-4oprogressivefeedbackengineeringeducationlearner-statemodelingAIhints
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

PAPPL is an intelligent tutoring system for engineering education that records every student attempt, detects recurring misconceptions, and uses GPT-4o to generate hints that become progressively more specific as errors persist. The paper's central empirical claim is that this history-aware feedback changes how students learn an unfamiliar STEM subject. In a case study with graduate students answering pavement-engineering questions, the group receiving PAPPL hints needed fewer attempts per question, spent more time per question, and improved from 30% first-attempt success to 55% second-attempt success, while the no-hint baseline rose only from 30% to 40%. The authors interpret more time, fewer attempts, and the larger second-attempt jump as evidence of a deeper and more effective learning process, with questionnaire scores above 5 on a 10-point scale across all five user-experience dimensions. The proposal matters because it offers a scalable, LLM-driven route to personalized tutoring across subjects without hand-built cognitive models per topic.

What carries the argument

The central mechanism is the multi-layered feedback core, built from the Learner-State Analyzer and the AI Hint Engine. The analyzer continuously aggregates each student's attempts, timestamps, correctness flags, and prior hint text into an evolving learner profile that detects recurring misconceptions. The hint engine then constructs a context-rich prompt from that profile, the problem statement, and instructor notes, and calls GPT-4o with a deliberately low temperature so hints become progressively more specific while never disclosing the solution. This profile-to-prompt loop is what carries the paper's claim of adaptive, personalized, history-aware feedback.

What would settle it

Run both groups through a delayed, no-hint post-test of parallel pavement-engineering items two weeks after the session. If the PAPPL group's advantage over the baseline group disappears, the measured gains were session practice rather than durable learning.

Watch

Extended reading notes

Core claim

The paper introduces PAPPL as an intelligent tutoring system that uniquely combines the four classical ITS modules—expert, student, tutor, and user interface—around a large language model. Its feedback core synthesizes the current question, instructor-provided explanations, the student's prior errors, and related question history into a structured prompt that forbids revealing the answer and uses a low sampling temperature to reduce hallucination. The claimed discovery is that this progressive, temporally-aware hint mechanism produces measurably better learning on an unfamiliar topic: the PAPPL group solved questions in fewer attempts, spent more time reflecting, and achieved a much larger s

Load-bearing premise

The study assumes the two groups were comparable in prior knowledge and motivation before the test and that fewer attempts with more time per question indicate learning rather than a different guessing or strategy profile.

Editorial extensions

If this is right

  • A successful PAPPL shows that personalized, progressive AI feedback can be deployed in graduate engineering courses without custom cognitive modeling per topic.
  • Instructor analytics would allow course design to react to class-level error patterns during a course rather than after exams.
  • The second-attempt gain implies hint-supported retries can be part of graded assessment without simply giving away answers.
  • Because the AI core is LLM-agnostic, the same platform could ride future model improvements and be extended to interpret diagrams and plots.
  • If the pattern scales, it offers engineering education a path to the kind of personalization already seen in K-12 and foundational-science tutoring.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the reported pattern would be more convincing with a delayed post-test: a two-week, no-hint follow-up would tell whether the observed gains are durable knowledge or in-session task familiarity.
  • A reader might also expect the second-attempt jump to transfer to reworded, isomorphic problems; testing that would separate conceptual learning from learning the specific question wording.
  • The platform's personalization depends on accumulating a learner profile, so a cold-start problem is implied: the first attempt or two cannot be history-aware, and a bootstrap rule or generic Socratic prompt would be needed.
  • The survey's Accuracy dimension scored lowest (5.8/10), consistent with open-ended complaints about repetitive or unhelpful hints for multiple-choice items; a testable extension would tune question-type-specific instructions and measure whether accuracy ratings improve as hints become less repetitive.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents PAPPL, an intelligent tutoring system for engineering education that integrates the classical ITS modules (expert, student, tutor, interface) around a GPT-4o core. The system records student attempts, detects recurring misconceptions, and generates progressively more specific, context-aware hints. The authors report a case study with graduate students from UIUC and GWU in which one group used PAPPL and received AI-generated feedback while a baseline group completed the same 25 pavement-engineering questions without feedback. They report that the PAPPL group needed fewer attempts per question, spent more time per question, showed a larger first-to-second-attempt success improvement, and gave positive questionnaire ratings. From these observations they conclude that PAPPL's progressive hints lead to a deeper and more effective learning process. The paper also describes the platform architecture, instructor dashboard, and planned extensions.

