Pith. sign in

REVIEW 4 major objections 6 minor 142 references

Open Foundation Models in Healthcare: Challenges, Paradoxes, and Opportunities with GenAI Driven Personalized Prescription

T0 review · 4 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Open LLMs match GPT-4 on prescription checks when grounded with RAG

desk verdict A useful survey and a plausible but statistically underpowered case study; the RAG-benefit claim is not supported by the paper's own Table V. read the letter →

arxiv 2502.04356 v1 pith:DWJAWTLC submitted 2025-02-04 cs.CL cs.AIcs.LG

Mahdi Alkaeed , Sofiat Abioye , Adnan Qayyum , Yosra Magdi Mekki , Ilhem Berrou , Mohamad Abdallah , Ala Al-Fuqaha , Muhammad Bilal
show 1 more author
Junaid Qadir
This is my paper · ORCID
classification cs.CLcs.AIcs.LG
keywords open-sourceLLMshealthcarefoundationmodelsretrieval-augmentedgenerationpersonalizedprescriptionadversedrugreactionsMIMIC-IIIsummaryofproductcharacteristicsLLMevaluation
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

This paper argues that open-weight large language models, though weaker than proprietary ones out of the box, can reach comparable performance on a demanding clinical task when augmented with retrieval of the drug's official product information. To make the case, it surveys the landscape of open healthcare foundation models and runs a case study where five models (LLaMA-2, LLaMA-3, Mistral, Meditron, and GPT-4) judge whether a drug fits a patient across eight interaction classes such as contraindications, genetics, and pregnancy. The central result is that LLaMA-3 with RAG matches or exceeds GPT-4 on several checks, for instance 0.92 versus 0.75 accuracy on contraindications, and that RAG improves most open models on most metrics. If true, this matters because open models can be run locally, inspected, and adapted, making safer AI-assisted prescribing available without dependence on proprietary APIs. The paper also reports a clinician's subjective evaluation supporting the RAG-enabled models' safety and drug-interaction detection.

What carries the argument

The load-bearing mechanism is Retrieval-Augmented Generation over a vectorized corpus of Summary of Product Characteristics (SmPC) documents, the European Union regulatory label for each drug. For each patient-drug pair, the system embeds the patient profile and query, retrieves the most relevant SmPC passages by cosine similarity, and feeds that context to the LLM, which must answer each of eight checks as Suitable, Risky, or N/A with a reason. The RAG setup supplies the factual grounding that turns general-purpose open models into reliable checkers. The evaluation machinery is a dual validation: quantitative metrics against SmPC-derived ground truth, plus a subjective five-scale scoring by an expert clinician with more than ten years of experience.

What would settle it

Run the same five-drug, 25-profile protocol with an independent pharmacist blind to the models, on a larger stratified sample from MIMIC-III or MIMIC-IV, and check whether LLaMA-3 with RAG still beats GPT-4 without RAG on contraindications and genetics; if the advantage disappears on a larger or harder set, the central claim would fail.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes a claim: open LLMs can achieve performance comparable to proprietary models when paired with grounding techniques such as RAG. Concretely, the case study shows that LLaMA-3 with RAG matches or beats GPT-4 on several prescription-suitability checks, with contraindication accuracy 0.92 versus 0.75, genetics F1 0.96 versus 1.00, and comparable scores across age, comorbidity, and warnings. This is presented as evidence that the performance gap between open and proprietary models is largely a grounding gap, not an unrecoverable capability gap. The paper further claims that grounding via RAG consistently improves accuracy and F1 for both open and closed models across almost all interaction classes, and that an expert clinician rates RAG-enabled open models as excellent on drug interaction detection and overall prescription safety.

Load-bearing premise

The result rests on the assumption that the 25 curated MIMIC-III profiles and five drugs represent the range of real prescribing decisions, and that the SmPC-derived labels are unambiguous; if either breaks, the reported accuracy cannot support the broad claim.

Editorial extensions

If this is right

  • RAG can close much of the gap between open and proprietary models on prescribing-safety checks.
  • Even proprietary models benefit from grounding, so RAG should be considered a standard component for clinical LLM deployment.
  • Open models plus RAG may be a viable path to transparent, locally deployable prescription decision support, particularly where data privacy or cost rules out cloud APIs.
  • The high scores on contraindications, genetics, and pregnancy suggest that structured regulatory text can serve as a sufficient knowledge source for many adverse-drug-reaction checks.

Reading between the lines

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

  • If the result generalizes, institutions could run their own prescription-safety checks on local hardware, keeping patient data on-premises; this implication is mine, not stated in the paper.
  • The small curated test set makes the headline margin fragile; a larger, harder, and independently labeled benchmark could either reinforce or overturn it.
  • The RAG benefit may be largest where the model's parametric knowledge is weak, so future work should stratify results by how well represented a drug is in the training corpus.
  • Grounding only on SmPC leaves open whether combining retrieval with clinical guidelines, pharmacogenomic databases, or real-world evidence would further close the gap.
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 / 6 minor

Summary. The paper has two parts. First, it surveys open-source LLMs and AI foundation models for healthcare, organizing them into a taxonomy by application area (clinical NLP, imaging, education, telehealth, etc.) and reviewing safety considerations and open datasets. Second, it presents a case study on personalized prescription assessment: 25 patient profiles derived from MIMIC-III, five medications (Warfarin, Metformin, Levothyroxine, Lisinopril, Omeprazole), and eight interaction classes (age, comorbidities, contraindications, dose, genetics, lactation, pregnancy, warnings). Five models (LLaMA-2, LLaMA-3, Mistral, Meditron, GPT-4) are evaluated with and without retrieval-augmented generation (RAG) grounded in Summary of Product Characteristics (SmPC) documents. The main claim is that open models, when paired with RAG, can achieve performance comparable to GPT-4; a secondary claim is that RAG significantly improves performance across almost all interaction types. The paper also reports a subjective evaluation by an expert clinician who is a co-author.

Significance. If the empirical claims were well supported, the case study would be a useful contribution to the growing literature on grounding LLMs for clinical decision support, and the survey portion would provide a helpful entry point to open healthcare foundation models. The paper is honest about the exploratory nature of the case study and about the co-author involvement in the subjective assessment; those disclosures are to the authors' credit. However, the central comparative claim rests on Table V, whose small sample sizes, lack of statistical inference, and internally inconsistent RAG effects currently prevent the results from being conclusive. The survey part is broad but largely derivative; its main value is as a catalog rather than as a synthesis. On balance, the significance is moderate and conditional on a substantially strengthened evaluation.

