Pith. sign in

REVIEW 5 major objections 5 minor 62 references

Can ChatGPT Diagnose Alzheimer's Disease?

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

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

Pith's one-line read The paper claims ChatGPT reaches 94.6 percent accuracy classifying Alzheimer's-related electronic health records when given 100 labeled examples per class and both MRI and cognitive-test features at a 75 percent confidence threshold.

desk verdict A large-scale but methodologically flawed evaluation: the headline accuracy probably reflects participant-level leakage from the multi-shot examples, so the central claim is unsupported. read the letter →

arxiv 2502.06907 v1 pith:QKN4G3FE submitted 2025-02-10 cs.LG cs.AI

Quoc-Toan Nguyen , Linh Le , Xuan-The Tran , Thomas Do , Chin-Teng Lin This is my paper · ORCID
classification cs.LGcs.AI
keywords Alzheimer'sdiseaseChatGPTlargelanguagemodelsmulti-shotpromptingzero-shotdiagnosticaccuracyconfidencecalibrationelectronichealthrecords
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

The paper asks whether ChatGPT, a general-purpose chatbot, can classify electronic health records as normal cognition, mild cognitive impairment, or Alzheimer's disease using MRI volumes and cognitive-test scores. Its central claim is that it can, provided the prompt includes 100 labeled examples of each class: under that multi-shot condition, with both data modalities, ChatGPT reaches 94.6 percent accuracy, 95.0 percent recall, and 94.6 percent precision at a 75 percent confidence threshold. The result is offered as evidence that a chatbot used as a black box, without any access to its internal states, could act as a supportive diagnostic tool, particularly where dementia specialists are scarce and automating part of the workflow could speed up triage. This is why the question matters: a chat interface plus a spreadsheet might become a first-pass screening step for Alzheimer's disease.

What carries the argument

The load-bearing technique is multi-shot prompting, a prompting strategy in which the text prompt embeds labeled example records alongside the unlabeled records to be classified. Here the examples are 100 complete records from each of the three diagnostic classes, normal cognition, mild cognitive impairment, and Alzheimer's disease, and the input fields are seven MRI volume features plus twelve cognitive-test features; ChatGPT is asked to give a class and a confidence score from 0 to 100 for every uploaded record. The 75 percent threshold selects only predictions the model is at least 75 percent sure about, which is the setting that produces the 0.946 accuracy. The technique does the work by giving the model a small in-context training set, so the same general-purpose model can align the numeric columns with the requested labels without any fine-tuning.

What would settle it

Re-run the multi-shot condition with a participant-level exclusion: remove every record belonging to any participant whose record was one of the 300 examples, classify the remaining records, and compare accuracy at the 75 percent threshold; a large drop below 0.946 would show the reported accuracy depends on overlap between examples and test records rather than on the model's diagnostic ability.

Watch

Extended reading notes

Core claim

The paper's central claim is that ChatGPT can serve as a supportive tool to diagnose Alzheimer's disease, and the measured basis for that claim is the multi-shot prompting condition. With 100 labeled examples per class in the prompt and both MRI and cognitive-test features, ChatGPT reaches accuracy 0.946, recall 0.950, precision 0.946, and F1 0.948 at the 75 percent confidence threshold, with an expected calibration error of 0.066 and a maximum calibration error of 0.140. The same input setup without examples, zero-shot, reaches only 0.744 accuracy on the combined data, and the combined data beat either modality alone in every condition, with cognitive tests alone at 0.809 and MRI alone at 0.606 for multi-shot. The paper reads this as evidence that in-context examples and multimodal input jointly make a general-purpose chat model accurate enough to be considered a supportive diagnostic tool.

Load-bearing premise

The 94.6 percent figure stands only if the 300 example records put in the prompt come from participants who are entirely absent from the records being scored, and the paper never states that this is the case.

Editorial extensions

If this is right

  • If the claim holds, a clinic with no local AI team could screen Alzheimer's cases by preparing a spreadsheet of MRI and cognitive-test values and running a chat prompt, since the method needs no model training or dedicated infrastructure.
  • The gain from zero-shot to multi-shot on combined data, from 0.744 to 0.946 accuracy, shows that a modest number of labeled examples, not task-specific fine-tuning, is what unlocks the diagnostic behavior.
  • The consistent advantage of combining MRI volumes with cognitive-test scores implies that diagnostic prompts should include both modalities rather than relying on a single test.
  • The reported calibration, with an expected calibration error of 0.066 at the 75 percent threshold, implies that the model's confidence values could be used to separate high-certainty predictions, which might be accepted automatically, from low-certainty ones, which should be referred to a specialist.
  • Repeating the best condition five times with cleared chat histories produces accuracy 0.946 ± 0.001, so the result appears stable across independent runs of the same prompt.

