Pith. sign in

REVIEW 4 major objections 4 minor 3 cited by

GPT-4o can score introductory-physics argument essays in line with multiple-choice performance, and students find its feedback useful and accurate.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Students who answered a physics multiple-choice question correctly received higher GPT-4o essay scores, and most students rated the AI feedback as mostly useful and mostly accurate.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A modest, honest exploratory study: student perception data are solid, but the 'LLM scoring works' claim rests on an unvalidated MC-correctness proxy. the 4 major comments →

arxiv 2508.14759 v1 pith:FA4TB6RU submitted 2025-08-20 physics.ed-ph

Students' Perceptions to a Large Language Model's Generated Feedback and Scores of Argumentation Essays

Winter Allen , Anand Shanker , N. Sanjay Rebello This is my paper
classification physics.ed-ph
keywords LLM feedbackscientific argumentationintroductory physicsGPT-4oclaim-evidence-reasoningstudent perceptionsautomated scoringlarge-enrollment courses
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper asks whether a large language model can take over the feedback burden that keeps large introductory physics courses from giving every student individualized comments on written work. Working with essays written under a claim-evidence-reasoning scaffold, the authors used GPT-4o with a five-point rubric to score each essay and write a short physics-focused comment, then gave the score and comment back to students. On both a hard and an easy energy quiz, the model's average score was higher for students who answered the underlying multiple-choice question correctly. In a delayed survey, students rated the comments as mostly useful and mostly accurate, regardless of whether they had gotten the question right. The paper concludes that LLM-generated feedback on written scientific arguments is a viable option for large-enrollment courses.

Core claim

The central claim is that a large language model can produce scores and feedback for students' written physics arguments that behave the way a useful grader's would: on the hard quiz the LLM gave students with correct multiple-choice answers an average 2.14/5 versus 1.12/5 for incorrect responders, and on the easy quiz 2.77 versus 2.53, both differences statistically significant. In the survey, the majority of students selected 'mostly useful' and 'mostly accurate' on Likert scales. The authors take this as evidence that the LLM scoring works for both high- and low-difficulty problems and that LLM feedback is a viable path to individualized feedback at scale.

What carries the argument

The machinery is a GPT-4o grading pipeline. Each essay goes to the model with the original problem, an example of an ideal answer, a five-point rubric built on the claim-evidence-reasoning framework, and a prompt that restricts feedback to the physics content in at most 100 words. The rubric allots two to three points to identifying the relevant physics claim and two to three points to appropriate evidence, mirroring the scaffolding students practiced in recitation. The validation instrument is the comparison of LLM scores for students who answered the underlying multiple-choice question correctly versus incorrectly, and the outcome measure is the student survey of perceived usefulness and a

Load-bearing premise

The load-bearing assumption is that getting the multiple-choice question right is a good stand-in for writing a good argument, so an LLM that scores correct-answer essays higher counts as 'working'—but the paper never checks LLM scores against human expert scores.

What would settle it

Score the same essays with two independent human raters using the same five-point CER rubric and compare their scores to GPT-4o's, for example with quadratic weighted kappa. If human–LLM agreement is below roughly 0.5, or if human raters agree with each other far more than with the LLM, the claim that the LLM scoring works is falsified. A second check: feed the LLM essays that contain no physics content but are padded with CER-sounding sentences; if those score at or above the average student essay, the scores are not tracking argument quality.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Large-enrollment physics courses could give every student individualized written feedback on argumentation essays without additional graders.
  • LLM scores carry signal about essay quality, at least on hard problems, since they separate correct and incorrect responders with a large effect size.
  • Feedback reaches struggling students too: students who answered incorrectly rated it as useful and accurate as those who answered correctly.
  • Moving from delayed end-of-semester feedback to feedback delivered within days of each quiz is a natural next deployment step.
  • The same prompt-and-rubric recipe can be reused across quizzes and, with modest changes, across different physics topics.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper's validation shortcut, comparing LLM scores to multiple-choice correctness, is never checked against human expert scoring, so an obvious extension is to compare LLM and human rubric scores on the same essays.
  • Because the survey measures perception, not learning, a future study should measure whether students who act on the feedback improve their later argumentation essays.
  • A possible confound the authors do not test is essay length or verbosity: if the LLM implicitly rewards longer essays, the score difference could partly reflect effort rather than argument quality.
  • The approach could be tested across other STEM argumentation genres, such as lab conclusions or design justifications, to see whether the CER rubric transfers.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. This manuscript reports an exploratory classroom study in which GPT-4o generated rubric-based scores and delayed feedback for students' words-only CER-style argumentation essays in an introductory calculus-based physics course, for two quizzes (Quiz 08 and Quiz 09). The authors compare average LLM scores for students who answered the associated multiple-choice question correctly versus incorrectly (Table I) and report Likert ratings of perceived usefulness and accuracy of the LLM feedback. They conclude that the LLM scoring 'works' for both a high-difficulty and a low-difficulty problem, and that students perceive the feedback as mostly useful and accurate, suggesting that LLM-based feedback is a viable option in large-enrollment courses.