major comments (4)
  1. [§IV-C2 and Table V] The claim that RAG "significantly enhances their performance across almost all interaction types" is contradicted by several cells in Table V. For example, LLaMA-2 Lactation accuracy falls from 1.00 to 0.63, LLaMA-3 Lactation from 1.00 to 0.50, LLaMA-2 Pregnancy from 1.00 to 0.84, and Mistral Genetics F1 from 1.00 to 0.67 while accuracy drops from 1.00 to 0.95. These are large swings, not small fluctuations, and they are inconsistent with a systematic grounding benefit. The evaluation uses only 25 profiles and five drugs, so per-cell denominators are likely very small (many profiles are male or otherwise not applicable for pregnancy/lactation), and no confidence intervals, standard errors, per-cell counts, or raw outputs are provided. Without these, the accuracy values cannot support comparative statements such as "LLaMA-3 with RAG outperforms GPT-4 on contraindications (0.92 vs 0.75)" or the overall conclusion that open models become comparable to GPT-4. Please report the number of applicable cases per interaction class, provide uncertainty estimates or significance tests, and either reconcile the sharp RAG-related degradations or temper the claim to reflect per-class heterogeneity.
  2. [§IV-B1 (Data Description)] The case study draws general conclusions from 25 curated MIMIC-III profiles and five selected drugs. The manuscript does not report how the profiles were sampled from MIMIC-III, how many profiles are applicable to each interaction class, or how the five drugs were chosen. If, for instance, only four profiles are relevant to lactation, a single changed answer moves accuracy from 1.00 to 0.75; the current presentation gives no way to assess this. The representativeness of this small set is load-bearing for the paper's central claim about open models' capabilities, and the paper should at least report the distribution of patient characteristics and the applicable-sample sizes per drug and interaction class. Absent that, the results are illustrative rather than evidential.
  3. [§IV-C3 (Subjective Assessment)] The qualitative validation is performed by a single expert clinician who is also a co-author and who "has been actively involved in the project since its inception." The evaluation is not blinded, no inter-rater reliability is reported, and the criteria (MSA, DID, PSDA, PSS) are not defined operationally in a way that another clinician could apply independently. This makes the subjective scores a weak basis for the claim that RAG-enabled models "consistently achieve higher average scores." I recommend either recruiting independent clinicians, providing a pre-specified scoring rubric with example cases, or clearly labeling the qualitative results as a pilot assessment and moving the claim to the limitations section.
  4. [§IV-C2 and §IV-B3 (RAG setting)] The RAG condition retrieves from the same SmPC documents that were used to derive the ground truth for validation, making the comparison an open-book test by design. This is not circular in a hidden way, but it is a design choice that should be discussed as a limitation: improved RAG performance may partly reflect the model copying the retrieved answer rather than reasoning from the patient profile. I suggest adding a baseline that gives the same retrieved SmPC text to the model without the patient profile, or a retrieval-only baseline that outputs the SmPC statement directly, to establish that the LLM is adding value beyond information retrieval. This is particularly important because the paper's headline comparison (LLaMA-3 with RAG 0.92 vs GPT-4 with RAG 0.75 on contraindications) could otherwise be driven by retrieval sensitivity or prompt formatting rather than model capability.
minor comments (6)
  1. [§IV-C2 text] In the paragraph discussing Table V, "LLaMA-3's accuracy in handling age interactions improves from 0.85% to 0.88%" should read "0.85 to 0.88", since the table reports proportions, not percentages.
  2. [§III-A8] The text says "AI models, including CCNs, such as Flan-PaLM"; "CCNs" appears to be a typo for "CNNs", and Flan-PaLM is not a CNN, so the sentence should be rephrased.
  3. [§IV-C2 text] The phrase "LLaMA-3 maintains immersive performance" should likely be "impressive performance".
  4. [Table I] Several entries in Table I are problematic: the "Code Level" column mixes terms such as "Link APIs", "Link CR", and "Link FOS" without a complete legend, and entries like "Clinical BERT (2023)" cite the model as developed in 2023 although the original Clinical BERT was released in 2019; please verify the dates and references.
  5. [References] Reference [9] appears as an incomplete citation: "Zero-shot clinical trial patient matching with LLMs" lacks author, venue, and year; several other references (e.g., [83]) are given as arXiv preprints without the arXiv identifier, which makes verification difficult.
  6. [§IV-C3 and Figure 5] Figure 5 aggregates the subjective scores across models and profiles, but the color intensity alone cannot encode the 1–5 scale reliably; adding numerical labels or a clearer legend would improve interpretability.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity: the RAG benefit is measured against labels drawn from the same SmPC documents that RAG retrieves, making the grounding evaluation an open-book test on the answer key; the open-vs-proprietary comparison itself is not forced.

  1. self definitional [Section IV-B3 and IV-B4 (Evaluating LLMs for Prescription; Results Validation)]
    "we utilized the Retrieval Augmented Generation (RAG) technique... It computes the Cosine distance between the input query and the vectorized SmPC document of the corresponding medication to retrieve the appropriate contextual information to generate accurate responses. ... The results generated by the LLMs were validated through a dual-validation approach that involves evaluating identified risks quantitatively using the ground truth information from the SmPC and qualitative assessment."

    The correct labels for the quantitative evaluation are the SmPC-derived risk classifications, while the RAG pipeline injects the same vectorized SmPC documents as retrieval context. Thus the reported RAG benefit (e.g., LLaMA-3 contraindications accuracy rising from 0.48 to 0.92) is, by construction, an open-book test on the answer key: the model is handed the very source from which the ground-truth labels were extracted. This does not force the open-model-versus-GPT-4 comparison, because both arms use the same protocol, so the circularity is partial and is concentrated in the claim that RAG itself 'significantly enhances' performance.

full rationale

The survey/taxonomy portion is independent and not circular, and the central open-versus-proprietary model comparison (LLaMA-3 with RAG vs GPT-4 with RAG) is not forced by construction: both models are evaluated under the same open-book protocol, and the outcome could have gone either way. The only structural circularity is in the RAG-enhancement claim: the ground truth is derived from the SmPCs, and the RAG retriever supplies those same SmPCs as context, so the RAG-vs-no-RAG gap partly measures the effect of handing the model the answer source. This is transparent (the paper describes the SmPC retrieval and the SmPC-based validation explicitly), but it is still an answer-key-in-the-open-book design. No parameters are fitted, no load-bearing self-citations were found, and no uniqueness theorem is imported from the authors' prior work. The internal contradictions in Table V (e.g., LLaMA-3 Lactation accuracy falling from 1.00 to 0.50 with RAG, Mistral Genetics F1 falling from 1.00 to 0.67) and the co-author subjective assessment are validity/independence concerns, not circularity, and therefore do not further raise the circularity score.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim relies on the curation of patient profiles and SmPC-based labels rather than on fitted model parameters. No new entities are introduced. The main burden is on the representativeness of the 25 cases and the objectivity of the clinical assessment.

assumptions (5)
  • domain assumption MIMIC-III patient records can be used to construct realistic polypharmacy profiles.
    Section IV-B1 assumes the curated 25 profiles reflect real-world clinical scenarios.
  • domain assumption SmPC documents constitute authoritative ground truth for medication suitability.
    Section IV-A and IV-C2 use SmPC-derived labels as the reference for accuracy and F1.
  • domain assumption The LLM's JSON outputs can be reliably parsed and compared against ground truth labels.
    Section IV-C2 reports metrics without discussing parse failures or ambiguous outputs.
  • ad hoc to paper The expert clinician's subjective scores are an unbiased measure of clinical usefulness.
    The expert is a co-author and was involved since inception (Section IV-C3), which may introduce bias.
  • domain assumption Default LLM parameters and the same prompt template produce fair comparisons.
    Section IV-C1 states default parameters were used, but no sensitivity analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Open Foundation Models in Healthcare: Challenges, Paradoxes, and Opportunities with GenAI Driven Personalized Prescription." pith.science (2026). https://pith.science/paper/DWJAWTLC

@misc{pith2026250204356,
  author       = {Pith},
  title        = {Pith review of: Open Foundation Models in Healthcare: Challenges, Paradoxes, and Opportunities with GenAI Driven Personalized Prescription},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DWJAWTLC}},
  note         = {Machine review of arXiv:2502.04356}
}
read the original abstract

In response to the success of proprietary Large Language Models (LLMs) such as OpenAI's GPT-4, there is a growing interest in developing open, non-proprietary LLMs and AI foundation models (AIFMs) for transparent use in academic, scientific, and non-commercial applications. Despite their inability to match the refined functionalities of their proprietary counterparts, open models hold immense potential to revolutionize healthcare applications. In this paper, we examine the prospects of open-source LLMs and AIFMs for developing healthcare applications and make two key contributions. Firstly, we present a comprehensive survey of the current state-of-the-art open-source healthcare LLMs and AIFMs and introduce a taxonomy of these open AIFMs, categorizing their utility across various healthcare tasks. Secondly, to evaluate the general-purpose applications of open LLMs in healthcare, we present a case study on personalized prescriptions. This task is particularly significant due to its critical role in delivering tailored, patient-specific medications that can greatly improve treatment outcomes. In addition, we compare the performance of open-source models with proprietary models in settings with and without Retrieval-Augmented Generation (RAG). Our findings suggest that, although less refined, open LLMs can achieve performance comparable to proprietary models when paired with grounding techniques such as RAG. Furthermore, to highlight the clinical significance of LLMs-empowered personalized prescriptions, we perform subjective assessment through an expert clinician. We also elaborate on ethical considerations and potential risks associated with the misuse of powerful LLMs and AIFMs, highlighting the need for a cautious and responsible implementation in healthcare.

Figures

Figures reproduced from arXiv: 2502.04356 by the authors.