Reading between the lines

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

  • I infer that a participant-level split would be the decisive generalization test: the paper mixes 9,300 records from 1,480 participants with repeated visits, and it never states that the 300 example participants are excluded from the evaluated file.
  • I would expect the accuracy to be sensitive to the number and composition of examples, since only the 100-per-class case is tested; trying 20 or 50 examples per class would show the practical floor for the method.
  • The same in-context classification template could plausibly transfer to other tabular medical data such as lab values or vital signs, but the paper provides no evidence for that transfer.
  • Because predictions go through a natural-language chat interface, small wording changes in the prompt could shift results; the paper does not test prompt sensitivity.
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

5 major / 5 minor

Summary. The paper evaluates whether ChatGPT (GPT-4-turbo) can classify ADNI electronic health records into Normal Control, Mild Cognitive Impairment, or Alzheimer's Disease using MRI features, cognitive test scores, or both, under zero-shot and multi-shot prompting. In the multi-shot condition with 100 labeled examples per class and combined MRI+cognitive features, the authors report accuracy of 0.946, recall 0.950, precision 0.946, and F1 0.948 at a 75% confidence threshold (Table 5). They conclude that 'ChatGPT can be a supportive tool to diagnose AD' (Section 7). The paper includes calibration metrics (ECE/MCE) and repeated runs (five per condition) with chat history cleared between runs.

Significance. If the reported performance were robust, the result would be notable: a black-box, general-purpose LLM achieving high diagnostic accuracy on structured EHR data with only a few hundred in-context examples, with potential implications for resource-limited settings facing a shortage of AD specialists. However, the significance is substantially tempered by the absence of any non-LLM baseline, the post hoc selection of the 75% confidence threshold, and the lack of a documented participant-level split between the in-context examples and the evaluated records. The paper does provide a clear task formulation, multiple performance and calibration metrics, and repeated runs per condition, but the experimental design as reported does not support the central claim as stated.

major comments (5)
  1. [§3 and §5.2] The multi-shot evaluation has a serious participant-leakage risk. Section 3 states that 1,480 participants contribute 9,300 records via repeated visits, but Section 5.2 only says that 100 samples per class were selected for the prompt, without stating whether those participants were excluded from the evaluated file (Multi_Shot_ADNI.csv). If any participant appears in both the prompt examples and the evaluation set, the high accuracy (0.946) could be inflated by the model recognizing highly correlated repeated visits from the same person, rather than diagnosing unseen participants. The paper must either provide evidence of a participant-level split or rerun the experiments with a held-out participant set.
  2. [§6.2, Table 5] No baseline classifier is reported. The paper compares zero-shot versus multi-shot prompting only, but the claim that ChatGPT is a 'supportive tool' requires comparison with a simple supervised model (e.g., logistic regression, random forest) on the same features. Without such a baseline, it is impossible to judge whether 0.946 accuracy reflects a distinctive capability of the LLM or merely the high signal contained in the cognitive-test and MRI features.
  3. [§5.2 and §7] The 75% confidence threshold is described in Section 7 as 'optimal', but it is selected post hoc from the reported 25/50/75% thresholds. Additionally, the paper does not report how many records remain after applying each threshold; if the metrics are computed only on high-confidence records, accuracy is inflated by design. The authors should pre-register the threshold or select it on a validation split, and they should report coverage (the number of classified records) for every threshold.
  4. [§5.1, Tables 3 and 5] The evaluation metrics are defined only for binary classification (TP/TN/FP/FN), but the task is three-class (NC/MCI/AD). The paper does not specify whether precision, recall, and F1 are macro-averaged, micro-averaged, or computed one-vs-rest, nor how the confidence threshold is applied to three-class predictions. This ambiguity hinders reproducibility and makes the reported numbers difficult to interpret.
  5. [§7 vs. Table 4] Section 7 states that 'multi-shot prompting achieves ECE and MCE values of 0.066 and 0.005, respectively', but Table 4 reports MCE = 0.140 ± 0.005 for the same condition. The '0.005' appears to be a typo (perhaps the standard deviation), but it is a factual inconsistency in the conclusion that a reader would rely on.
minor comments (5)
  1. [Figure 1] The prompt text in Figure 1 contains typos in feature names: 'WholeBrai' should be 'WholeBrain' and 'RAVLT_perc_fO' has inconsistent capitalization; these should be standardized.
  2. [§4.1 and §4.2] The probability-based equations g(r,c; T, P, K) are not operational: they do not describe how the model's confidence score is parsed from the free-text output or how the arg max is computed. A short pseudocode or a description of the output parsing would be more useful.
  3. [§6.1 and §6.2] The tables show MCE as identical across thresholds for some modalities (e.g., Table 2, MRI: MCE=0.495 for all thresholds), which is suspicious if MCE is recomputed on threshold-filtered subsets. Please clarify whether MCE is computed on all predictions or only on those meeting the threshold.
  4. [References] Reference [39] is missing page/article numbers; also, the reference list mixes preprint and published versions (e.g., [15] and [44] are the same paper). Please harmonize.
  5. [Abstract] The abstract says '9300 electronic health records (EHRs)'; Section 3 correctly notes that these are 9,300 records from 1,480 participants with repeated visits. The wording in the abstract is slightly misleading and should explicitly mention the longitudinal structure.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical benchmark, not a derivation, and no result reduces to its inputs by construction.