Significance. If the conclusions were fully supported, the paper would be a useful proof-of-concept for scalable, individualized feedback in large introductory physics courses. The study has real strengths: it uses a genuine classroom setting, has large score samples (N=730 and N=565), uses simple and transparent statistical analyses, and explicitly acknowledges several limitations, including the delayed timing of feedback and the absence of a human-grader comparison. The perception data are direct and relevant. However, the score-validity evidence is indirect, because it relies entirely on multiple-choice correctness as a proxy for written-argument quality, and survey response rates are not reported. The contribution is therefore suggestive rather than confirmatory.

major comments (4)
  1. [Section IV, Table I and Section V] The claim that 'LLM scored students as expected... which indicates that the LLM scoring works for both high and low difficulty problems' is not supported by the evidence presented. The only external criterion used is whether the student answered the associated multiple-choice question correctly. The MC item tests selection of a final answer, while the LLM scores a words-only CER argument; these are different constructs. The observed score gap could be driven by essay length, keyword usage, writing fluency, or general course ability rather than argumentation quality. No human-expert scores are reported, and the paper itself lists human-grader comparison as future work (Section V). The wording should be restrained to 'scores differed by MC correctness' unless criterion validity is established.
  2. [Table I, Quiz 09 row; Section IV] The text states 'A very small p-value and large Cohen’s d' for Quiz 09, but Table I reports d = 0.22, which is conventionally a small effect size. This is not merely a wording issue: the small effect undermines the claim that LLM scoring 'works' equally for low-difficulty problems. The authors should report the effect size accurately and interpret the Quiz 09 result as weak evidence of a scoring difference, not as confirmation.
  3. [Section III.B] No reliability information is given for the LLM-generated scores. The API temperature was set to 0.80, so the model is nondeterministic, yet it appears each essay was scored only once. Without repeating the scoring or reporting a stability metric (e.g., agreement across runs), the reported differences in Table I may partly reflect prompt/temperature noise. This matters for the conclusion that LLM scoring 'works.'
  4. [Section III.C and Figure 1] The survey response rate is not reported. Students received bonus points for completing the delayed post-survey, but the manuscript does not state how many of the 730 (Quiz 08) or 565 (Quiz 09) students actually responded, nor does Figure 1 show sample sizes. If response rates were low, self-selection could bias the perception ratings, which are central to the paper's positive conclusion about student perceptions. Please report response counts and, ideally, compare respondents to nonrespondents.
minor comments (4)
  1. [Table I] Typographical and table-formatting issues: 'Correct A verage' is missing a space, and the column headers could be more readable. Adding sample sizes per subgroup (correct/incorrect) would also help.
  2. [Section IV] The sentence 'Due to the very small p-value and large Cohen’s d' appears in the Quiz 08 paragraph and again with 'large Cohen’s d' in the Quiz 09 paragraph. The Quiz 09 wording contradicts the reported d=0.22 and should be corrected.
  3. [Figure 1] The panels are described only in the caption as (a) and (b) etc., but the axis labels, legend, sample sizes, and exact Likert anchors are not visible in the text. Please include these details for readability.
  4. [Section III.C] The manuscript states that one question was a six-point Likert scale and another a four-point Likert scale, but the exact question wording is not provided. Reporting the item stems would strengthen reproducibility.