Figure 1
Figure 1. Levels of Open-Source Access for LLMs and AIFMs. The figure shows the spectrum from Fully Closed Source to Fully Open Source, detailing key aspects and descriptions for each level. (Adapted from [21]). of open-source AI models, encompassing everything from training data to the model itself. In contrast, Kapoor et al. [29] argued that open access to model weights alone suffices to classify a model as open AI. However… view at source ↗
Figure 2
Figure 2. Applications of LLMs in Healthcare. This figure shows six key areas: Clinical Documentation, Disease Prediction and Diagnosis, Drug Discovery, Virtual Health Assistants, NLP in EHRs, and Telemedicine Support. clinical decision support, extracting phenotypic information for research, detecting adverse drug events, expediting clin￾ical trial recruitment, understanding temporal relationships in patient records, de-iden… view at source ↗
Figure 3
Figure 3. Structure and Purpose of the SmPC (Summary of Product Characteristics) [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Illustration of the proposed methodology for evaluating open and closed-source LLMs using two settings, i.e., with and without Retrieval Augmented [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Comparative analysis of the qualitative assessment scores across different metrics for each patient profile. Each subplot represents a specific metric, [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

142 extracted references · 33 canonical work pages

  1. [1]

    Heterogeneous data and big data analytics,

    L. Wang, “Heterogeneous data and big data analytics,” Automatic Control and Information Sciences , vol. 3, no. 1, pp. 8–15, 2017

    2017

  2. [2]

    Language models are unsupervised multitask learners,

    A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, I. Sutskever et al., “Language models are unsupervised multitask learners,” OpenAI blog, vol. 1, no. 8, p. 9, 2019

    2019

  3. [3]

    Language models are few-shot learners,

    B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhariwal, A. Neelakan- tan, P. Shyam, G. Sastry, A. Askell, S. Agarwal et al. , “Language models are few-shot learners,” arXiv preprint arXiv:2005.14165, 2020

    arXiv 2005

  4. [4]

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity,

    W. Fedus, B. Zoph, and N. Shazeer, “Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity,” The Journal of Machine Learning Research, vol. 23, no. 1, pp. 5232–5270, 2022

    2022

  5. [5]

    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

    arXiv 2023

  6. [6]

    GLUE: A multi-task benchmark and analysis platform for natural language understanding,

    A. Wang, A. Singh, J. Michael, F. Hill, O. Levy, and S. R. Bowman, “GLUE: A multi-task benchmark and analysis platform for natural language understanding,” arXiv preprint arXiv:1804.07461 , 2018

    arXiv 2018

  7. [7]

    Winogrande: An adversarial winograd schema challenge at scale,

    K. Sakaguchi, R. L. Bras, C. Bhagavatula, and Y . Choi, “Winogrande: An adversarial winograd schema challenge at scale,” Communications of the ACM , vol. 64, no. 9, pp. 99–106, 2021

    2021

  8. [8]

    Language mod- els are few-shot learners,

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell et al., “Language mod- els are few-shot learners,” Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020

    1901

Show all 142 references
  1. [9]

    Zero-shot clinical trial patient matching with LLMs

    “Zero-shot clinical trial patient matching with LLMs.”

  2. [10]

    Recall and learn: Fine-tuning deep pretrained language models with less forgetting,

    S. Chen, Y . Hou, Y . Cui, W. Che, T. Liu, and X. Yu, “Recall and learn: Fine-tuning deep pretrained language models with less forgetting,” arXiv preprint arXiv:2004.12651 , 2020

    2004 arXiv

  3. [11]

    How fine-tuning allows for effective meta-learning,

    K. Chua, Q. Lei, and J. D. Lee, “How fine-tuning allows for effective meta-learning,” Advances in Neural Information Processing Systems , vol. 34, pp. 8871–8884, 2021

    2021

  4. [12]

    Robust fine-tuning of deep neural net- works with hessian-based generalization guarantees,

    H. Ju, D. Li, and H. R. Zhang, “Robust fine-tuning of deep neural net- works with hessian-based generalization guarantees,” in International Conference on Machine Learning . PMLR, 2022, pp. 10 431–10 461

    2022

  5. [13]

    Lexfit: Lexical fine-tuning of pretrained language models,

    I. Vuli ´c, E. M. Ponti, A. Korhonen, and G. Glava ˇs, “Lexfit: Lexical fine-tuning of pretrained language models,” in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Proces...

    2021

  6. [14]

    Revisit- ing few-sample BERT fine-tuning,

    T. Zhang, F. Wu, A. Katiyar, K. Q. Weinberger, and Y . Artzi, “Revisit- ing few-sample BERT fine-tuning,” arXiv preprint arXiv:2006.05987 , 2020

    2006 arXiv

  7. [15]

    Open- sourcing highly capable foundation models: An evaluation of risks, benefits, and alternative methods for pursuing open-source objectives,

    E. Seger, N. Dreksler, R. Moulange, E. Dardaman, J. Schuett, K. Wei, C. Winter, M. Arnold, S. ´O. h ´Eigeartaigh, A. Korinek et al. , “Open- sourcing highly capable foundation models: An evaluation of risks, benefits, and alternative methods for pursuing open-source objectives...

    2023 arXiv

  8. [16]

    Position paper: On the societal impact of open foundation models,

    S. Kapoor, R. Bommasani, K. Klyman, S. Longpre, A. Ramaswami, P. Cihon, A. K. Hopkins, K. Bankston, S. Biderman, M. Bogen et al., “Position paper: On the societal impact of open foundation models,” in Forty-first International Conference on Machine Learning

  9. [17]

    Foundation models in healthcare: Opportunities, risks & strategies forward,

    A. Thieme, A. Nori, M. Ghassemi, R. Bommasani, T. O. Andersen, and E. Luger, “Foundation models in healthcare: Opportunities, risks & strategies forward,” in Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems , 2023, pp. 1–4

    2023

  10. [18]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

    2017

  11. [19]

    A survey of large language models for healthcare: from data, tech- nology, and applications to accountability and ethics,

    K. He, R. Mao, Q. Lin, Y . Ruan, X. Lan, M. Feng, and E. Cambria, “A survey of large language models for healthcare: from data, tech- nology, and applications to accountability and ethics,” arXiv preprint arXiv:2310.05694, 2023

    arXiv

  12. [20]

    On the opportunities and risks of foundation models,

    R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, S. von Arx, M. S. Bernstein, J. Bohg, A. Bosselut, E. Brunskill et al. , “On the opportunities and risks of foundation models,” arXiv preprint arXiv:2108.07258, 2021

    2021 arXiv

  13. [21]

    The gradient of generative AI release: Methods and considerations,

    I. Solaiman, “The gradient of generative AI release: Methods and considerations,” in Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency, 2023, pp. 111–122

    2023

  14. [22]

    Code llama: Open foundation models for code,

    B. Roziere, J. Gehring, F. Gloeckle, S. Sootla, I. Gat, X. E. Tan, Y . Adi, J. Liu, T. Remez, J. Rapin et al., “Code llama: Open foundation models for code,” arXiv preprint arXiv:2308.12950 , 2023

    2023 arXiv

  15. [23]

    Competition in generative artificial intelligence founda- tion models,

    C. Carugati, “Competition in generative artificial intelligence founda- tion models,” Bruegel Working Paper, Tech. Rep., 2023

    2023

  16. [24]

    D2. 6: Ontology of licencing, ownership and conditions of use (v1. 0),

    M. Gurrieri, E. Daga, A. Scharnhorst, and J. Carvalho, “D2. 6: Ontology of licencing, ownership and conditions of use (v1. 0),” Ph.D. dissertation, Open University, 2023

    2023

  17. [25]

    Open-source large language models outperform crowd workers and approach ChatGPT in text-annotation tasks,

    M. Alizadeh, M. Kubli, Z. Samei, S. Dehghani, J. D. Bermeo, M. Korobeynikova, and F. Gilardi, “Open-source large language models outperform crowd workers and approach ChatGPT in text-annotation tasks,” arXiv preprint arXiv:2307.02179 , 2023

    2023 arXiv

  18. [26]

    The open source definition (osd),

    Open Source Initiative, “The open source definition (osd),” 1998, accessed: 2024-12-01. [Online]. Available: https://opensource.org/ history