Significance. If the empirical claims were well supported, PAPPL would be a notable step toward an integrated, LLM-based ITS that uses interaction history to generate progressive hints, a contribution relevant to the AI-in-education community. The architecture is clearly described, and the choice of GPT-4o with low-temperature prompting, instructor context, and misconception detection is sensible. However, the paper's central claim—that the system improves learning—rests entirely on a small, uncontrolled, under-reported case study. The authors themselves state in §4.2 that 'the dataset is not large enough for a definitive assessment.' Because the main contribution is evaluated only by this pilot study and the observed differences are not statistically validated, the significance of the paper as a research contribution is currently not established.

major comments (4)
  1. [§4.1 and §4.2] The central causal claim in §4.2—that the combination of increased time, fewer attempts, and improved second-attempt success 'suggests a deeper and more effective learning process'—is not supported by the reported study design. No sample sizes are given, no randomization procedure is described, no pre-test is administered, and no inferential statistics (standard errors, confidence intervals, or hypothesis tests) are reported. The percentages '30%, 55%, 40%' are aggregate point estimates; with a small sample, chance could easily produce the observed differences. The paper's own admission in §4.2 that the dataset is not large enough for a definitive assessment undercuts the abstract's and conclusion's stronger claims. This is a load-bearing problem because the educational-effectiveness claim is the paper's primary empirical result.
  2. [§3.2.3 and §4.1] The experimental contrast conflates receiving any AI-generated hints with receiving PAPPL's progressive, history-based hints. The PAPPL group received GPT-4o feedback, while the baseline group 'completed the same assessment without receiving any hints or feedback.' Therefore any observed difference could be due to the mere availability of hints, the additional time-on-task, or motivational effects, rather than to the progressive personalization mechanism that the paper claims as novel (contribution 1). A static-hint control condition or a condition with non-progressive hints would be needed to isolate the mechanism. Without such a control, the data cannot distinguish PAPPL's claimed innovation from generic hint provision.
  3. [§4.2] The outcome measures used to infer 'deeper learning' are process and self-report metrics: number of attempts, time per question, second-attempt success, and a Likert questionnaire. None of these directly measures learning gain; there is no pre/post content test or retention measure. The time-on-task difference is especially ambiguous: more time may reflect reflection, as the authors suggest, but it may also reflect confusion or poorly adapted hints. The qualitative responses in §4.2 even indicate that some students found hints repetitive or unhelpful for multiple-choice questions. The claim that the observed pattern uniquely supports the progressive-learning explanation is therefore not justified.
  4. [Table 2 and §4.2] The questionnaire analysis lacks essential psychometric and reporting details. Items are described as using a 5-point Likert scale, yet Figure 6 reports scores 'rescaled to a 10-point scale' without specifying the transformation, and no response distributions, standard deviations, or sample size are given. The duplicate-consistency check is mentioned but no numerical results are reported beyond a passing statement. Because the questionnaire is the only direct evidence for perceived effectiveness, engagement, and adaptivity, this lack of detail further weakens the empirical support.
minor comments (5)
  1. [Throughout] The text contains numerous spacing/formatting errors from the PDF conversion (e.g., 'sampleandsingle-sessiondesignoffervaluableearlyinsights', missing spaces between words). A careful copy edit is needed.
  2. [§4.2, Figure 5] The y-axis of Figure 5 is labeled 'Average time spent per question' but no units are given; specify seconds or minutes. Also clarify whether the average is over questions or over participants.
  3. [§3.2.3] The paper states that low temperature is used to mitigate hallucination, but no evaluation of hint correctness is provided. The Accuracy dimension in the questionnaire is self-reported; a manual or automated correctness audit of a sample of hints would strengthen the system description.
  4. [§4.1] The study involved human participants but no IRB approval, consent, or anonymization details are reported. This is standard for education research and should be included.
  5. [Abstract/Conclusion] The abstract and conclusion state that PAPPL 'significantly enhances' learning, while §4.2 explicitly cautions that the dataset is not large enough for a definitive assessment. The conclusions should be calibrated to the evidence presented.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's empirical claims are measured outcomes with experimental-design caveats, not derivations that reduce to their inputs.

full rationale

PAPPL is a systems-and-case-study paper. Its central claim—that the PAPPL group's increased time, fewer attempts, and higher second-attempt success suggest a deeper learning process—rests on observed aggregate statistics, not on a derivation whose output equals an input. No parameter is fitted to a subset of data and then reported as a prediction: the attempt counts, success rates, and questionnaire scores are all measured outcomes. The adaptive/progressive mechanism is a design feature of the system, and the paper's evaluation does not use that mechanism's definition to compute the reported results. There is no load-bearing self-citation: the references to ITS frameworks and GPT-4o capabilities are external background, and no claimed 'uniqueness theorem' or prior author result is invoked to force a choice. The paper even includes a limitation statement ('the dataset is not large enough for a definitive assessment'), further distancing the claims from forced conclusions. The genuine concerns here are experimental-design issues—no pre-test, no randomization, small sample, no inferential statistics, and the confound between receiving any hints and receiving progressive hints—but those concern causal identifiability, not circularity. Applying the hard rule that circularity requires a specific reduction (e.g., Eq. X = Eq. Y by construction, or a fitted parameter renamed as a prediction), no such reduction is present. Therefore the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 1 invented entities

