Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 103 of the 103 references it cites were found in a registry or an index; 13 of the 108 sources its claims cite were checked for support.

On opportunities and challenges of large multimodal foundation models in education

Open access

Abstract

Recently, the option to use large language models as a middleware connecting various AI tools and other large language models led to the development of so-called large multimodal foundation models, which have the power to process spoken text, music, images and videos. In this overview, we explain a new set of opportunities and challenges that arise from the integration of large multimodal foundation models in education.

Overview

CoChat Summary

TL;DR

This overview argues that large multimodal foundation models (LMFMs) such as ChatGPT-4-Turbo and Gemini create new educational opportunities for learners, teachers, researchers, and developers by processing text, speech, images, and video, while also introducing challenges around accuracy, bias, privacy, access, and overreliance. It proposes mitigation strategies emphasizing AI literacy, human oversight, transparent documentation, and equitable access.

Key findings

  • LMFMs extend large language models by enabling input and output via written text, spoken text, images, and videos, and can act as middleware connecting different AI tools; examples named include ChatGPT-4-Turbo and Gemini.
  • Opportunities span personalized learning, barrier-free interaction, multimedia learning, representational competence training, experimentation and code-generation support, teacher lesson planning and assessment, researcher data analysis and literature review, and developer creation of customized learning applications.
  • Challenges include output and source verification, difficulty maintaining an overview of involved AI models, potential segregation due to access costs, ethical issues and inherent biases, dependency and overuse, personal data leaks, plagiarism risks, digital divide, erosion of human relations, reliability of information, and need for AI literacy.
  • The paper reports that GPT-4 Turbo’s expanded context length can consider and modify extensive source code, potentially encompassing 128k tokens, described as more than 300 pages of text.

Methods

  • Design: narrative overview/perspective that synthesizes prior literature, examples, and arguments about LMFMs in education; it is not an empirical study.
  • Data or sample: Not stated in the paper; no dataset, participants, or systematic sampling procedure is reported.
  • Main analysis: Not stated in the paper; no formal quantitative or qualitative analysis is described.

Limitations

  • The paper does not state limitations of its own overview.
  • Evident from the text: it is a non-systematic overview/perspective, so its claims about opportunities and challenges are not empirically tested in this manuscript.
  • Evident from the text: the authors note increasingly short innovation cycles in AI, which makes specific model examples and claims potentially time-sensitive.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41539-025-00301-w

Cites

103 of 103 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

9 Valid · 1 Not valid · 66 Not checked · 108 Own results and interpretations

60 more claims · 3 with their cited sources checked

Create a free account to see all 184 claims and to verify the 95 references still open.

See all claims — free account
  • It may state two separate claims as one. 61
  • Probably not a substantial claim. 90

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 76 references

Research works printed but not linked: 9

  1. Fauzi, F., Tuhuteru, L., Sampe, F., Ausat, A. M. A. & Hatta, H. R. Analysing the role of ChatGPT in improving student productivity in higher education. J. Educ. 5, 14886–14891 (2023).
  2. Tavangarian, D. et al. Is e-Learning the Solution for Individual Learning? Electron. J. E learn. 2, 265–272 (2004).
  3. Hutson, J. & Robertson, B. Exploring the Educational Potential of AI Generative Art in 3D Design Fundamentals: A Case Study on Prompt Engineering and Creative Workflows. Glob. J. Hum.-Soc. Sci. 23, 485 (2023).
  4. Meske, C. & Bunde, E. Design principles for user interfaces in ai-based decision support systems: The case of explainable hate speech detection. Inf. Syst. Front. 25, 743–773 (2023).
  5. Goodman, B. & Flaxman, S. European union regulations on algorithmic decision-making and a “right to explanation”. AI Mag. 38, 50–57 (2017).
  6. Anderson, J., Rainie, L. & Luchsinger, A. Artificial intelligence and the future of humans. Pew Res. Center 10, 12 (2018).
  7. Trust, T., Whalen, J. & Mouza, C. Chatgpt: Challenges, opportunities, and implications for teacher education. Contemp. Issues Technol. Teach. Educ. 23, 1–23 (2023).
  8. Leemann, T., Rong, Y., Nguyen, T.-T., Kasneci, E. & Kasneci, G. Caution to the exemplars: On the intriguing effects of example choice on human trust in XAI. In XAI in Action: Past, Present, and Future Applications (NIPS, 2023).
  9. Borisov, V., Sessler, K., Leemann, T., Pawelczyk, M. & Kasneci, G., Language models are realistic tabular data generators. In: The Eleventh International Conference on Learning Representations (ICLR, 2022).

Cited by

See all 48

Metadata

Authors
Küchemann, S., Avila, K. E., Dinc, Y., Hortmann, C., Revenga, N., Ruf, V., Stausberg, N., Steinert, S., Fischer, F., Fischer, M., Kasneci, E., Kasneci, G., Kuhr, T., Kutyniok, G., Malone, S., Sailer, M., Schmidt, A., Stadler, M., Weller, J., & Kuhn, J.
Affiliations
  • Ludwig-Maximilians-Universität München (DE)
  • Institut für Didaktik und Ausbildungsforschung in der Medizin (DE)
  • Technical University of Munich (DE)
  • Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE)
  • Saarland University (DE)
  • University of Augsburg (DE)
Corresponding author
Küchemann, Stefan
Venue
npj Science of Learning
Published
2025-02-26
Volume
10
Issue
1
Pages
11-11
ISSN
2056-7936 (ISSN-L)
OpenAlex
W4407928447
Cited by
47
References
76
Citations by year
2026: 32 · 2025: 15
FWCI
4.92
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Science of Learning (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Open access LMU (Ludwid Maxmilian's Universitat Munchen) (submittedVersion)
  • OPUS (Augsburg University) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) (submittedVersion, other-oa)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
  • Social Sciences › Education › Artificial Intelligence in Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
Keywords
large language models, artificial intelligence in education, multimodal AI, educational technology integration, opportunities and challenges

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-04, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-04 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: On opportunities and challenges of large multimodal foundation models in education. https://app.cochat.ai/papers/10.1038/s41539-025-00301-w
The paper
Küchemann, S., Avila, K. E., Dinc, Y., Hortmann, C., Revenga, N., Ruf, V., Stausberg, N., Steinert, S., Fischer, F., Fischer, M., Kasneci, E., Kasneci, G., Kuhr, T., Kutyniok, G., Malone, S., Sailer, M., Schmidt, A., Stadler, M., Weller, J., & Kuhn, J. (2025). On opportunities and challenges of large multimodal foundation models in education. npj Science of Learning. https://doi.org/10.1038/s41539-025-00301-w

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.