    1998

  19. [27]

    DiBona and S

    C. DiBona and S. Ockman, Open sources: Voices from the open source revolution. ” O’Reilly Media, Inc.”, 1999

    1999

  20. [28]

    Llama 2: Open foundation and fine-tuned chat models,

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale et al. , “Llama 2: Open foundation and fine-tuned chat models,” arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  21. [29]

    On the societal impact of open foundation models,

    S. Kapoor, R. Bommasani, K. Klyman, S. Longpre, A. Ramaswami, P. Cihon, A. Hopkins, K. Bankston, S. Biderman, M. Bogen et al., “On the societal impact of open foundation models,” 2024

    2024

  22. [30]

    Accelerating LLM inference with staged speculative decoding,

    B. Spector and C. Re, “Accelerating LLM inference with staged speculative decoding,” arXiv preprint arXiv:2308.04623 , 2023

    2023 arXiv

  23. [31]

    Privacy- preserving artificial intelligence in healthcare: Techniques and applica- tions,

    N. Khalid, A. Qayyum, M. Bilal, A. Al-Fuqaha, and J. Qadir, “Privacy- preserving artificial intelligence in healthcare: Techniques and applica- tions,” Computers in Biology and Medicine , p. 106848, 2023

    2023

  24. [32]

    Training data leakage analysis in language models,

    H. A. Inan, O. Ramadan, L. Wutschitz, D. Jones, V . R ¨uhle, J. Withers, and R. Sim, “Training data leakage analysis in language models,” arXiv preprint arXiv:2101.05405, 2021

    2021 arXiv

  25. [33]

    Extracting training data from large language models,

    N. Carlini, F. Tramer, E. Wallace, M. Jagielski, A. Herbert-V oss, K. Lee, A. Roberts, T. Brown, D. Song, U. Erlingssonet al., “Extracting training data from large language models,” in 30th USENIX Security Symposium (USENIX Security 21) , 2021, pp. 2633–2650

    2021

  26. [34]

    Privacy-preserving artificial intelligence techniques in biomedicine,

    R. Torkzadehmahani, R. Nasirigerdeh, D. B. Blumenthal, T. Kacprowski, M. List, J. Matschinske, J. Spaeth, N. K. Wenke, and J. Baumbach, “Privacy-preserving artificial intelligence techniques in biomedicine,” Methods of Information in Medicine , vol. 61, pp. e12–e27, 2022. 15

    2022

  27. [35]

    Privacy analysis of deep learning in the wild: Membership inference attacks against transfer learning,

    Y . Zou, Z. Zhang, M. Backes, and Y . Zhang, “Privacy analysis of deep learning in the wild: Membership inference attacks against transfer learning,” arXiv preprint arXiv:2009.04872 , 2020

    2009 arXiv

  28. [36]

    Privacy risks of general-purpose language models,

    X. Pan, M. Zhang, S. Ji, and M. Yang, “Privacy risks of general-purpose language models,” in 2020 IEEE Symposium on Security and Privacy (SP). IEEE, 2020, pp. 1314–1331

    2020

  29. [37]

    Generative language models and automated influence operations: Emerging threats and potential mitigations,

    J. A. Goldstein, G. Sastry, M. Musser, R. DiResta, M. Gentzel, and K. Sedova, “Generative language models and automated influence operations: Emerging threats and potential mitigations,” arXiv preprint arXiv:2301.04246, 2023

    2023 arXiv

  30. [38]

    On the dangers of stochastic parrots: Can language models be too big?

    E. M. Bender, T. Gebru, A. McMillan-Major, and S. Shmitchell, “On the dangers of stochastic parrots: Can language models be too big?” in Proceedings of the 2021 ACM conference on fairness, accountability, and transparency, 2021, pp. 610–623

    2021

  31. [39]

    A machine and human reader study on AI diagnosis model safety under attacks of adversarial images,

    Q. Zhou, M. Zuley, Y . Guo, L. Yang, B. Nair, A. Vargo, S. Ghannam, D. Arefan, and S. Wu, “A machine and human reader study on AI diagnosis model safety under attacks of adversarial images,” Nature communications, vol. 12, no. 1, p. 7281, 2021

    2021

  32. [40]

    Secure and robust machine learning for healthcare: A survey,

    A. Qayyum, J. Qadir, M. Bilal, and A. Al-Fuqaha, “Secure and robust machine learning for healthcare: A survey,” IEEE Reviews in Biomedical Engineering, vol. 14, pp. 156–180, 2020

    2020

  33. [41]

    Adversarial attacks against medical deep learning systems,

    S. G. Finlayson, H. W. Chung, I. S. Kohane, and A. L. Beam, “Adversarial attacks against medical deep learning systems,” arXiv preprint arXiv:1804.05296, 2018

    2018 arXiv

  34. [42]

    Adversarial attacks on medical machine learning,

    S. G. Finlayson, J. D. Bowers, J. Ito, J. L. Zittrain, A. L. Beam, and I. S. Kohane, “Adversarial attacks on medical machine learning,” Science, vol. 363, no. 6433, pp. 1287–1289, 2019

    2019

  35. [43]

    Generalizability vs. robustness: adversarial examples for medical imaging,

    M. Paschali, S. Conjeti, F. Navarro, and N. Navab, “Generalizability vs. robustness: adversarial examples for medical imaging,” arXiv preprint arXiv:1804.00504, 2018

    2018 arXiv

  36. [44]

    Deep learning models for electrocardiograms are susceptible to adversarial attack,

    X. Han, Y . Hu, L. Foschini, L. Chinitz, L. Jankelson, and R. Ran- ganath, “Deep learning models for electrocardiograms are susceptible to adversarial attack,” Nature medicine , vol. 26, no. 3, pp. 360–363, 2020

    2020

  37. [45]

    Ethical considerations of using ChatGPT in health care,

    C. Wang, S. Liu, H. Yang, J. Guo, Y . Wu, and J. Liu, “Ethical considerations of using ChatGPT in health care,” Journal of Medical Internet Research, vol. 25, p. e48009, 2023

    2023

  38. [46]

    Re- enacting machine learning practices to enquire into the moral issues they pose,

    J.-M. John-Mathews, R. De Mourat, D. Ricci, and M. Cr ´epel, “Re- enacting machine learning practices to enquire into the moral issues they pose,” Convergence, vol. 30, no. 1, pp. 66–93, 2024

    2024

  39. [47]

    HIPAA, HIPPA, or HIPPO: What really is the heath in- surance portability and accountability act?

    A. Oakley, “HIPAA, HIPPA, or HIPPO: What really is the heath in- surance portability and accountability act?” Biotechnology Law Report, 2023

    2023

  40. [48]

    The european general data protection regulation (GDPR) in mhealth: Theoretical and practical aspects for practitioners’ use,

    L. Carmi, M. Zohar, and G. M. Riva, “The european general data protection regulation (GDPR) in mhealth: Theoretical and practical aspects for practitioners’ use,” Medicine, Science and the Law , vol. 63, no. 1, pp. 61–68, 2023

    2023

  41. [49]

    Chain-of-thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhou et al. , “Chain-of-thought prompting elicits reasoning in large language models,” Advances in Neural Information Processing Systems, vol. 35, pp. 24 824–24 837, 2022

    2022

  42. [50]

    Large language models encode clinical knowledge,

    K. Singhal, S. Azizi, T. Tu, S. S. Mahdavi, J. Wei, H. W. Chung, N. Scales, A. Tanwani, H. Cole-Lewis, S. Pfohl et al., “Large language models encode clinical knowledge,” arXiv preprint arXiv:2212.13138 , 2022

    2022 arXiv

  43. [51]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning . PMLR, 2021, pp. 8748–8763

    2021

  44. [52]

    Visionfm: a multi-modal multi-task vision foundation model for generalist ophthalmic artificial intelligence,

    J. Qiu, J. Wu, H. Wei, P. Shi, M. Zhang, Y . Sun, L. Li, H. Liu, H. Liu, S. Hou et al. , “Visionfm: a multi-modal multi-task vision foundation model for generalist ophthalmic artificial intelligence,” arXiv preprint arXiv:2310.04992, 2023

    arXiv

  45. [53]

    A foundation model for generalizable disease detection from retinal images,

    Y . Zhou, M. A. Chia, S. K. Wagner, M. S. Ayhan, D. J. Williamson, R. R. Struyven, T. Liu, M. Xu, M. G. Lozano, P. Woodward-Court et al. , “A foundation model for generalizable disease detection from retinal images,” Nature, vol. 622, no. 7981, pp. 156–163, 2023

    2023

  46. [54]

    Lvm-med: Learn- ing large-scale self-supervised vision models for medical imaging via second-order graph matching,

    D. MH Nguyen, H. Nguyen, N. Diep, T. N. Pham, T. Cao, B. Nguyen, P. Swoboda, N. Ho, S. Albarqouni, P. Xie et al. , “Lvm-med: Learn- ing large-scale self-supervised vision models for medical imaging via second-order graph matching,” Advances in Neural Information Processing Sys...