The central claim rests on assumptions about LLM reliability, group comparability, and the validity of self-report and attempt proxies. There are no explicit fitted constants, but the system has implicit hand-chosen parameters (temperature, misconception criterion). The platform itself is the sole invented entity, with no independent evidence.

free parameters (2)
  • LLM sampling temperature
    The AI Hint Engine sets GPT-4o temperature 'deliberately low' to reduce hallucination (Section 3.2.3). The exact value is not reported, and it controls the variability of generated hints, an implicit hand-chosen parameter.
  • Misconception detection criterion
    The Learner-State Analyzer 'detects recurring misconceptions' (Section 3.2.2) but no operational threshold (e.g., number of repeated errors, similarity score) is specified, leaving an implicit modeling choice that affects when hints become more specific.
assumptions (4)
  • domain assumption GPT-4o produces pedagogically sound, context-sensitive hints when given structured prompts and a low temperature.
    Used in Section 3.2.3 and Background (Section 2) as the basis of the tutor module; cited literature (Tithi et al., Levonian et al.) warns that LLM hints can hallucinate, so this is an unproven assumption for this system.
  • domain assumption Self-reported Likert questionnaire responses are a valid measure of learning effectiveness.
    Section 4.2 uses questionnaire scores (all above 5 on a 10-point scale) as evidence of effectiveness; no objective pre/post learning measure is reported.
  • domain assumption Participants in both groups had near-zero background in pavement engineering, ensuring comparability.
    Section 4.1 asserts 'a topic with which the participants were unfamiliar' without a pre-test; group equivalence is assumed, not measured.
  • domain assumption Fewer attempts and more time spent per question indicate deeper learning.
    Section 4.2 interprets the attempt and time patterns as 'a deeper and more effective learning process'; these proxies are not validated against learning outcomes.
invented entities (1)
  • PAPPL platform
    purpose: A new ITS architecture integrating expert, student, tutor, and UI modules with an LLM-based progressive feedback core.
    The platform's effectiveness is evaluated only within this paper's case study; no external benchmark or falsifiable prediction independent of the paper is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PAPPL: Personalized AI-Powered Progressive Learning Platform." pith.science (2026). https://pith.science/paper/E2QJ3IXW

@misc{pith2026250814109,
  author       = {Pith},
  title        = {Pith review of: PAPPL: Personalized AI-Powered Progressive Learning Platform},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E2QJ3IXW}},
  note         = {Machine review of arXiv:2508.14109}
}
read the original abstract

Engineering education has historically been constrained by rigid, standardized frameworks, often neglecting students' diverse learning needs and interests. While significant advancements have been made in online and personalized education within K-12 and foundational sciences, engineering education at both undergraduate and graduate levels continues to lag in adopting similar innovations. Traditional evaluation methods, such as exams and homework assignments, frequently overlook individual student requirements, impeding personalized educational experiences. To address these limitations, this paper introduces the Personalized AI-Powered Progressive Learning (PAPPL) platform, an advanced Intelligent Tutoring System (ITS) designed specifically for engineering education. It highlights the development of a scalable, data-driven tutoring environment leveraging cutting-edge AI technology to enhance personalized learning across diverse academic disciplines, particularly in STEM fields. PAPPL integrates core ITS components including the expert module, student module, tutor module, and user interface, and utilizes GPT-4o, a sophisticated large language model (LLM), to deliver context-sensitive and pedagogically sound hints based on students' interactions. The system uniquely records student attempts, detects recurring misconceptions, and generates progressively targeted feedback, providing personalized assistance that adapts dynamically to each student's learning profile. Additionally, PAPPL offers instructors detailed analytics, empowering evidence-based adjustments to teaching strategies. This study provides a fundamental framework for the progression of Generative ITSs scalable to all education levels, delivering important perspectives on personalized progressive learning and the wider possibilities of Generative AI in the field of education.

Figures

Figures reproduced from arXiv: 2508.14109 by the authors.