full rationale

The paper makes no formal derivation; its central claim is an empirical benchmark of ChatGPT on ADNI records. The load-bearing 'multi-shot' accuracy (Table 5) is a measurement, not a fitted quantity: 100 labeled examples per class are placed in the prompt, which is the standard mechanism of in-context learning, and the reported metrics are computed on a separate uploaded file (Multi_Shot_ADNI.csv). No equation in the paper defines the target metric in terms of the prompt examples, and no fitted parameter is renamed as a prediction. Self-citations [4]-[6] are contextual and not load-bearing. A methodological limitation—that the paper does not explicitly document participant-level disjointness between the 300 in-context examples and the evaluated records—would be a validity or leakage concern, not circularity, because there is no construction-level identity between the examples and the scored labels. Therefore no circular step can be exhibited under the required standard, and the appropriate finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The evaluation depends on ADNI labels as ground truth, on the selected features being sufficient for diagnosis, on treating ChatGPT's self-reported confidence as a probability, and on the assumption that in-context example records are independent of test records at the participant level. No new physical or mathematical entities are introduced.

free parameters (2)
  • confidence threshold = 0.75
    Evaluated at 25%, 50%, 75% on the test data and then reported as the optimal threshold in the conclusion.
  • number of in-context examples per class = 100
    Chosen without ablation; the multi-shot accuracy depends on this arbitrary count.
assumptions (4)
  • domain assumption ADNI clinical labels (NC, MCI, AD) are correct gold-standard diagnoses
    The evaluation treats the DX column as ground truth (Section 3 and prompt in Figure 1).
  • domain assumption The selected tabular features are sufficient for AD diagnosis
    Only 7 MRI volumes and 12 cognitive test scores are used; no history, biomarkers, or imaging interpretation (Section 3, Figure 1).
  • domain assumption ChatGPT's self-reported confidence score behaves like a probability
    ECE/MCE and threshold filtering rely on the 0-100% confidence output as a calibrated probability (Section 5.1).
  • domain assumption Multi-shot example records and evaluated records are independent at the participant level
    No participant-level split is described; without this assumption, accuracy can be inflated (Sections 3 and 5.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Can ChatGPT Diagnose Alzheimer's Disease?." pith.science (2026). https://pith.science/paper/QKN4G3FE

@misc{pith2026250206907,
  author       = {Pith},
  title        = {Pith review of: Can ChatGPT Diagnose Alzheimer's Disease?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QKN4G3FE}},
  note         = {Machine review of arXiv:2502.06907}
}
read the original abstract

Can ChatGPT diagnose Alzheimer's Disease (AD)? AD is a devastating neurodegenerative condition that affects approximately 1 in 9 individuals aged 65 and older, profoundly impairing memory and cognitive function. This paper utilises 9300 electronic health records (EHRs) with data from Magnetic Resonance Imaging (MRI) and cognitive tests to address an intriguing question: As a general-purpose task solver, can ChatGPT accurately detect AD using EHRs? We present an in-depth evaluation of ChatGPT using a black-box approach with zero-shot and multi-shot methods. This study unlocks ChatGPT's capability to analyse MRI and cognitive test results, as well as its potential as a diagnostic tool for AD. By automating aspects of the diagnostic process, this research opens a transformative approach for the healthcare system, particularly in addressing disparities in resource-limited regions where AD specialists are scarce. Hence, it offers a foundation for a promising method for early detection, supporting individuals with timely interventions, which is paramount for Quality of Life (QoL).