    2024

  47. [55]

    GLORIA-A globally representative hyperspectral in situ dataset for optical sensing of water quality,

    M. K. Lehmann, D. Gurlin, N. Pahlevan, K. Alikas, T. Conroy, J. Anstee, S. V . Balasubramanian, C. C. Barbosa, C. Binding, A. Bracher et al. , “GLORIA-A globally representative hyperspectral in situ dataset for optical sensing of water quality,” Scientific Data , vol. 10, no. ...

    2023

  48. [56]

    LAMM: Language-assisted multi-modal instruction-tuning dataset, framework, and benchmark,

    Z. Yin, J. Wang, J. Cao, Z. Shi, D. Liu, M. Li, L. Sheng, L. Bai, X. Huang, Z. Wang et al. , “LAMM: Language-assisted multi-modal instruction-tuning dataset, framework, and benchmark,” arXiv preprint arXiv:2306.06687, 2023

    2023 arXiv

  49. [57]

    PLIP: Language-image pre-training for person representation learning,

    J. Zuo, C. Yu, N. Sang, and C. Gao, “PLIP: Language-image pre-training for person representation learning,” arXiv preprint arXiv:2305.08386, 2023

    2023 arXiv

  50. [58]

    Praveen and V

    S. Praveen and V . Vajrobol, “Understanding the perceptions of health- care researchers regarding ChatGPT: a study based on bidirectional encoder representation from transformers (BERT) sentiment analysis and topic modeling,” Annals of Biomedical Engineering, pp. 1–3, 2023

    2023

  51. [59]

    An analysis on large language mod- els in healthcare: A case study of BioBERT,

    S. Sharaf and V . Anoop, “An analysis on large language mod- els in healthcare: A case study of BioBERT,” arXiv preprint arXiv:2310.07282, 2023

    2023 arXiv

  52. [60]

    ChatGPT: Is this version good for healthcare and research?

    R. Vaishya, A. Misra, and A. Vaish, “ChatGPT: Is this version good for healthcare and research?” Diabetes & Metabolic Syndrome: Clinical Research & Reviews, vol. 17, no. 4, p. 102744, 2023

    2023

  53. [61]

    The future landscape of large language models in medicine,

    J. Clusmann, F. R. Kolbinger, H. S. Muti, Z. I. Carrero, J.-N. Eckardt, N. G. Laleh, C. M. L. L ¨offler, S.-C. Schwarzkopf, M. Unger, G. P. Veldhuizen et al. , “The future landscape of large language models in medicine,” Communications Medicine, vol. 3, no. 1, p. 141, 2023

    2023

  54. [62]

    The shaky foundations of large language models and foundation models for electronic health records,

    M. Wornow, Y . Xu, R. Thapa, B. Patel, E. Steinberg, S. Fleming, M. A. Pfeffer, J. Fries, and N. H. Shah, “The shaky foundations of large language models and foundation models for electronic health records,” npj Digital Medicine , vol. 6, no. 1, p. 135, 2023

    2023

  55. [63]

    AnyPredict: Foundation Model for Tabular Prediction,

    Z. Wang, C. Gao, C. Xiao, and J. Sun, “AnyPredict: Foundation Model for Tabular Prediction,” arXiv preprint arXiv:2305.12081 , 2023

    2023 arXiv

  56. [64]

    Ehr foundation models improve robustness in the presence of temporal distribution shift,

    L. L. Guo, E. Steinberg, S. L. Fleming, J. Posada, J. Lemmon, S. R. Pfohl, N. Shah, J. Fries, and L. Sung, “Ehr foundation models improve robustness in the presence of temporal distribution shift,” Scientific Reports, vol. 13, no. 1, p. 3767, 2023

    2023

  57. [65]

    Large language models encode clinical knowledge,

    K. Singhal, S. Azizi, T. Tu, S. S. Mahdavi, J. Wei, H. W. Chung, N. Scales, A. Tanwani, H. Cole-Lewis, S. Pfohl et al., “Large language models encode clinical knowledge,” Nature, vol. 620, no. 7972, pp. 172–180, 2023

    2023

  58. [66]

    DeID-GPT: zero-shot medical text de- identification by GPT-4,

    Z. Liu, X. Yu, L. Zhang, Z. Wu, C. Cao, H. Dai, L. Zhao, W. Liu, D. Shen, Q. Li et al. , “DeID-GPT: zero-shot medical text de- identification by GPT-4,” arXiv, 2023

    2023

  59. [67]

    GPT-4: a new era of artificial intelligence in medicine,

    E. Waisberg, J. Ong, M. Masalkhi, S. A. Kamran, N. Zaman, P. Sarker, A. G. Lee, and A. Tavakkoli, “GPT-4: a new era of artificial intelligence in medicine,” Irish Journal of Medical Science (1971-) , pp. 1–4, 2023

    1971

  60. [68]

    ChatGPT and Large Language Models (LLMs) in Healthcare: Opportunities and Risks,

    H. Ali, J. Qadir, and Z. Shah, “ChatGPT and Large Language Models (LLMs) in Healthcare: Opportunities and Risks,” 2023

    2023

  61. [69]

    Domain-specific language model pretraining for biomedical natural language processing,

    Y . Gu, R. Tinn, H. Cheng, M. Lucas, N. Usuyama, X. Liu, T. Naumann, J. Gao, and H. Poon, “Domain-specific language model pretraining for biomedical natural language processing,” ACM Transactions on Computing for Healthcare (HEALTH) , vol. 3, no. 1, pp. 1–23, 2021

    2021

  62. [70]

    PathAsst: Redefining Pathology through Generative Foundation AI Assistant for Pathology,

    Y . Sun, C. Zhu, S. Zheng, K. Zhang, Z. Shui, X. Yu, Y . Zhao, H. Li, Y . Zhang, R. Zhao et al. , “PathAsst: Redefining Pathology through Generative Foundation AI Assistant for Pathology,” arXiv preprint arXiv:2305.15072, 2023

    2023 arXiv

  63. [71]

    Conversational AI models for ophthalmic diagnosis: Comparison of ChatGPT and the Isabel pro differential diagnosis generator,

    M. Balas and E. B. Ing, “Conversational AI models for ophthalmic diagnosis: Comparison of ChatGPT and the Isabel pro differential diagnosis generator,” JFO Open Ophthalmology , vol. 1, p. 100005, 2023

    2023

  64. [72]

    The diagnostic and triage accuracy of the GPT-3 artificial intelligence model,

    D. M. Levine, R. Tuwani, B. Kompa, A. Varma, S. G. Finlayson, A. Mehrotra, and A. Beam, “The diagnostic and triage accuracy of the GPT-3 artificial intelligence model,” medRxiv, pp. 2023–01, 2023

    2023

  65. [73]

    Faithful AI in Medicine: A Systematic Review with Large Language Models and Beyond,

    Q. Xie, E. J. Schenck, H. S. Yang, Y . Chen, Y . Peng, and F. Wang, “Faithful AI in Medicine: A Systematic Review with Large Language Models and Beyond,”Medrxiv: the Preprint Server for Health Sciences, 2023

    2023

  66. [74]

    ChatGPT and the future of medical writing,

    S. Biswas, “ChatGPT and the future of medical writing,” p. e223312, 2023

    2023

  67. [75]

    Segment Anything Model (SAM) for Digital Pathology: Assess Zero-shot Seg- mentation on Whole Slide Imaging,

    R. Deng, C. Cui, Q. Liu, T. Yao, L. W. Remedios, S. Bao, B. A. Landman, L. E. Wheless, L. A. Coburn, K. T. Wilson et al., “Segment Anything Model (SAM) for Digital Pathology: Assess Zero-shot Seg- mentation on Whole Slide Imaging,” arXiv preprint arXiv:2304.04155, 2023

    2023 arXiv

  68. [76]

    The segment anything foundation model achieves favorable brain tumor autosegmen- tation accuracy on MRI to support radiotherapy treatment planning,

    F. Putz, J. Grigo, T. Weissmann, P. Schubert, D. Hoefler, A. Gomaa, H. B. Tkhayat, A. Hagag, S. Lettmaier, B. Frey et al. , “The segment anything foundation model achieves favorable brain tumor autosegmen- tation accuracy on MRI to support radiotherapy treatment planning,” arX...