Figure 1
Figure 1. Component Design for PAPPL Together, these components create a responsive environment where students receive context-aware hints while instructors gain actionable insights into class progress. 3.1. User Interface Layer The platform provides a comprehensive and intuitive web portal for both learners and instructors. Students can browse registered courses, attempt questions, and immediately receive AI-generated hints … view at source ↗
Figure 2
Figure 2. User Interface for the PAPPL: (a) Home Page, (b) Instructor Dashboard, (c) Question Defining [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. PAPPL in action within a STEM-designated field, providing feedback on a pavement-engineering [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Number of attempts made by each group during the case study. [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Average time spent per question [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Average user-experience scores across five dimensions (rescaled to a 10-point scale). Higher scores [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

81 extracted references · 70 canonical work pages

  1. [1]

    Barnes and J

    T. Barnes and J. Stamper. Automatic hint generation for logic proof tutoring using historical data.Educational Technology & Society, 13(1):3–12, 2010

    2010

  2. [2]

    Levonian, O

    Z. Levonian, O. Henkel, C. Li, and M. E. Postle. Designing safe and relevant gener- ative chats for math learning in intelligent tutoring systems.Journal of Educational Data Mining, 17(1), 2025

    work page 2025

  3. [3]

    D. A. Joyner and A. K. Goel. Improving inquiry-driven modeling in science educa- tion through interaction with intelligent tutoring agents. InProceedings of the 20th International Conference on Intelligent User Interfaces, pages 5–16. ACM, 2015

    work page 2015

  4. [4]

    Piech, M

    C. Piech, M. Sahami, J. Huang, and L. Guibas. Autonomously generating hints by inferring problem-solving policies. InProceedings of the Second ACM Conference on Learning@Scale, pages 195–204. ACM, 2015

    work page 2015

  5. [5]

    A. C. Graesser, X. Hu, B. D. Nye, K. VanLehn, R. Kumar, C. Heffernan, and V. Rus. Electronixtutor: An intelligent tutoring system with multiple learning resources for electronics. International Journal of STEM Education, 5:15, 2018

    work page 2018

  6. [6]

    Howdoesexamanxietyaffecttheperformanceofuniver- sity students?Mediterranean Journal of Social Sciences, 2(2):93–100, 2011

    A.TrifoniandM.Shahini. Howdoesexamanxietyaffecttheperformanceofuniver- sity students?Mediterranean Journal of Social Sciences, 2(2):93–100, 2011

    work page 2011

  7. [7]

    Grassi, A

    A. Grassi, A. Gaggioli, and G. Riva. New technologies to manage exam anxiety. In B. K. Wiederhold and G. Riva, editors,Annual Review of Cybertherapy and Telemedicine 2011, volume 167, pages 57–62. IOS Press, 2011. doi: 10.3233/ 978-1-60750-766-6-57. 16

    work page 2011

  8. [8]

    Exploringthestateoftheartinadaptivedistributed learning environments

    B.CiloglugilandM.M.Inceoglu. Exploringthestateoftheartinadaptivedistributed learning environments. InInternational Conference on Computational Science and Its Applications, pages 556–569. Springer, 2010

    work page 2010

Show all 81 references
  1. [9]

    Graesser, P

    A.C. Graesser, P. Chipman, B.C. Haynes, and A. Olney. Autotutor: an intelligent tutoring system with mixed-initiative dialogue.IEEE Transactions on Education, 48 (4):612–618, 2005. doi: 10.1109/TE.2005.856149

    2005

  2. [10]

    Vu, Robert Belfer, Joelle Pineau, Aaron Courville, Laurent Charlin, and Yoshua Bengio

    Iulian Vlad Serban, Varun Gupta, Ekaterina Kochmar, Dung D. Vu, Robert Belfer, Joelle Pineau, Aaron Courville, Laurent Charlin, and Yoshua Bengio. A large-scale, open-domain, mixed-interface dialogue-based its for stem. arXiv preprint arXiv:2005.06616, 2020. URL https://doi.or...

    2005 doi arXiv

  3. [11]

    Integrating technology and pedagogy for inquiry- based learning: The stanford mobile inquiry-based learning environment (smile)

    Elizabeth Buckner and Paul Kim. Integrating technology and pedagogy for inquiry- based learning: The stanford mobile inquiry-based learning environment (smile). Prospects, 44(1):99–118, 2014. doi: 10.1007/s11125-013-9269-7

    2014 doi

  4. [12]

    Designinglearningbyteachingagents: The betty’sbrainsystem

    KrittayaLeelawongandGautamBiswas. Designinglearningbyteachingagents: The betty’sbrainsystem. InternationalJournalofArtificialIntelligenceinEducation ,18 (3):181–208, 2008

    2008

  5. [13]