Figures

Figures reproduced from arXiv: 2502.06907 by the authors.

Figure 1
Figure 1. The Workflow of Exploring ChatGPT’s Potential in Diagnosing AD. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Visualisation of Performance Metrics of Zero-Shot and Multi-Shot Prompting with ChatGPT for Detecting [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Visualisation of Calibration Metrics of Zero-Shot and Multi-Shot Prompting with ChatGPT for Detecting AD. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Accurate Samples with Different Thresholds from Zero-Shot and Multi-Shot Prompting for Detecting AD. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Accurate Samples with Confidence Scores (%) Distribution from Zero-Shot and Multi-Shot Prompting for [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 55 canonical work pages

  1. [1]

    Dementia,

    World Health Organization, “Dementia,” 2023. Accessed: 2024-08-19. Available at: https://www.who.int/ news-room/fact-sheets/detail/dementia

    work page 2023

  2. [2]

    2023 alzheimer’s disease facts and figures,

    “2023 alzheimer’s disease facts and figures,” Alzheimer’s & Dementia, vol. 19, no. 4, pp. 1598–1695, 2023

    work page 2023

  3. [3]

    Prevalence of alzheimer’s disease and vascular dementia: association with education. the rotterdam study,

    A. Ott, M. M. Breteler, F. Van Harskamp, J. J. Claus, T. J. Van Der Cammen, D. E. Grobbee, and A. Hofman, “Prevalence of alzheimer’s disease and vascular dementia: association with education. the rotterdam study,”Bmj, vol. 310, no. 6985, pp. 970–973, 1995

    work page 1995

  4. [4]

    Fairad-xai: Evaluation framework for explainable ai methods in alzheimer’s disease detection with fairness-in-the-loop,

    Q.-T. Nguyen, L. Le, X.-T. Tran, T. Do, and C.-T. Lin, “Fairad-xai: Evaluation framework for explainable ai methods in alzheimer’s disease detection with fairness-in-the-loop,” in Companion of the 2024 on ACM International Joint Conference on Pervasive and Ubiquitous Computing, pp. 870–876, 2024

    work page 2024

  5. [5]

    EEG-SSM: Leveraging State-Space Model for Dementia Detection,

    X.-T. Tran, L. Le, Q. T. Nguyen, T. Do, and C.-T. Lin, “EEG-SSM: Leveraging State-Space Model for Dementia Detection,” 2024

    work page 2024

  6. [6]

    Advancing early alzheimer’s disease detection in underdeveloped areas with fair explainable ai methods,

    Q.-T. Nguyen, “Advancing early alzheimer’s disease detection in underdeveloped areas with fair explainable ai methods,” in Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, vol. 7, pp. 47–49, 2024

    work page 2024

  7. [7]

    Towards a future where alzheimer’s disease pathology is stopped before the onset of dementia,

    W. M. van der Flier, M. E. de Vugt, E. M. Smets, M. Blom, and C. E. Teunissen, “Towards a future where alzheimer’s disease pathology is stopped before the onset of dementia,” Nature aging, vol. 3, no. 5, pp. 494–505, 2023

    work page 2023

  8. [8]

    Timely diagnosis for alzheimer’s disease: a literature review on benefits and challenges,

    B. Dubois, A. Padovani, P. Scheltens, A. Rossi, and G. Dell’Agnello, “Timely diagnosis for alzheimer’s disease: a literature review on benefits and challenges,” Journal of Alzheimer’s disease, vol. 49, no. 3, pp. 617–631, 2016

    work page 2016

Show all 62 references
  1. [9]

    Early recognition and treatment of neuropsychiatric symptoms to improve quality of life in early alzheimer’s disease: Protocol of the beat-it study,

    W. S. Eikelboom, E. Singleton, E. Van Den Berg, M. Coesmans, F. Mattace Raso, R. L. Van Bruchem, J. A. Goudzwaard, F. J. De Jong, M. Koopmanschap, T. Den Heijer, et al., “Early recognition and treatment of neuropsychiatric symptoms to improve quality of life in early alzheimer...

    2019

  2. [10]

    Duration of preclinical, prodromal, and dementia stages of alzheimer’s disease in relation to age, sex, and apoe genotype,

    L. Vermunt, S. A. Sikkes, A. Van Den Hout, R. Handels, I. Bos, W. M. Van Der Flier, S. Kern, P.-J. Ousset, P. Maruff, I. Skoog, et al., “Duration of preclinical, prodromal, and dementia stages of alzheimer’s disease in relation to age, sex, and apoe genotype,” Alzheimer’s & De...