    2023 arXiv

  69. [77]

    Clinical applications of deep learning in breast MRI,

    X. Zhao, J.-W. Bai, Q. Guo, K. Ren, and G.-J. Zhang, “Clinical applications of deep learning in breast MRI,” Biochimica et Biophysica Acta (BBA)-Reviews on Cancer , p. 188864, 2023. 16

    2023

  70. [78]

    Pancreatic cancer detection on CT scans with deep learning: a nationwide population-based study,

    P.-T. Chen, T. Wu, P. Wang, D. Chang, K.-L. Liu, M.-S. Wu, H. R. Roth, P.-C. Lee, W.-C. Liao, and W. Wang, “Pancreatic cancer detection on CT scans with deep learning: a nationwide population-based study,” Radiology, vol. 306, no. 1, pp. 172–182, 2023

    2023

  71. [79]

    Capabilities of GPT-4 on medical challenge problems,

    H. Nori, N. King, S. M. McKinney, D. Carignan, and E. Horvitz, “Capabilities of GPT-4 on medical challenge problems,” arXiv preprint arXiv:2303.13375, 2023

    2023 arXiv

  72. [80]

    Artificial general intelligence for medical imaging,

    X. Li, L. Zhang, Z. Wu, Z. Liu, L. Zhao, Y . Yuan, J. Liu, G. Li, D. Zhu, P. Yanet al., “Artificial general intelligence for medical imaging,”arXiv preprint arXiv:2306.05480, 2023

    2023 arXiv

  73. [81]

    On the challenges and perspectives of foundation models for medical image analysis,

    S. Zhang and D. Metaxas, “On the challenges and perspectives of foundation models for medical image analysis,” arXiv preprint arXiv:2306.05705, 2023

    2023 arXiv

  74. [82]

    Generalist vision foundation models for medical imaging: A case study of segment anything model on zero-shot medical segmentation,

    P. Shi, J. Qiu, S. M. D. Abaxi, H. Wei, F. P.-W. Lo, and W. Yuan, “Generalist vision foundation models for medical imaging: A case study of segment anything model on zero-shot medical segmentation,” Diagnostics, vol. 13, no. 11, p. 1947, 2023

    1947

  75. [83]

    Segment anything is not always perfect: An investigation of sam on different real-world applications. arxiv 2023,

    W. Ji, J. Li, Q. Bi, W. Li, and L. Cheng, “Segment anything is not always perfect: An investigation of sam on different real-world applications. arxiv 2023,” arXiv preprint arXiv:2304.05750

    2023 arXiv

  76. [84]

    Evaluation of GPT-3.5 and GPT-4 for supporting real-world informa- tion needs in healthcare delivery,

    D. Dash, R. Thapa, J. M. Banda, A. Swaminathan, M. Cheatham, M. Kashyap, N. Kotecha, J. H. Chen, S. Gombar, L. Downing et al., “Evaluation of GPT-3.5 and GPT-4 for supporting real-world informa- tion needs in healthcare delivery,” arXiv preprint arXiv:2304.13714 , 2023

    2023 arXiv

  77. [85]

    Implications of large language models such as ChatGPT for dental medicine,

    F. Eggmann, R. Weiger, N. U. Zitzmann, and M. B. Blatz, “Implications of large language models such as ChatGPT for dental medicine,” Journal of Esthetic and Restorative Dentistry , 2023

    2023

  78. [86]

    The role of ChatGPT, generative language models, and artificial intelligence in medical education: a conversation with ChatGPT and a call for papers,

    G. Eysenbach et al. , “The role of ChatGPT, generative language models, and artificial intelligence in medical education: a conversation with ChatGPT and a call for papers,” JMIR Medical Education, vol. 9, no. 1, p. e46885, 2023

    2023

  79. [87]

    Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models,

    T. H. Kung, M. Cheatham, A. Medenilla, C. Sillos, L. De Leon, C. Elepa ˜no, M. Madriaga, R. Aggabao, G. Diaz-Candido, J. Maningo et al., “Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models,” PLoS digital health , vol. 2, no...

    2023

  80. [88]

    Van Bulck and P

    L. Van Bulck and P. Moons, “What if your patient switches from dr. google to dr. ChatGPT? a vignette-based survey of the trustwor- thiness, value, and danger of ChatGPT-generated responses to health questions,” European Journal of Cardiovascular Nursing , p. zvad038, 2023

    2023

  81. [89]

    A review of ChatGPT applications in education, marketing, software engineering, and healthcare: Benefits, drawbacks, and research directions,

    M. Fraiwan and N. Khasawneh, “A review of ChatGPT applications in education, marketing, software engineering, and healthcare: Benefits, drawbacks, and research directions,” arXiv preprint arXiv:2305.00237, 2023

    2023 arXiv

  82. [90]

    Large language models in medical education: Opportunities, challenges, and future directions,

    A. Abd-Alrazaq, R. AlSaad, D. Alhuwail, A. Ahmed, P. M. Healy, S. Latifi, S. Aziz, R. Damseh, S. A. Alrazak, J. Sheikh et al., “Large language models in medical education: Opportunities, challenges, and future directions,” JMIR Medical Education , vol. 9, no. 1, p. e48291, 2023

    2023

  83. [91]

    Im- proved performance of ChatGPT-4 on the OKAP exam: A comparative study with ChatGPT-3.5,

    S. Teebagy, L. Colwell, E. Wood, A. Yaghy, and M. Faustina, “Im- proved performance of ChatGPT-4 on the OKAP exam: A comparative study with ChatGPT-3.5,” medRxiv, pp. 2023–04, 2023

    2023

  84. [92]

    Evaluating the performance of ChatGPT in ophthalmology: An analysis of its successes and shortcomings,

    F. Antaki, S. Touma, D. Milad, J. El-Khoury, and R. Duval, “Evaluating the performance of ChatGPT in ophthalmology: An analysis of its successes and shortcomings,” Ophthalmology Science, p. 100324, 2023

    2023

  85. [93]

    ChatGPT: Vision and challenges,

    S. S. Gill and R. Kaur, “ChatGPT: Vision and challenges,” Internet of Things and Cyber-Physical Systems , vol. 3, pp. 262–271, 2023

    2023

  86. [94]

    AI-based large language models are ready to transform psychological health assessment,

    O. Kjell, K. Kjell, and H. A. Schwartz, “AI-based large language models are ready to transform psychological health assessment,” 2023

    2023

  87. [95]

    ChatGPT in healthcare: A taxonomy and systematic review,

    J. Li, A. Dada, J. Kleesiek, and J. Egger, “ChatGPT in healthcare: A taxonomy and systematic review,” medRxiv, pp. 2023–03, 2023

    2023

  88. [96]

    ChatGPT: A comprehensive review on background, ap- plications, key challenges, bias, ethics, limitations and future scope,

    P. P. Ray, “ChatGPT: A comprehensive review on background, ap- plications, key challenges, bias, ethics, limitations and future scope,” Internet of Things and Cyber-Physical Systems , 2023

    2023

  89. [97]

    Chatdoctor: A medical chat model fine-tuned on llama model using medical domain knowledge,