    Itssql: Intelligenttutoringsystemforsql,2023

    Sören Aguirre Reid, Frank Kammer, Johannes Kunz, Timon Pellekoorne, Markus Siepermann,andJonasWölfer. Itssql: Intelligenttutoringsystemforsql,2023. URL https://arxiv.org/abs/2311.10730

    2023 arXiv

  6. [14]

    Ma.Intelligent tutoring systems and learning outcomes: Two systematic reviews

    W. Ma.Intelligent tutoring systems and learning outcomes: Two systematic reviews. Doctoral dissertation, Faculty of Education, 2017

    2017

  7. [15]

    H. S. Nwana. Intelligent tutoring systems: An overview. Artificial Intelligence Review, 4(4):251–277, 1990. doi: 10.1007/bf00168958

    1990 doi

  8. [16]

    Advancing genera- tive intelligent tutoring systems with gpt-4: Design, evaluation, and a modular framework for future learning platforms

    Siyang Liu, Xiaorong Guo, Xiangen Hu, and Xin Zhao. Advancing genera- tive intelligent tutoring systems with gpt-4: Design, evaluation, and a modular framework for future learning platforms. Electronics, 13(24):4876, 2024. doi: 10.3390/electronics13244876

    2024 doi

  9. [17]

    B. Martin. Constraint-based modelling: Representing student knowledge. New Zealand Journal of Computing, 7(2):30–38, 1996. 17

    1996

  10. [18]

    Woolf, Esma Aïmeur, Roger Nkambou, and Susanne Lajoie, editors

    Beverley P. Woolf, Esma Aïmeur, Roger Nkambou, and Susanne Lajoie, editors. Intelligent Tutoring Systems: 9th International Conference on Intelligent Tutoring Systems,ITS2008,Montreal,Canada,June23–27,2008,Proceedings ,volume5091 of Lecture Notes in Computer Science, 2008. Spr...

    2008 doi

  11. [19]

    A. R. Salman. The use of intelligent tutoring system for developing web-based learningcommunities. InternationalJournalofComputerScienceIssues(IJCSI) ,10 (6):156, 2013

    2013

  12. [20]

    Sharma, S

    S. Sharma, S. Ghorpade, A. Sahni, and N. Saluja. Survey of intelligent tutoring systems: Areviewonthedevelopmentofexpert/intelligenttutoringsystems,various teaching strategies and expert tutoring system design suggestions. International Journal of Engineering Research & Techno...

    2014

  13. [21]

    J. J. Williams, J. Kim, A. Rafferty, S. Maldonado, K. Z. Gajos, W. S. Lasecki, and N. Heffernan. Axis: Generating explanations at scale with learnersourcing and machinelearning. InProceedingsoftheThirdACMConferenceonLearning@Scale , pages 379–388. ACM, 2016

    2016

  14. [22]

    Wumpusadvisor1: Afirstimplemen- tation program that tutors logical and probabilistic reasoning skills, 1980

    J.L.Stansfield,B.P.Carr,andI.P.Goldstein. Wumpusadvisor1: Afirstimplemen- tation program that tutors logical and probabilistic reasoning skills, 1980

    1980

  15. [23]

    Aneffectivemetacognitivestrategy: Learningby doing and explaining with a computer-based cognitive tutor.Cognitive Science, 26 (2):147–179, 2002

    V.A.AlevenandK.R.Koedinger. Aneffectivemetacognitivestrategy: Learningby doing and explaining with a computer-based cognitive tutor.Cognitive Science, 26 (2):147–179, 2002

    2002

  16. [24]

    Intelligenttutoring goes to school in the big city.International Journal of Artificial Intelligence in Education, 8:30–43, 1997

    K.R.Koedinger,J.R.Anderson,W.H.Hadley,andM.A.Mark. Intelligenttutoring goes to school in the big city.International Journal of Artificial Intelligence in Education, 8:30–43, 1997

    1997

  17. [25]

    C. R. Wolfe, C. L. Widmer, V. F. Reyna, X. Hu, E. M. Cedillos, C. R. Fisher, and A. M. Weil. The development and analysis of tutorial dialogues in autotutor lite. Behavior Research Methods, 45(3):623–636, 2013

    2013

  18. [26]

    Mitrovic, B

    A. Mitrovic, B. Martin, and P. Suraweera. Intelligent tutors for all: The constraint- based approach.IEEE Intelligent Systems, 22(4):38–45, 2007

    2007

  19. [27]

    Mitrovic

    A. Mitrovic. Fifteen years of constraint-based tutors: What we have achieved and where we are going.User Modeling and User-Adapted Interaction, 22(1–2):39–72, 2012. 18

    2012

  20. [28]

