Revolutionizing Health Care: The Transformative Impact of Large Language Models in Medicine

Abstract

Large language models (LLMs) are rapidly advancing medical artificial intelligence, offering revolutionary changes in health care. These models excel in natural language processing (NLP), enhancing clinical support, diagnosis, treatment, and medical research. Breakthroughs, like GPT-4 and BERT (Bidirectional Encoder Representations from Transformer), demonstrate LLMs’ evolution through improved computing power and data. However, their high hardware requirements are being addressed through technological advancements. LLMs are unique in processing multimodal data, thereby improving emergency, elder care, and digital medical procedures. Challenges include ensuring their empirical reliability, addressing ethical and societal implications, especially data privacy, and mitigating biases while maintaining privacy and accountability. The paper emphasizes the need for human-centric, bias-free LLMs for personalized medicine and advocates for equitable development and access. LLMs hold promise for transformative impacts in health care.

Introduction

Recent advancements in artificial intelligence (AI) have catalyzed the development and significant breakthroughs of large language models (LLMs), placing them at the forefront of AI research [1-4]. LLMs are deep learning models that generate human-like text by predicting the next word in a sequence based on statistical patterns learned from vast text data. These models leverage deep learning algorithms to interpret and generate natural language, using extensive corpus data to enhance pretrained language models, a cornerstone of natural language processing (NLP) [5,6]. Characterized by their immense scale, these models often consist of hundreds of millions to billions of parameters and are trained on vast textual datasets [7,8]. Their ability to efficiently process natural language data with minimal human intervention, capturing intricate grammatical structures, lexical nuances, and semantic contexts, is noteworthy. Globally recognized LLMs include the ChatGPT series, BERT (Bidirectional Encoder Representations from Transformer), PaLM, LaMDA, and Meta’s Llama series, with China contributing models such as Baidu’s “Wenxin Yiyan,” 360’s LLM, Alibaba’s “Tongyi Qianwen,” and SenseTime’s LLM [9]. The evolution of LLMs represents over 7 years of relentless technological innovation and research, marking a significant milestone in AI development since the inception of the Turing machine.

LLMs primarily function to comprehend, generate, and interact through language. In NLP tasks, such as text classification, named entity recognition, and sentiment analysis, their proficiency is unparalleled [10-12]. Beyond these applications, LLMs are expanding their influence. In mathematics, they assist in solving complex problems and contributing to mathematical proofs [13]. In software development, their capabilities include automatic code generation, debugging assistance, and complex algorithm explanation [14]. Intriguingly, LLMs are venturing into artistic creation, exhibiting talent in generating poetry, stories, and music [15,16].

In the medical domain, LLMs are poised to revolutionize clinical decision support. They can assist health care professionals in diagnosing diseases with enhanced accuracy and speed, provide treatment recommendations, and facilitate the analysis of medical records by processing large volumes of medical data [17-20]. They are instrumental in swiftly navigating vast medical literature, providing health care professionals with essential research, guidelines, and information, thus saving time and grounding medical treatments in current knowledge [21-25]. Additionally, LLMs can interact directly with patients, offer medical consultations, and handle document processing efficiently [26-28]. For example, health care professionals use LLMs to assist in diagnosing diseases by quickly processing and interpreting large volumes of patient data such as electronic health records and imaging results. Clinicians also leverage LLMs for treatment planning, where the models suggest potential treatment options based on the latest medical guidelines and patient-specific data. Moreover, LLMs are used in streamlining administrative tasks, such as generating and managing medical documentation, allowing clinicians to spend more time with their patients. Their role in drug research and development is also emerging, aiding in new drug discoveries through detailed analysis of chemical and biological data [29,30]. As such, LLMs are reshaping research methodologies and applications across various fields, particularly in medicine, equipping doctors with advanced tools for more accurate and efficient diagnosis and treatment, while offering patients more convenient and effective medical services. The potential for broader applications of LLMs in the medical field is vast, and there is a strong rationale to expect their significant impact on future health care advancements (Figure 1).

Figure 1. Timeline of mainstream LLMs commercially available to the public. The technological evolution of LLMs, highlighting several key technologies and models. It includes RNNs and LSTMs from the 1990s, Google’s Transformer model introduced in 2017, Google’s BERT model released in 2018, and the GPT series by OpenAI. Specific emphasis is placed on three major milestones: the first open-source LLM—GPT-2, and the first widely acclaimed LLM—GPT-3. These developments signify major advancements in LLMs within the field of natural language processing. BERT: Bidirectional Encoder Representations from Transformers; LLM: large language model; LSTM: long short-term memory network; RNN: recurrent neural network.