    2019

  3. [11]

    Mild cognitive impairment and its management in older people,

    S. A. Eshkoor, T. A. Hamid, C. Y . Mun, and C. K. Ng, “Mild cognitive impairment and its management in older people,” Clinical interventions in aging, pp. 687–693, 2015

    2015

  4. [12]

    Chatgpt,

    OpenAI, “Chatgpt,” 2023. Accessed: 2024-11-25

    2023

  5. [13]

    Model card and evaluations for claude models,

    C. Models, “Model card and evaluations for claude models,” 2023

    2023

  6. [14]

    Sparks of artificial general intelligence: Early experiments with gpt-4,

    S. Bubeck, V . Chandrasekaran, R. Eldan, J. Gehrke, E. Horvitz, E. Kamar, P. Lee, Y . T. Lee, Y . Li, S. Lundberg, et al., “Sparks of artificial general intelligence: Early experiments with gpt-4,” arXiv preprint arXiv:2303.12712, 2023

    2023 arXiv

  7. [15]

    Language models are few-shot learners,

    T. B. Brown, “Language models are few-shot learners,” arXiv preprint arXiv:2005.14165, 2020

    2005 arXiv

  8. [16]

    Toolformer: Language models can teach themselves to use tools,

    T. Schick, J. Dwivedi-Yu, R. Dessì, R. Raileanu, M. Lomeli, E. Hambro, L. Zettlemoyer, N. Cancedda, and T. Scialom, “Toolformer: Language models can teach themselves to use tools,”Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  9. [17]

    A word sense disambiguation method applied to natural language processing for the portuguese language,

    C. H. do Nascimento, V . C. Garcia, and R. de Andrade Araújo, “A word sense disambiguation method applied to natural language processing for the portuguese language,” IEEE Open Journal of the Computer Society, vol. 5, pp. 268–277, 2024

    2024

  10. [18]

    Cd-llmcars: Cross domain fine-tuned large language model for context-aware recommender systems,

    A. A. Cheema, M. S. Sarfraz, U. Habib, Q. uz Zaman, and E. Boonchieng, “Cd-llmcars: Cross domain fine-tuned large language model for context-aware recommender systems,” IEEE Open Journal of the Computer Society, 2024. 11 Can ChatGPT Diagnose Alzheimer’s?

    2024

  11. [19]

    Is chatgpt a good translator? yes with gpt-4 as the engine,

    W. Jiao, W. Wang, J.-t. Huang, X. Wang, S. Shi, and Z. Tu, “Is chatgpt a good translator? yes with gpt-4 as the engine,” arXiv preprint arXiv:2301.08745, 2023

    2023 arXiv

  12. [20]

    A multitask, multilingual, multimodal evaluation of chatgpt on reasoning, hallucination, and interactivity,

    Y . Bang, S. Cahyawijaya, N. Lee, W. Dai, D. Su, B. Wilie, H. Lovenia, Z. Ji, T. Yu, W. Chung,et al., “A multitask, multilingual, multimodal evaluation of chatgpt on reasoning, hallucination, and interactivity,”In Proceedings of the 13th International Joint Conference on Natur...

    2023

  13. [21]

    Is chatgpt a general-purpose natural language processing task solver?,

    C. Qin, A. Zhang, Z. Zhang, J. Chen, M. Yasunaga, and D. Yang, “Is chatgpt a general-purpose natural language processing task solver?,” in Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp. 1339–1384, 2023

    2023

  14. [22]

    Generative agents: Interactive simulacra of human behavior,

    J. S. Park, J. O’Brien, C. J. Cai, M. R. Morris, P. Liang, and M. S. Bernstein, “Generative agents: Interactive simulacra of human behavior,” in Proceedings of the 36th annual acm symposium on user interface software and technology, pp. 1–22, 2023

    2023

  15. [23]

    Survey of hallucination in natural language generation,

    Z. Ji, N. Lee, R. Frieske, T. Yu, D. Su, Y . Xu, E. Ishii, Y . J. Bang, A. Madotto, and P. Fung, “Survey of hallucination in natural language generation,” ACM Computing Surveys, vol. 55, no. 12, pp. 1–38, 2023

    2023

  16. [24]

    Inference-time intervention: Eliciting truthful answers from a language model,

    K. Li, O. Patel, F. Viégas, H. Pfister, and M. Wattenberg, “Inference-time intervention: Eliciting truthful answers from a language model,” Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  17. [25]