Circularity Check

0 steps flagged

No significant circularity; LLM scores are compared to an external criterion (MC correctness) and perception ratings are self-reported survey data.

full rationale

The paper's central empirical claims are (1) LLM-generated scores differ between students who answered the underlying multiple-choice question correctly versus incorrectly, and (2) students mostly perceived the LLM feedback as useful and accurate. Neither claim reduces to the paper's inputs by construction. The LLM scores were generated by prompting GPT-4o with a rubric, problem statement, and an ideal example; the multiple-choice correctness variable was not fed into the scoring prompt, so the observed score differences are an external comparison rather than a fitted parameter being renamed a prediction. The rubric design choices may influence scores, but they do not force the reported t-test results or effect sizes. The perception data are self-reported Likert ratings and are not derived from the scoring model. The limitations section openly acknowledges the absence of human-grader comparison and delayed feedback, and lists human-grader comparison as future work; this is a validity limitation, not an indication of circularity. Self-citations (e.g., Rebello et al. for argumentation effectiveness, Allen et al. for the scaffolding implementation) appear only in background or methodological context and are not load-bearing for the paper's main conclusions. No circular step could be identified from the text.

Axiom & Free-Parameter Ledger

1 free parameters · 4 axioms · 0 invented entities

The central claims rest on several unvalidated assumptions, chiefly that MC correctness is a proxy for argument quality, that the CER framework is valid, and that the survey sample is representative. No new physical or conceptual entities are introduced.

free parameters (1)
  • GPT-4o temperature = 0.80
    LLM sampling temperature set by hand to 0.80, not varied or optimized; it controls stochasticity of scores and feedback, which are not reported for consistency.
axioms (4)
  • domain assumption Students who answer the MC question correctly tend to write better scientific arguments.
    Used in Results to interpret the LLM score difference between correct and incorrect groups as evidence that LLM scoring works.
  • domain assumption The CER (claim, evidence, reasoning) framework is an appropriate and valid model for scientific argumentation in this context.
    The paper builds the rubric and prompt on CER, citing McNeill and Krajcik; if the framework is not valid, the scores may not measure argument quality.
  • domain assumption The survey respondents are representative of the full class.
    The survey was voluntary with bonus points; no response rate is reported, so non-response bias could skew perception ratings.
  • standard math The t-test assumptions (independence, approximate normality) hold for the LLM scores.
    The paper uses t-tests to compare means; no normality tests or adjustment for multiple comparisons are reported.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Students' Perceptions to a Large Language Model's Generated Feedback and Scores of Argumentation Essays." pith.science (2026). https://pith.science/paper/FA4TB6RU

@misc{pith2026250814759,
  author       = {Pith},
  title        = {Pith review of: Students' Perceptions to a Large Language Model's Generated Feedback and Scores of Argumentation Essays},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FA4TB6RU}},
  note         = {Machine review of arXiv:2508.14759}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Students in introductory physics courses often rely on ineffective strategies, focusing on final answers rather than understanding underlying principles. Integrating scientific argumentation into problem-solving fosters critical thinking and links conceptual knowledge with practical application. By facilitating learners to articulate their scientific arguments for solving problems, and by providing real-time feedback on students' strategies, we aim to enable students to develop superior problem-solving skills. Providing timely, individualized feedback to students in large-enrollment physics courses remains a challenge. Recent advances in Artificial Intelligence (AI) offer promising solutions. This study investigates the potential of AI-generated feedback on students' written scientific arguments in an introductory physics class. Using Open AI's GPT-4o, we provided delayed feedback on student written scientific arguments and surveyed them about the perceived usefulness and accuracy of this feedback. Our findings offer insights into the viability of implementing real-time AI feedback to enhance students' problem-solving and metacognitive skills in large-enrollment classrooms.

Figures

Figures reproduced from arXiv: 2508.14759 by Anand Shanker, N. Sanjay Rebello, Winter Allen.

Figure 1
Figure 1. Figure 1: FIG. 1: Students Perception of the LLM Feedback. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Students' Epistemological Beliefs and their Chatbot Preferences in AI-mediated Physics Learning

    physics.ed-ph 2026-07 conditional novelty 5.0

    Students who chose a combination chatbot (guided inquiry then answers) scored slightly higher on an epistemological-beliefs survey than answer-preference students, but the difference was not robust to multiple-testing...

  2. Feedback That Clicks: Introductory Physics Students' Valued Features in AI Feedback Generated From Self-Crafted and Engineered Prompts

    physics.ed-ph 2025-09 conditional novelty 5.0

    Introductory physics students rarely use prompt engineering on their own, but they rate AI feedback as most useful when the prompt explicitly requests evaluation, the correct answer, and improvement suggestions.

  3. Creating and Evaluating K-12 GenAI Assessment Graders Through Context Engineering

    cs.CY 2026-05 unverdicted novelty 3.0

    LLM graders achieve substantial human agreement on math and science MCAS items but vary on ELA, performing best as sources of formative narrative feedback rather than summative numerical scores.