    VanLehn, C

    K. VanLehn, C. Lynch, K. Schulze, J. A. Shapiro, R. Shelby, L. Taylor, and M. Win- tersgill. The andes physics tutoring system: Lessons learned.International Journal of Artificial Intelligence in Education, 15(3):147–204, 2005

    2005

  21. [29]

    S. Ohlsson. Learning from performance errors.Psychological Review, 103(2):241– 262, 1996

    1996

  22. [30]

    Ottmar, D

    E. Ottmar, D. Landy, R. L. Goldstone, and E. Weitnauer. Getting from here to there!: Testingtheeffectivenessofaninteractivemathematicsinterventionembedding perceptual learning. InCogSci 2015, 2015

    2015

  23. [31]

    Mathematicsintelligenttutoringsystemswithhandwritteninput: ascoping review.Education and Information Technologies, 29(9):11183–11209, 2024

    L.Rodrigues,F.D.Pereira,M.Marinho,V.Macario,I.I.Bittencourt,S.Isotani,and R.Mello. Mathematicsintelligenttutoringsystemswithhandwritteninput: ascoping review.Education and Information Technologies, 29(9):11183–11209, 2024

    2024

  24. [32]

    Q. Brown. Mobile intelligent tutoring system: Moving intelligent tutoring systems off the desktop, 2006

    2006

  25. [33]

    Le Corre, C

    F. Le Corre, C. Fauvel, C. Hoareau, R. Querrec, and C. Buche. Chrysaor: An agent- basedintelligenttutoringsysteminvirtualenvironment. In InternationalConference on Virtual Learning, pages 39–45, 2012

    2012

  26. [34]

    Pérez-Ramírez, N

    M. Pérez-Ramírez, N. J. Ontiveros-Hernández, C. A. Ochoa-Ortíz, J. A. Hernández- Aguilar, and B. E. Zayas-Pérez. Intelligent tutoring systems based on virtual reality for the electrical domain.Research in Computing Science, 122:163–174, 2017

    2017

  27. [35]

    Sdmentor: Avirtual reality-based intelligent tutoring system for surgical decision making in dentistry

    N.Vannaprathip,P.Haddawy,H.Schultheis,andS.Suebnukarn. Sdmentor: Avirtual reality-based intelligent tutoring system for surgical decision making in dentistry. Artificial Intelligence in Medicine, 162:103092, 2025

    2025

  28. [36]

    H. Ateş. Integrating augmented reality into intelligent tutoring systems to enhance science education outcomes.Education and Information Technologies, 30(4):4435– 4470, 2025

    2025

  29. [37]

    N. J. Ahuja, S. Dutt, S. L. Choudhary, and M. Kumar. Intelligent tutoring system ineducationfordisabledlearnersusinghuman–computerinteractionandaugmented reality. International Journal of Human–Computer Interaction, 41(3):1804–1816, 2025

    2025

  30. [38]

    C. Conati. Intelligent tutoring systems: New challenges and directions. InIJCAI, pages 2–7, 2009. 19

    2009

  31. [39]

    Ritter, J

    S. Ritter, J. R. Anderson, K. R. Koedinger, and A. Corbett. Cognitive tutor: Applied researchinmathematicseducation. PsychonomicBulletin&Review ,14(2):249–255, 2007

    2007

  32. [40]

    Barnes and J

    T. Barnes and J. Stamper. Toward automatic hint generation for logic proof tutoring using historical student data. In International Conference on Intelligent Tutoring Systems, pages 373–382. Springer, 2008

    2008

  33. [41]

    Buildingadaptivetutoringmodelusingar- tificialneuralnetworksandreinforcementlearning

    G.Fenza,F.Orciuoli,andD.G.Sampson. Buildingadaptivetutoringmodelusingar- tificialneuralnetworksandreinforcementlearning. In 17thInternationalConference on Advanced Learning Technologies (ICALT), pages 460–462. IEEE, 2017

    2017

  34. [42]

    Y. Qin, S. Hu, Y. Lin, W. Chen, N. Ding, G. Cui, and M. Sun. Tool learning with foundation models. ACM Computing Surveys, 57(4):Article 101, 2025. doi: 10.1145/3704435

    2025 doi

  35. [43]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    AshishVaswani,NoamShazeer,NikiParmar,JakobUszkoreit,LlionJones,AidanN. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. InProceed- ings of the 31st International Conference on Neural Information Processing Systems (NeurIPS), pages 6000–6010. Curran Associ...