    Language models (mostly) know what they know,

    S. Kadavath, T. Conerly, A. Askell, T. Henighan, D. Drain, E. Perez, N. Schiefer, Z. Hatfield-Dodds, N. DasSarma, E. Tran-Johnson, et al., “Language models (mostly) know what they know,”arXiv preprint arXiv:2207.05221, 2022

    2022 arXiv

  18. [26]

    Teaching models to express their uncertainty in words,

    S. Lin, J. Hilton, and O. Evans, “Teaching models to express their uncertainty in words,” arXiv preprint arXiv:2205.14334, 2022

    2022 arXiv

  19. [27]

    Does chatgpt know that it does not know? evaluating the black-box calibration of chatgpt,

    Y . Yuan, W. Wang, Q. Guo, Y . Xiong, C. Shen, and P. He, “Does chatgpt know that it does not know? evaluating the black-box calibration of chatgpt,” inProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-...

    2024

  20. [28]

    Few-shot fairness: Unveiling llm’s potential for fairness-aware classification,

    G. Chhikara, A. Sharma, K. Ghosh, and A. Chakraborty, “Few-shot fairness: Unveiling llm’s potential for fairness-aware classification,” arXiv preprint arXiv:2402.18502, 2024

    2024 arXiv

  21. [29]

    Using chatgpt for clinical practice and medical education: cross-sectional survey of medical students’ and physicians’ perceptions,

    P. Tangadulrat, S. Sono, B. Tangtrakulwanich,et al., “Using chatgpt for clinical practice and medical education: cross-sectional survey of medical students’ and physicians’ perceptions,” JMIR Medical Education, vol. 9, no. 1, p. e50658, 2023

    2023

  22. [30]

    “doctor chatgpt, can you help me?

    J. Armbruster, F. Bussmann, C. Rothhaas, N. Titze, P. A. Grützner, and H. Freischmidt, ““doctor chatgpt, can you help me?” the patient’s perspective: Cross-sectional study,”Journal of Medical Internet Research, vol. 26, p. e58831, 2024

    2024

  23. [31]

    Exploring the role of chatgpt in patient care (diagnosis and treatment) and medical research: A systematic review,

    R. K. Garg, V . L. Urs, A. A. Agarwal, S. K. Chaudhary, V . Paliwal, and S. K. Kar, “Exploring the role of chatgpt in patient care (diagnosis and treatment) and medical research: A systematic review,”Health Promotion Perspectives, vol. 13, no. 3, p. 183, 2023

    2023

  24. [32]

    Diagnostic accuracy of differential-diagnosis lists generated by generative pretrained transformer 3 chatbot for clinical vignettes with common chief complaints: a pilot study,

    T. Hirosawa, Y . Harada, M. Yokose, T. Sakamoto, R. Kawamura, and T. Shimizu, “Diagnostic accuracy of differential-diagnosis lists generated by generative pretrained transformer 3 chatbot for clinical vignettes with common chief complaints: a pilot study,” International journa...

    2023

  25. [33]

    Assessing the utility of chatgpt throughout the entire clinical workflow: development and usability study,

    A. Rao, M. Pang, J. Kim, M. Kamineni, W. Lie, A. K. Prasad, A. Landman, K. Dreyer, and M. D. Succi, “Assessing the utility of chatgpt throughout the entire clinical workflow: development and usability study,”Journal of Medical Internet Research, vol. 25, p. e48659, 2023

    2023

  26. [34]

    Evaluating gpt as an adjunct for radiologic decision making: Gpt-4 versus gpt-3.5 in a breast imaging pilot,

    A. Rao, J. Kim, M. Kamineni, M. Pang, W. Lie, K. J. Dreyer, and M. D. Succi, “Evaluating gpt as an adjunct for radiologic decision making: Gpt-4 versus gpt-3.5 in a breast imaging pilot,” Journal of the American College of Radiology, vol. 20, no. 10, pp. 990–997, 2023

    2023

  27. [35]

    Chatgpt as a diagnostic aid in alzheimer’s disease: an exploratory study,

    M. El Haj, C. Boutoleau-Bretonnière, K. Gallouj, N. Wagemann, P. Antoine, D. Kapogiannis, and G. Chapelet, “Chatgpt as a diagnostic aid in alzheimer’s disease: an exploratory study,”Journal of Alzheimer’s Disease Reports, vol. 8, no. 1, pp. 495–500, 2024

    2024

  28. [36]

    Geriatrics workforce by the numbers

    American Geriatrics Society, “Geriatrics workforce by the numbers.” https://www.americangeriatrics. org/geriatrics-profession/about-geriatrics/geriatrics-workforce-numbers . Accessed: 2024- 12-11

    2024

  29. [37]

    Geographic variation in shortfalls of dementia specialists in the united states,

    J. L. Liu, L. Baker, A. Y .-A. Chen, and J. Wang, “Geographic variation in shortfalls of dementia specialists in the united states,” Health Affairs Scholar, vol. 2, no. 7, p. qxae088, 2024. 12 Can ChatGPT Diagnose Alzheimer’s?