Reference graph

Works this paper leans on

57 extracted references · 55 canonical work pages · cited by 3 Pith papers · 1 internal anchor

  1. [1]

    National Research Council, Next Generation Science Stan- dards: For States, By States (The National Academies Press, Washington, DC, 2013)

    work page 2013

  2. [2]

    W. J. Leonard, R. J. Dufresne, and J. P. Mestre, Using quali- tative problem-solving strategies to highlight the role of con- ceptual knowledge in solving problems, American Journal of Physics 64, 1495 (1996)

    work page 1996

  3. [3]

    D. P. Maloney, An overview of physics education research on problem solving, Getting started in PER 2, 1 (2011)

    work page 2011

  4. [4]

    R. J. Dufresne, W. J. Gerace, and W. J. Leonard, Solv- ing physics problems with multiple representations, Physics Teacher 35, 270 (1997)

    work page 1997

  5. [5]

    Tuminaro and E

    J. Tuminaro and E. F. Redish, Elements of a cognitive model of physics problem solving: Epistemic games, Phys. Rev. ST Phys. Educ. Res. 3, 020101 (2007)

    2007

  6. [6]

    C. M. Rebello, Using a hybrid of argumentation and problem solving prompts to facilitate undergraduates’ problem solving performance and confidence, in The 13th Conference of the European Science Education Research Association (ESERA) (2019)

    work page 2019

  7. [7]

    E. A. Siverling, T. J. Moore, E. Suazo-Flores, C. A. Mathis, and S. S. Guzey, What initiates evidence-based reasoning?: Situ- ations that prompt students to support their design ideas and decisions, Journal of Engineering Education 110, 294 (2021)

    work page 2021

  8. [8]

    Kortemeyer, J

    G. Kortemeyer, J. Nöhl, and D. Onishchuk, Grading assistance for a handwritten thermodynamics exam using artificial intelli- gence: An exploratory study, Phys. Rev. Phys. Educ. Res. 20, 020144 (2024)

    work page 2024

  9. [9]

    Can Large Language Models Make the Grade? An Empirical Study Evaluating LLMs Ability to Mark Short Answer Questions in K-12 Education

    O. Henkel, A. Boxer, L. Hills, and B. Roberts, Can large lan- guage models make the grade? an empirical study evaluating llms ability to mark short answer questions in k-12 education (2024), arXiv:2405.02985 [cs.CL]

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  10. [10]

    Chen and T

    Z. Chen and T. Wan, Achieving human level partial credit grad- ing of written responses to physics conceptual question using gpt-3.5 with only prompt engineering, in Physics Education Research Conference 2024 , PER Conference (Boston, MA,

    work page 2024

  11. [11]

    L. K. Berland and B. J. Reiser, Making sense of argumentation and explanation, Science education 93, 26 (2009)

    work page 2009

  12. [12]

    Kuhn, Science as argument: Implications for teaching and learning scientific thinking, Science education 77, 319 (1993)

    D. Kuhn, Science as argument: Implications for teaching and learning scientific thinking, Science education 77, 319 (1993)

    work page 1993

  13. [13]

    L. K. Berland and K. L. McNeill, A learning progression for scientific argumentation: Understanding student work and de- signing supportive instructional contexts, Science Education 94, 765 (2010)

    work page 2010

  14. [14]

    Kuhn, Teaching and learning science as argument, Science Education 94, 810 (2010)

    D. Kuhn, Teaching and learning science as argument, Science Education 94, 810 (2010)

    work page 2010

  15. [15]

    you’re going to want to find out which and prove it

    E. A. Forman, J. Larreamendy-Joerns, M. K. Stein, and C. A. Brown, “you’re going to want to find out which and prove it”: Collective argumentation in a mathematics classroom, Learn- ing and instruction 8, 527 (1998)

    work page 1998

  16. [16]

    doing the lesson

    M. P. Jiménez-Aleixandre, A. Bugallo Rodríguez, and R. A. Duschl, “doing the lesson” or “doing science”: Argument in high school genetics, Science education 84, 757 (2000)

    work page 2000

  17. [17]

    D. H. Jonassen and B. Kim, Arguing to learn and learning to argue: Design justifications and guidelines, Educational Tech- nology Research and Development 58, 439 (2010)

    work page 2010

  18. [18]