    L. Yunxiang, L. Zihan, Z. Kai, D. Ruilong, and Z. You, “Chatdoctor: A medical chat model fine-tuned on llama model using medical domain knowledge,” arXiv preprint arXiv:2303.14070 , 2023

    2023 arXiv

  90. [98]

    Chatdoctor: A medical chat model fine-tuned on a large language model meta-ai (llama) using medical domain knowledge,

    Y . Li, Z. Li, K. Zhang, R. Dan, S. Jiang, and Y . Zhang, “Chatdoctor: A medical chat model fine-tuned on a large language model meta-ai (llama) using medical domain knowledge,”Cureus, vol. 15, no. 6, 2023

    2023

  91. [99]

    Pmc-llama: Further finetuning llama on medical papers,

    C. Wu, X. Zhang, Y . Zhang, Y . Wang, and W. Xie, “Pmc-llama: Further finetuning llama on medical papers,” arXiv preprint arXiv:2304.14454, 2023

    2023 arXiv

  92. [100]

    AI-generated medical advice—GPT and beyond,

    C. E. Haupt and M. Marks, “AI-generated medical advice—GPT and beyond,” Jama, vol. 329, no. 16, pp. 1349–1350, 2023

    2023

  93. [101]

    ChatGPT for healthcare ser- vices: An emerging stage for an innovative perspective,

    M. Javaid, A. Haleem, and R. P. Singh, “ChatGPT for healthcare ser- vices: An emerging stage for an innovative perspective,” BenchCouncil Transactions on Benchmarks, Standards and Evaluations, vol. 3, no. 1, p. 100105, 2023

    2023

  94. [102]

    As- sessing the accuracy and reliability of AI-generated medical responses: an evaluation of the chat-gpt model,

    D. Johnson, R. Goodman, J. Patrinely, C. Stone, E. Zimmerman, R. Donald, S. Chang, S. Berkowitz, A. Finn, E. Jahangir et al. , “As- sessing the accuracy and reliability of AI-generated medical responses: an evaluation of the chat-gpt model,” 2023

    2023

  95. [103]

    ChatGPT: can artificial intelligence language models be of value for cardiovascular nurses and allied health professionals,

    P. Moons and L. Van Bulck, “ChatGPT: can artificial intelligence language models be of value for cardiovascular nurses and allied health professionals,” European journal of cardiovascular nursing , 2023

    2023

  96. [104]

    DocGPT: Impact of ChatGPT-3 on health services as a virtual doctor,

    L. Iftikhar et al., “DocGPT: Impact of ChatGPT-3 on health services as a virtual doctor,” EC Paediatrics, vol. 12, no. 1, pp. 45–55, 2023

    2023

  97. [105]

    Can ChatGPT be considered an author of a medical article?

    K. Ide, P. Hawke, and T. Nakayama, “Can ChatGPT be considered an author of a medical article?” Journal of Epidemiology, p. JE20230030, 2023

    2023

  98. [106]

    Exploring medical breakthroughs: A systematic review of ChatGPT applications in healthcare,

    F. Mufti ´c, M. Kaduni ´c, A. Mu ˇsinbegovi´c, and A. Abd Almisreb, “Exploring medical breakthroughs: A systematic review of ChatGPT applications in healthcare,” Southeast Europe Journal of Soft Comput- ing, vol. 12, no. 1, pp. 13–41, 2023

    2023

  99. [107]

    ChatGPT and the rise of large language models: the new AI-driven infodemic threat in public health,

    L. De Angelis, F. Baglivo, G. Arzilli, G. P. Privitera, P. Ferragina, A. E. Tozzi, and C. Rizzo, “ChatGPT and the rise of large language models: the new AI-driven infodemic threat in public health,” Frontiers in Public Health , vol. 11, p. 1166120, 2023

    2023

  100. [108]

    Artificial intelligence: Augmenting telehealth with large language models,

    C. L. Snoswell, A. J. Snoswell, J. T. Kelly, L. J. Caffery, and A. C. Smith, “Artificial intelligence: Augmenting telehealth with large language models,” Journal of telemedicine and telecare , p. 1357633X231169055, 2023

    2023

  101. [109]

    Deep learning for surgical instrument recognition and segmentation in robotic-assisted surgeries: A systematic review,

    F. A. Ahmed, M. Yousef, M. A. Ahmed, H. O. Ali, A. Mahboob, H. Ali, Z. Shah, O. Aboumarzouk, A. A. Ansari, and S. Balakrishnan, “Deep learning for surgical instrument recognition and segmentation in robotic-assisted surgeries: A systematic review,” arXiv preprint arXiv:2410.07...

    arXiv

  102. [110]

    Can generalist foundation models outcompete special-purpose tuning? case study in medicine,

    H. Nori, Y . T. Lee, S. Zhang, D. Carignan, R. Edgar, N. Fusi, N. King, J. Larson, Y . Li, W. Liu et al. , “Can generalist foundation models outcompete special-purpose tuning? case study in medicine,” arXiv preprint arXiv:2311.16452, 2023

    2023 arXiv

  103. [111]

    Large lan- guage models and artificial intelligence: A primer for plastic surgeons on the demonstrated & potential applications, promises, and limitations of ChatGPT,

    J. Abi-Rafeh, H. H. Xu, R. Kazan, R. Tevlin, and H. Furnas, “Large lan- guage models and artificial intelligence: A primer for plastic surgeons on the demonstrated & potential applications, promises, and limitations of ChatGPT,” Aesthetic Surgery Journal , p. sjad260, 2023

    2023

  104. [112]

    Biobert: a pre-trained biomedical language representation model for biomedical text mining,

    J. Lee, W. Yoon, S. Kim, D. Kim, S. Kim, C. H. So, and J. Kang, “Biobert: a pre-trained biomedical language representation model for biomedical text mining,” Bioinformatics, vol. 36, no. 4, pp. 1234–1240, 2020

    2020

  105. [113]

    Biomegatron: Larger biomedical domain language model,

    H.-C. Shin, Y . Zhang, E. Bakhturina, R. Puri, M. Patwary, M. Shoeybi, and R. Mani, “Biomegatron: Larger biomedical domain language model,” arXiv preprint arXiv:2010.06060 , 2020

    2010 arXiv

  106. [114]

    Med-bert: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction,

    L. Rasmy, Y . Xiang, Z. Xie, C. Tao, and D. Zhi, “Med-bert: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction,” NPJ digital medicine , vol. 4, no. 1, p. 86, 2021

    2021

  107. [115]

    Bioelectra: pretrained biomedical text encoder using discriminators,

    K. raj Kanakarajan, B. Kundumani, and M. Sankarasubbu, “Bioelectra: pretrained biomedical text encoder using discriminators,” in Proceed- ings of the 20th Workshop on Biomedical Language Processing , 2021, pp. 143–154

    2021

  108. [116]

    Improving biomed- ical pretrained language models with knowledge,

    Z. Yuan, Y . Liu, C. Tan, S. Huang, and F. Huang, “Improving biomed- ical pretrained language models with knowledge,” arXiv preprint arXiv:2104.10344, 2021

    2021 arXiv

  109. [117]

    Clinicalt5: A generative language model for clinical text,

    Q. Lu, D. Dou, and T. Nguyen, “Clinicalt5: A generative language model for clinical text,” in Findings of the Association for Computa- tional Linguistics: EMNLP 2022 , 2022, pp. 5436–5443

    2022

  110. [118]

    Linkbert: Pretraining lan- guage models with document links,

    M. Yasunaga, J. Leskovec, and P. Liang, “Linkbert: Pretraining lan- guage models with document links,” arXiv preprint arXiv:2203.15827, 2022

    2022 arXiv

  111. [119]

    BioGPT: generative pre-trained transformer for biomedical text generation and mining,

    R. Luo, L. Sun, Y . Xia, T. Qin, S. Zhang, H. Poon, and T.-Y . Liu, “BioGPT: generative pre-trained transformer for biomedical text generation and mining,” Briefings in Bioinformatics , vol. 23, no. 6, p. bbac409, 2022

    2022

  112. [120]

    RadBERT: Adapting transformer-based language models to radiology,

    A. Yan, J. McAuley, X. Lu, J. Du, E. Y . Chang, A. Gentili, and C.- N. Hsu, “RadBERT: Adapting transformer-based language models to radiology,” Radiology: Artificial Intelligence, vol. 4, no. 4, p. e210258, 2022