    2024

  30. [38]

    Assessing the quality of chatgpt responses to dementia caregivers’ questions: Qualitative analysis,

    A. Aguirre, R. Hilsabeck, T. Smith, B. Xie, D. He, Z. Wang, and N. Zou, “Assessing the quality of chatgpt responses to dementia caregivers’ questions: Qualitative analysis,” JMIR aging, vol. 7, p. e53019, 2024

    2024

  31. [39]

    Chatgpt vs google for queries related to dementia and other cognitive decline: comparison of results,

    V . Hristidis, N. Ruggiano, E. L. Brown, S. R. R. Ganta, and S. Stewart, “Chatgpt vs google for queries related to dementia and other cognitive decline: comparison of results,” Journal of Medical Internet Research, vol. 25, pp. e489664, publisher=JMIR Publications Toronto, Canada

  32. [40]

    Leveraging chatgpt-like large language models for alzheimer’s disease: Enhancing care, advancing research, and overcoming challenges,

    S. Thapa and S. Adhikari, “Leveraging chatgpt-like large language models for alzheimer’s disease: Enhancing care, advancing research, and overcoming challenges,” inSmart Healthcare Systems, pp. 265–275, CRC Press, 2024

    2024

  33. [41]

    Adni data,

    “Adni data,” 2024. Accessed: 2024-12-11, https://adni.loni.usc.edu/data-samples/adni-data/

    2024

  34. [42]

    The alzheimer’s disease neuroimaging initiative (adni): Mri methods,

    C. R. Jack Jr, M. A. Bernstein, N. C. Fox, P. Thompson, G. Alexander, D. Harvey, B. Borowski, P. J. Britson, J. L. Whitwell, C. Ward, et al., “The alzheimer’s disease neuroimaging initiative (adni): Mri methods,”Journal of Magnetic Resonance Imaging: An Official Journal of the...

    2008

  35. [43]

    Ppad: A deep learning architecture to predict progression of alzheimer’s disease,

    M. Al Olaimat, J. Martinez, F. Saeed, S. Bozdag, and A. D. N. Initiative, “Ppad: A deep learning architecture to predict progression of alzheimer’s disease,”Bioinformatics, vol. 39, no. Supplement_1, pp. i149–i157, 2023

    2023

  36. [44]

    Language models are few-shot learners,

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, and e. a. Sastry, “Language models are few-shot learners,” in Advances in Neural Information Processing Systems (H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, eds.), v...

    1901

  37. [45]

    Health-llm: Large language models for health prediction via wearable sensor data,

    Y . Kim, X. Xu, D. McDuff, C. Breazeal, and H. W. Park, “Health-llm: Large language models for health prediction via wearable sensor data,” in Proceedings of the fifth Conference on Health, Inference, and Learning(T. Pollard, E. Choi, P. Singhal, M. Hughes, E. Sizikova, B. Mor...

    2024

  38. [46]

    The accuracy and appropriateness of chatgpt responses on nonmelanoma skin cancer information using zero-shot chain of thought prompting,

    R. O’Hagan, D. Poplausky, J. N. Young, N. Gulati, M. Levoska, B. Ungar, and J. Ungar, “The accuracy and appropriateness of chatgpt responses on nonmelanoma skin cancer information using zero-shot chain of thought prompting,” JMIR dermatology, vol. 6, p. e49889, 2023

    2023

  39. [47]

    Zero-shot information extraction from radiological reports using chatgpt,

    D. Hu, B. Liu, X. Zhu, X. Lu, and N. Wu, “Zero-shot information extraction from radiological reports using chatgpt,” International Journal of Medical Informatics, vol. 183, p. 105321, 2024

    2024

  40. [48]

    Innovation and application of large language models (llms) in dentistry–a scoping review,

    F. Umer, I. Batool, and N. Naved, “Innovation and application of large language models (llms) in dentistry–a scoping review,”Nature BDJ open, vol. 10, no. 1, p. 90, 2024