    2017

  36. [44]

    Akkaya, M

    I. Akkaya, M. Andrychowicz, M. Chociej, M. Litwin, B. McGrew, A. Petron, and L.Zhang. Solvingrubik’scubewitharobothand. arXivpreprintarXiv:1910.07113 , 2019

    1910 arXiv

  37. [45]

    V. Mnih, K. Kavukcuoglu, D. Silver, A. Graves, I. Antonoglou, D. Wierstra, and M. Riedmiller. Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602, 2013

    2013 arXiv

  38. [46]

    D. A. Pomerleau. Alvinn: An autonomous land vehicle in a neural network. In Advances in Neural Information Processing Systems, volume 1, 1988

    1988

  39. [47]

    Predicting learning performance with large language models: A study in adult literacy

    L.Zhang,J.Lin,C.Borchers,J.Sabatini,J.Hollander,M.Cao,andX.Hu. Predicting learning performance with large language models: A study in adult literacy. In R.A. Sottilare and J. Schwarz, editors,Proceedings of the Adaptive Instructional Systems, pages 333–353, Washington, DC, USA...

    2024

  40. [48]

    Combining large language models with tutoring system intelligence: A case study in caregiver homework support

    Devika Venugopalan, Ziwen Yan, Conrad Borchers, Jionghao Lin, and Vincent Aleven. Combining large language models with tutoring system intelligence: A case study in caregiver homework support. InProceedings of the 15th International 20 LearningAnalyticsandKnowledgeConference(L...

    2025

  41. [49]

    Canlargelanguagemodelsmatchtutoringsystem adaptivity? a benchmarking study, 2025

    ConradBorchersandTianzeShou. Canlargelanguagemodelsmatchtutoringsystem adaptivity? a benchmarking study, 2025. URLhttps://arxiv.org/abs/2504. 05570. Accepted as full paper to the 26th International Conference on Artificial Intelligence in Education (AIED 2025)

    2025

  42. [50]

    Llm intelligent agent tutoring in higher education courses using a rag approach

    HoriaModran,IoanaCorinaBogdan,DoruUrsut ,iu,CornelSamoila,andPaulLivius Modran. Llm intelligent agent tutoring in higher education courses using a rag approach. Preprints, July 2024. doi: 10.20944/preprints202407.0519.v1. URL https://doi.org/10.20944/preprints202407.0519.v1

    2024

  43. [51]

    Taylor, and Alan Tay

    Michael Guevarra, Indronil Bhattacharjee, Srijita Das, Christabel Wayllace, Car- rie Demmans Epp, Matthew E. Taylor, and Alan Tay. An llm-guided tutoring system for social skills training. InProceedings of the AAAI Conference on Arti- ficial Intelligence, volume 39, pages 29506...

    2025 doi

  44. [52]

    A systematic review of ai- drivenintelligenttutoringsystems(its)ink-12education

    AngéliqueLétourneau,MarionDeslandesMartineau,PatrickCharland,JohnAlexan- der Karran, Jared Boasen, and Pierre Majorique Léger. A systematic review of ai- drivenintelligenttutoringsystems(its)ink-12education. npjScienceofLearning ,10 (29), 2025. doi: 10.1038/s41539-025-00320-7....

    2025 doi

  45. [53]

    The promise and limits of llms in construct- ing proofs and hints for logic problems in intelligent tutoring systems

    Sutapa Dey Tithi, Arun Kumar Ramesh, Clara DiMarco, Xiaoyi Tian, Nazia Alam, Kimia Fazeli, and Tiffany Barnes. The promise and limits of llms in construct- ing proofs and hints for logic problems in intelligent tutoring systems. arXiv preprint arXiv:2505.04736, 2025. URL https...

    2025

  46. [54]

    Designing safe and relevant generative chats for math learning in intelligent tutoring systems

    Zachary Levonian, Owen Henkel, Chenglu Li, and Millie-Ellen Postle. Designing safe and relevant generative chats for math learning in intelligent tutoring systems. Journal of Educational Data Mining, 17(1), 2025. doi: 10.5281/zenodo.14751365. URL https://doi.org/10.5281/zenodo...

    2025 doi

  47. [55]

    Gpt-4technicalreport,2023

    OpenAI. Gpt-4technicalreport,2023. Retrievedfromhttps://openai.com/index/gpt- 4/. 21

    2023

  48. [56]

    Schmucker, M

    R. Schmucker, M. Xia, A. Azaria, and T. Mitchell. Ruffle & riley: Insights from designing and evaluating a large language model-based conversational tutoring sys- tem. In A.M. Olney, I.-A. Chounta, Z. Liu, O.C. Santos, and I.I. Bittencourt, editors, Proceedings of the Artifici...