    Xun and S

    G. Xun and S. M. Land, A conceptual framework for scaf- folding iii-structured problem-solving processes using ques- tion prompts and peer interactions, Educational technology re- search and development 52, 5 (2004)

    work page 2004

  19. [19]

    Cho and D

    K.-L. Cho and D. H. Jonassen, The effects of argumentation scaffolds on argumentation and problem solving, Educational Technology Research and Development 50, 5 (2002)

    work page 2002

  20. [20]

    Christodoulou and J

    A. Christodoulou and J. Osborne, The science classroom as a site of epistemic talk: A case study of a teacher’s attempts to teach science based on argument, Journal of Research in Sci- ence Teaching 51, 1275 (2014)

    work page 2014

  21. [21]

    K. L. McNeill and D. S. Pimentel, Scientific discourse in three urban classrooms: The role of the teacher in engaging high school students in argumentation, Science Education 94, 203 (2010)

    work page 2010

  22. [22]

    Schworm and A

    S. Schworm and A. Renkl, Learning argumentation skills through the use of prompts for self-explaining examples., Jour- nal of Educational Psychology 99, 285 (2007)

    work page 2007

  23. [23]

    W. N. Wampler, The relationship between students’ problem solving frames and epistemological beliefs , Ph.D. thesis, Pur- due University (2013)

    work page 2013

  24. [24]

    C. M. Rebello, Scaffolding evidence-based reasoning in a tech- nology supported engineering design activity, inThe 13th Con- ference of the European Science Education Research Associa- tion (ESERA) (2019)

    work page 2019

  25. [25]

    K. L. McNeill and J. S. Krajcik, Supporting grade 5-8 students in constructing explanations in science: The claim, evidence, and reasoning framework for talk and writing., Pearson (2011)

    work page 2011

  26. [26]

    S. E. Toulmin, The uses of argument (Cambridge university press, 2003)

    work page 2003

  27. [27]

    K. L. McNeill and J. Krajcik, Inquiry and scientific explana- tions: Helping students use evidence and reasoning, Science as inquiry in the secondary setting 121, 34 (2008)

    work page 2008

  28. [28]

    Wang, Scrutinising the positions of students and teacher en- gaged in argumentation in a high school physics classroom, International Journal of Science Education 42, 25 (2020)

    J. Wang, Scrutinising the positions of students and teacher en- gaged in argumentation in a high school physics classroom, International Journal of Science Education 42, 25 (2020)

    work page 2020

  29. [29]

    Hattie and H

    J. Hattie and H. Timperley, The power of feedback, Review of Educational Research 77, 81 (2007)

    work page 2007

  30. [30]

    has shown that feedback is one of the important drivers of learning. Feedback can facilitate improvements in learn- ers’ understanding and skills [31–34] by informing learners about their progress, reinforcing good practice, and moti- vating them to engage in self-regulation [34, 35]. It facil- itates self-assessment and reflection on performance, which c...

    work page

  31. [31]

    Henderson, T

    M. Henderson, T. Ryan, D. Boud, P. Dawson, M. Phillips, E. Molloy, and P. Mahoney, The usefulness of feedback, Ac- tive Learning in Higher Education 22, 229 (2021)

    work page 2021

  32. [32]

    Burgess, C

    A. Burgess, C. van Diggele, C. Roberts, and C. Mellis, Feed- back in the clinical setting, BMC Medical Education 20, 1 (2020)

    work page 2020

  33. [33]

    Sadler, Beyond feedback: Developing student capability in complex appraisal, Assessment & Evaluation in Higher Educa- tion 35, 535 (2010)

    R. Sadler, Beyond feedback: Developing student capability in complex appraisal, Assessment & Evaluation in Higher Educa- tion 35, 535 (2010)

    work page 2010

  34. [34]

    Nicol and D

    D. Nicol and D. Macfarlane-Dick, Formative assessment and self-regulated learning: A model and seven principles of good feedback practice, Studies in Higher Education31, 199 (2006)

    work page 2006

  35. [35]

    Burgess and C

    A. Burgess and C. Mellis, Receiving feedback from peers: medical students’ perceptions, Clinical Teacher12, 245 (2015)

    work page 2015

  36. [36]

    L. A. Shepard, The role of assessment in a learning culture, Educational Researcher 29, 4 (2000)

    2000

  37. [37]

    A. N. Kluger and A. DeNisi, The effects of feedback interven- tions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory, Psychological Bul- letin 119, 254 (1996)

    work page 1996

  38. [38]