    2024

  41. [49]

    Multi-prompts learning with cross-modal alignment for attribute-based person re-identification,

    Y . Zhai, Y . Zeng, Z. Huang, Z. Qin, X. Jin, and D. Cao, “Multi-prompts learning with cross-modal alignment for attribute-based person re-identification,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, pp. 6979–6987, 2024

    2024

  42. [50]

    explainable ai with gpt4 for story analysis and generation: A novel framework for diachronic sentiment analysis,

    J. Chun and K. Elkins, “explainable ai with gpt4 for story analysis and generation: A novel framework for diachronic sentiment analysis,” International Journal of Digital Humanities, vol. 5, no. 2, pp. 507–532, 2023

    2023

  43. [51]

    Large language models in cryptocurrency securities cases: can a gpt model meaningfully assist lawyers?,

    A. Trozze, T. Davies, and B. Kleinberg, “Large language models in cryptocurrency securities cases: can a gpt model meaningfully assist lawyers?,” Artificial Intelligence and Law, pp. 1–47, 2024

    2024

  44. [52]

    Inclusive prompt engineering: A methodology for hacking biased ai image generation,

    R. Skilton and A. Cardinal, “Inclusive prompt engineering: A methodology for hacking biased ai image generation,” in Proceedings of the 42nd ACM International Conference on Design of Communication, pp. 76–80, 2024

    2024

  45. [53]

    On evaluation metrics for medical applications of artificial intelligence,

    S. A. Hicks, I. Strümke, V . Thambawita, M. Hammou, M. A. Riegler, P. Halvorsen, and S. Parasa, “On evaluation metrics for medical applications of artificial intelligence,” Scientific reports, vol. 12, no. 1, p. 5979, 2022

    2022

  46. [54]

    Obtaining well calibrated probabilities using bayesian binning,

    M. P. Naeini, G. Cooper, and M. Hauskrecht, “Obtaining well calibrated probabilities using bayesian binning,” in Proceedings of the AAAI conference on artificial intelligence, vol. 29, 2015

    2015

  47. [55]

    Predicting good probabilities with supervised learning,

    A. Niculescu-Mizil and R. Caruana, “Predicting good probabilities with supervised learning,” in Proceedings of the 22nd international conference on Machine learning, pp. 625–632, 2005

    2005

  48. [56]

    Measuring calibration in deep learning.,

    J. Nixon, M. W. Dusenberry, L. Zhang, G. Jerfel, and D. Tran, “Measuring calibration in deep learning.,” inCVPR, vol. 2, 2019

    2019

  49. [57]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat, et al., “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  50. [58]

    Quality of life in alzheimer disease,

    M. P. Lawton, “Quality of life in alzheimer disease,”Alzheimer Disease & Associated Disorders, vol. 8, pp. 138– 150, 1994. 13 Can ChatGPT Diagnose Alzheimer’s?

    1994

  51. [59]

    Is chatgpt fair for recommendation? evaluating fairness in large language model recommendation,

    J. Zhang, K. Bao, Y . Zhang, W. Wang, F. Feng, and X. He, “Is chatgpt fair for recommendation? evaluating fairness in large language model recommendation,” in Proceedings of the 17th ACM Conference on Recommender Systems, pp. 993–999, 2023

    2023

  52. [60]

    Quttainah, V

    M. Quttainah, V . Mishra, S. Madakam, Y . Lurie, S. Mark,et al., “Cost, usability, credibility, fairness, accountability, transparency, and explainability framework for safe and effective large language models in medical education: Narrative review and qualitative study,”JMIR ...

    2024

  53. [61]

    Systematic analysis of chatgpt, google search and llama 2 for clinical decision support tasks,

    S. Sandmann, S. Riepenhausen, L. Plagwitz, and J. Varghese, “Systematic analysis of chatgpt, google search and llama 2 for clinical decision support tasks,” Nature Communications, vol. 15, no. 1, p. 2050, 2024

    2024

  54. [62]

    Exploring ai-chatbots’ capability to suggest surgical planning in ophthalmology: Chatgpt versus google gemini analysis of retinal detachment cases,

    M. M. Carlà, G. Gambini, A. Baldascino, F. Giannuzzi, F. Boselli, E. Crincoli, N. C. D’Onofrio, and S. Rizzo, “Exploring ai-chatbots’ capability to suggest surgical planning in ophthalmology: Chatgpt versus google gemini analysis of retinal detachment cases,” British Journal o...

    2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.