  49. [57]

    Ahmed, K

    F. Ahmed, K. Shubeck, and X. Hu. Chatgpt in the generalized intelligent framework for tutoring. InProceedings of the 11th Annual Generalized Intelligent Framework for Tutoring (GIFT) Users Symposium (GIFTSym11), page 109, Orlando, FL, USA, 2023

    2023

  50. [58]

    Y. Dan, Z. Lei, Y. Gu, Y. Li, J. Yin, J. Lin, L. Ye, Z. Tie, Y. Zhou, Y. Wang, et al. Educhat: A large-scale language model-based chatbot system for intelligent education. arXiv preprint arXiv:2308.02773, 2023

    2023 arXiv

  51. [59]

    Abu-Rasheed, M.H

    H. Abu-Rasheed, M.H. Abdulsalam, C. Weber, and M. Fathi. Supporting student decisionsonlearningrecommendations: Anllm-basedchatbotwithknowledgegraph contextualization for conversational explainability and mentoring.arXiv preprint arXiv:2401.08517, 2024

    2024 arXiv

  52. [60]

    Edith Arámbula-Mercado, Ryan A. Hill, S. Caro, L. Manrique-Sánchez, E.S. Park, andE.Fernando. Understandingmechanismsofravelingtoextendopen-gradedfric- tioncourse(ogfc)servicelife. ResearchReportBDR74-977-04,FloridaDepartment of Transportation / Texas A&M Transportation Instit...

    2016

  53. [61]

    Researchonperformanceofsbs-ppaandsbr-ppacompoundmodifiedasphalts

    JianguoWei,SongShi,YumingZhou,ZhiyuanChen,FanYu,ZhuyiPeng,andXurui Duan. Researchonperformanceofsbs-ppaandsbr-ppacompoundmodifiedasphalts. Materials (Basel), 15(6):2112, March 2022. doi: 10.3390/ma15062112

    2022 doi

  54. [62]

    Impact of curling and warping on concrete pavement: Phase ii

    Halil Ceylan and Kasthurirangan Gopalakrishnan. Impact of curling and warping on concrete pavement: Phase ii. Final Research Report IHRB Project TR-749, Iowa DepartmentofTransportation;IowaHighwayResearchBoard;IowaStateUniversity InstituteforTransportation,Ames,IA,2023. URL ht...

    2023

  55. [63]

    Mohamed I. E. Attia and Manal A. Ahmed. Impact of vehicle class and tire pressure on pavement performance in mepdg.International Journal of Engineering Research & Applications, 4(10 Part 3):45–57, 2014. 22

    2014

  56. [64]

    James Wilde, Luke Thompson, and Thomas J

    W. James Wilde, Luke Thompson, and Thomas J. Wood. Cost-effective pavement preservation solutions for the real world. Final Report 2014-33, Minnesota Local RoadResearchBoard;MinnesotaDepartmentofTransportation,St.Paul,MN,2014. URL https://www.lrrb.org/pdf/201433.pdf. 23 Table ...

    2014

  57. [66]

    PAPPL improved my problem-solving skills for pavement- engineering problems

  58. [67]

    PAPPL deepened my understanding of pavement-engineering con- cepts

  59. [68]

    Engagement 5

    Using PAPPL enhanced my grasp of pavement-engineering princi- ples. Engagement 5. PAPPL’s interactive exercises kept me motivated to solve pavement- engineering problems

  60. [69]

    I felt engaged while solving pavement-engineering problems with PAPPL’s AI hints

  61. [70]

    I often felt distracted while using PAPPL to solve pavement- engineering problems

  62. [71]

    Adaptivity 9

    Using PAPPL kept me actively involved in solving pavement- engineering problems. Adaptivity 9. PAPPL adapted hints based on my responses and progress

  63. [72]

    PAPPL’s AI hints were often irrelevant to my specific errors

  64. [73]

    PAPPL provided a personalized learning experience tailored to my needs

  65. [74]

    Satisfaction 13

    PAPPL’sAIhintsprovidedanappropriatelevelofguidancewithout revealing answers. Satisfaction 13. PAPPL’s learning experience was valuable for my pavement- engineering studies

  66. [75]

    The PAPPL interface was easy to use and supported my learning

  67. [76]

    I would recommend PAPPL to peers studying pavement- engineering

  68. [77]

    Accuracy 17

    Using PAPPL was frustrating at times. Accuracy 17. PAPPL’s AI hints were relevant and helpful for solving pavement- engineering problems

  69. [78]

    PAPPL’s AI hints were free of inaccuracies or misleading informa- tion

  70. [79]

    I often ignored PAPPL’s AI hints because they were not helpful

  71. [80]

    Open-Ended 21

    PAPPL’s AI hints were often repetitive. Open-Ended 21. Which PAPPL features best supported your learning? Provide examples

  72. [81]

    What challenges did you face with PAPPL, and how could it be improved? 24

  73. [2024]

    Springer Nature Switzerland

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.