    2022

  113. [121]

    Fastfold: Reducing alphafold training time from 11 days to 67 hours,

    S. Cheng, X. Zhao, G. Lu, J. Fang, Z. Yu, T. Zheng, R. Wu, X. Zhang, J. Peng, and Y . You, “Fastfold: Reducing alphafold training time from 11 days to 67 hours,” arXiv preprint arXiv:2203.00854 , 2022. 17

    2022 arXiv

  114. [122]

    Enhancing phenotype recognition in clinical notes using large lan- guage models: Phenobcbert and phenogpt,

    J. Yang, C. Liu, W. Deng, D. Wu, C. Weng, Y . Zhou, and K. Wang, “Enhancing phenotype recognition in clinical notes using large lan- guage models: Phenobcbert and phenogpt,” Patterns, 2023

    2023

  115. [123]

    Chatcad: Interactive computer-aided diagnosis on medical image using large language models,

    S. Wang, Z. Zhao, X. Ouyang, Q. Wang, and D. Shen, “Chatcad: Interactive computer-aided diagnosis on medical image using large language models,” arXiv preprint arXiv:2302.07257 , 2023

    2023 arXiv

  116. [124]

    Cpllm: Clinical prediction with large language models,

    O. B. Shoham and N. Rappoport, “Cpllm: Clinical prediction with large language models,” arXiv preprint arXiv:2309.11295 , 2023

    2023 arXiv

  117. [125]

    Doctorglm: Fine-tuning your chinese doctor is not a herculean task,

    H. Xiong, S. Wang, Y . Zhu, Z. Zhao, Y . Liu, Q. Wang, and D. Shen, “Doctorglm: Fine-tuning your chinese doctor is not a herculean task,” arXiv preprint arXiv:2304.01097 , 2023

    2023 arXiv

  118. [126]

    Towards expert- level medical question answering with large language models,

    K. Singhal, T. Tu, J. Gottweis, R. Sayres, E. Wulczyn, L. Hou, K. Clark, S. Pfohl, H. Cole-Lewis, D. Neal et al. , “Towards expert- level medical question answering with large language models,” arXiv preprint arXiv:2305.09617, 2023

    2023 arXiv

  119. [127]

    Gatortron: A large language model for clinical natural language processing,

    X. Yang, N. PourNejatian, H. C. Shin, K. E. Smith, C. Parisien, C. Compas, C. Martin, M. G. Flores, Y . Zhang, T. Magoc et al. , “Gatortron: A large language model for clinical natural language processing,” medRxiv, pp. 2022–02, 2022

    2022

  120. [128]

    HuatuoGPT, towards Taming Language Model to Be a Doctor,

    H. Zhang, J. Chen, F. Jiang, F. Yu, Z. Chen, J. Li, G. Chen, X. Wu, Z. Zhang, Q. Xiao et al. , “HuatuoGPT, towards Taming Language Model to Be a Doctor,” arXiv preprint arXiv:2305.15075 , 2023

    2023 arXiv

  121. [129]

    Dragonfly: Multi-resolution zoom supercharges large visual- language model,

    K. Chen, R. Thapa, R. Chalamala, B. Athiwaratkun, S. L. Song, and J. Zou, “Dragonfly: Multi-resolution zoom supercharges large visual- language model,” arXiv preprint arXiv:2406.00977 , 2024

    2024 arXiv

  122. [130]

    Merlin: A vision language foundation model for 3d computed tomography,

    L. Blankemeier, J. P. Cohen, A. Kumar, D. Van Veen, S. J. S. Gardezi, M. Paschali, Z. Chen, J.-B. Delbrouck, E. Reis, C. Truyts et al., “Merlin: A vision language foundation model for 3d computed tomography,” arXiv preprint arXiv:2406.06512 , 2024

    2024

  123. [131]

    Chameleon: Mixed-modal early-fusion foundation models,

    C. Team, “Chameleon: Mixed-modal early-fusion foundation models,” arXiv preprint arXiv:2405.09818 , 2024

    2024 arXiv

  124. [132]

    Meditron-70b: Scaling medical pretraining for large language models,

    Z. Chen, A. H. Cano, A. Romanou, A. Bonnet, K. Matoba, F. Salvi, M. Pagliardini, S. Fan, A. K ¨opf, A. Mohtashami et al., “Meditron-70b: Scaling medical pretraining for large language models,” arXiv preprint arXiv:2311.16079, 2023

    2023 arXiv

  125. [133]

    Maira at rrg24: A specialised large multimodal model for radiology report generation,

    S. Srivastav, M. Ranjit, F. P ´erez-Garc´ıa, K. Bouzid, S. Bannur, D. C. Castro, A. Schwaighofer, H. Sharma, M. Ilse, V . Salvatelli et al. , “Maira at rrg24: A specialised large multimodal model for radiology report generation,” in Proceedings of the 23rd Workshop on Biomedic...

    2024

  126. [134]

    MedAlpaca–An Open-Source Collection of Medical Conversational AI Models and Training Data,

    T. Han, L. C. Adams, J.-M. Papaioannou, P. Grundmann, T. Oberhauser, A. L¨oser, D. Truhn, and K. K. Bressem, “MedAlpaca–An Open-Source Collection of Medical Conversational AI Models and Training Data,” arXiv preprint arXiv:2304.08247 , 2023

    2023 arXiv

  127. [135]

    Universal jailbreak backdoors from poisoned human feedback,

    J. Rando and F. Tram `er, “Universal jailbreak backdoors from poisoned human feedback,” arXiv preprint arXiv:2311.14455 , 2023

    2023 arXiv

  128. [136]

    Extracting training data from diffusion models,

    N. Carlini, J. Hayes, M. Nasr, M. Jagielski, V . Sehwag, F. Tram `er, B. Balle, D. Ippolito, and E. Wallace, “Extracting training data from diffusion models,” in 32nd USENIX Security Symposium (USENIX Security 23) . Anaheim, CA: USENIX Association, Aug. 2023, pp. 5253–5270. [O...

    2023

  129. [137]

    Safe RLHF: Safe reinforcement learning from human feedback,

    J. Dai, X. Pan, R. Sun, J. Ji, X. Xu, M. Liu, Y . Wang, and Y . Yang, “Safe RLHF: Safe reinforcement learning from human feedback,”arXiv preprint arXiv:2310.12773, 2023

    2023 arXiv

  130. [138]

    Towards safe self-distillation of internet-scale text-to-image diffusion models,

    S. Kim, S. Jung, B. Kim, M. Choi, J. Shin, and J. Lee, “Towards safe self-distillation of internet-scale text-to-image diffusion models,” arXiv preprint arXiv:2307.05977, 2023

    2023 arXiv

  131. [139]

    Generative AI and large language models in health care: pathways to implementation,

    M. M. Raza, K. P. Venkatesh, and J. C. Kvedar, “Generative AI and large language models in health care: pathways to implementation,” npj Digital Medicine , vol. 7, no. 1, p. 62, 2024

    2024

  132. [140]

    Guideline on summary of product characteristics,

    E. M. Agency, “Guideline on summary of product characteristics,” Official Journal of the European Union , 2009. [Online]. Available: https://ec.europa.eu/health/sites/default/files/files/eudralex/ vol-2/c/smpc guideline rev2 en.pdf

    2009

  133. [141]

    Mimic-iii, a freely accessible critical care database,

    A. E. Johnson, T. J. Pollard, L. Shen, L.-w. H. Lehman, M. Feng, M. Ghassemi, and R. G. Mark, “Mimic-iii, a freely accessible critical care database,” Scientific Data , vol. 3, p. 160035, 2016. [Online]. Available: https://doi.org/10.1038/sdata.2016.35

    2016 doi

  134. [142]

    Rag in health care: A novel framework for improving communication and decision-making by addressing llm limitations,

    K. K. Y . Ng, I. Matsuba, and P. C. Zhang, “Rag in health care: A novel framework for improving communication and decision-making by addressing llm limitations,” NEJM AI, p. AIra2400380, 2024

    2024

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.