    Wisniewski, K

    B. Wisniewski, K. Zierer, and J. Hattie, The power of feedback revisited: A meta-analysis of educational feedback research, Frontiers in Psychology 10, 3087 (2020)

    work page 2020

  39. [39]

    Brooks, A

    C. Brooks, A. Carroll, R. M. Gillies, and J. Hattie, A matrix of feedback for learning, Australian Journal of Teacher Education 44, 2 (2019)

    work page 2019

  40. [40]

    Hattie and S

    J. Hattie and S. Clarke, Visible Learning: Feedback (Rout- ledge, 2018)

    work page 2018

  41. [41]

    D. R. Ferris, The influence of teacher commentary on student revision, TESOL Quarterly 31, 315 (1997)

    work page 1997

  42. [42]

    W. T. Branch and A. Paranjape, Feedback and reflection: Teaching methods for clinical settings, Academic Medicine77, 1185 (2002)

    work page 2002

  43. [43]

    Hepplestone and G

    S. Hepplestone and G. Chikwa, Understanding how students process and use feedback to support their learning, Practitioner Research in Higher Education 8, 41 (2014)

    work page 2014

  44. [44]

    Price, K

    M. Price, K. Handley, J. Millar, and B. O’Donovan, Feedback: all that effort, but what is the effect?, Assessment & Evaluation in Higher Education 35, 277 (2010)

    work page 2010

  45. [45]

    T. Ryan, M. Henderson, and M. Phillips, Feedback modes mat- ter: Comparing student perceptions of digital and non-digital feedback modes in higher education, British Journal of Educa- tional Technology 50, 1507 (2019)

    work page 2019

  46. [46]

    N. E. Winstone, R. A. Nash, M. Parker, and J. R. and, Support- ing learners’ agentic engagement with feedback: A systematic review and a taxonomy of recipience processes, Educational Psychologist 52, 17 (2017)

    work page 2017

  47. [47]

    Court, Tutor feedback on draft essays: Developing students’ academic writing and subject knowledge, Journal of Further and Higher Education 38, 327 (2014)

    K. Court, Tutor feedback on draft essays: Developing students’ academic writing and subject knowledge, Journal of Further and Higher Education 38, 327 (2014)

    work page 2014

  48. [48]

    Boud and E

    D. Boud and E. Molloy, Feedback in Higher and Professional Education: Understanding it and doing it well (Routledge, 2013)

    work page 2013

  49. [49]

    Orlando, How to effectively assess online learning (Magna Publications, 2011)

    J. Orlando, How to effectively assess online learning (Magna Publications, 2011)

    work page 2011

  50. [50]

    Pitt and L

    E. Pitt and L. N. and, ‘now that’s the feedback i want!’ stu- dents’ reactions to feedback on graded work and what they do with it, Assessment & Evaluation in Higher Education 42, 499 (2017)

    work page 2017

  51. [51]

    Furnborough and M

    C. Furnborough and M. T. and, Adult beginner distance lan- guage learner perceptions and use of assignment feedback, Dis- tance Education 30, 399 (2009)

    work page 2009

  52. [52]

    T. D. Wolsey, E-feedback: An exploratory study of using email to provide feedback to students, Journal of Writing Assessment 4, 1 (2008)

    work page 2008

  53. [53]

    K. L. McNeill and D. M. Martin, Claims, evidence, and rea- soning, Science and Children 48, 52 (2011)

    work page 2011

  54. [54]

    Ortiz-Rodríguez, R

    M. Ortiz-Rodríguez, R. W. Telg, T. Irani, T. G. Roberts, and E. Rhoades, College students’perceptions of quality in distance education the importance of communication., Quarterly Re- view of Distance Education 6 (2005)

    work page 2005

  55. [55]

    completely useless

    using its application programming interface (API). For both quizzes, the LLM was given the prompt: You are an ed- ucator. Your goal is to provide useful feedback on the physics aspect of the essay. Your feedback should be no more than 100 words long. Focus only on the physics ideas and con- cepts. Do not include salutations. Do not comment on the grammar ...

    work page

  56. [56]

    Allen, C

    W. Allen, C. M. Rebello, and N. S. Rebello, Assessing physics students’ scientific argumentation using natural language pro- cessing, arXiv preprint arXiv:2504.08910 (2025)

    work page arXiv 2025

  57. [57]

    OpenAI, Gpt-4o (2024), large multimodal model

    work page 2024

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.