LLM Technical Background and Hardware Infrastructure

The evolution of LLMs, like OpenAI’s GPT-3 and Google’s BERT, has been monumental, driven by advancements in AI chip computing power and large, high-quality datasets [31]. The Transformer model, introduced by Google in 2016, underpins this progress, predicting words in sentences based on statistical correlations [32,33]. Notably, GPT-3 in 2020 showcased the significance of model size and data quality.

The operation and training of LLMs, such as ChatGPT, require substantial hardware infrastructure [34]. This includes graphics processing units (GPUs) or tensor processing units (TPUs) with thousands of cores, extensive RAM (several terabytes), over 48 GB of VRAM on GPUs, high-performance solid-state drives, and fast, low-latency networks (10 to 100 Gbps) [35,36]. Effective cooling systems and reliable power supplies are also essential. Compatibility with software frameworks, like TensorFlow and PyTorch, is necessary for optimizing training and deployment. The training of GPT-3, for instance, costs around US $1.4 million, and operational costs for models, like ChatGPT, can reach up to US $700,000 daily, with significant energy consumption.

Future technology advancements are expected to reduce the costs and improve the efficiency of LLMs. Progress in GPU and TPU technologies, along with hardware tailored for LLM training, will drive efficiency. Compact model structures through knowledge distillation, model pruning, transfer learning, energy-efficient practices, distributed training, and edge computing are anticipated. Semisupervised and self-supervised learning methods will also play a role in training models with fewer labeled datasets [37,38]. ChatGPT’s recent updates showcase improvements in response speed, handling complex queries, multimodal functionality, global language support, and enhanced privacy and security measures [39].

In health care, deploying large-scale medical models faces unique challenges due to data security and privacy concerns. Hospitals typically have CPUs for general computing, with limited access to GPUs. Medical LLMs, generally smaller than general-purpose LLMs, still require substantial investment in operational hardware [40,41]. For instance, a model with 13 billion parameters might cost under US $138,000 while larger models for entire hospitals may require advanced GPU solutions costing around RMB 10 million. Effective deployment demands careful consideration of model scale, computational resources, data security, and cost control (Figure 2).

Figure 2. The architectural designs of LLMs: a study of self-attention mechanisms and structural variations. The image depicts the hardware infrastructure for LLMs and their implementation in the BERT and GPT models. On the left, there is a network diagram showing servers and computing devices needed to run these models, labeled with hardware such as TPU and GPU. On the right, the structure of BERT and GPT is compared in detail, including positional encoding, self-attention mechanisms, feed-forward networks, addition and normalization layers, and the computation of output probabilities. Although these models have different approaches to processing text, both are large neural network models based on deep learning and self-attention mechanisms. BERT: Bidirectional Encoder Representations from Transformers. GPU: graphics processing unit; LLM: large language model. TPU: tensor processing unit.

Advancing the Integration of LLMs in Health Care: The Imperative for Evidence-Based Research and Collaborative Evaluation

Overview

In the contemporary health care landscape, the paradigm of evidence-based medicine is instrumental in shaping medical decision-making processes. This methodology integrates top-tier research evidence with clinical expertise and aligns it with patient values and expectations, thereby informing patient care decisions. Evidence-based medicine ensures that medical interventions are grounded in scientific evidence rather than solely relying on a physician’s experience or intuition, enhancing patient safety and the efficacy of treatments [42-45].

The integration of LLMs into the medical field introduces a significant challenge: the current scarcity of evidence-based medical research concerning the application of LLMs in health care settings [46]. Although LLMs have shown remarkable efficacy in various sectors, the unique context of medicine, with its direct implications for human life and health, necessitates a cautious approach to the introduction of untested technologies or methods into clinical practice [47]. Despite their robust data processing capabilities, LLMs present a potential risk for prediction errors in clinical environments. The medical domain, with its complex interplay of biology, physiology, and pathology, might be challenging for machine learning models to fully encapsulate, especially considering the intricacies and variability inherent in medical data [48]. Furthermore, the realm of medical decision-making often requires a high level of expertise and experience, aspects that may not be entirely replicable by LLMs. The consequences of medical decisions far surpass those in other sectors, where a misdiagnosis or incorrect treatment recommendation could directly jeopardize a patient’s life. Hence, it is imperative to back any new technological innovation, including LLMs, with solid scientific evidence before they are implemented in medical practice.

Currently, empirical studies examining the application of LLMs in the medical field are limited. This scarcity of research implies an inability to definitively assess the accuracy, reliability, and safety of LLMs within a health care context. Model reliability refers to the consistency and dependability of a model’s outputs across different datasets or under varying conditions. In medical applications, the reliability of LLMs is critical, as it directly affects the accuracy of diagnoses and treatment recommendations, where any inconsistency could have serious consequences for patient care. To comprehensively understand the potential benefits and risks associated with LLMs in medicine, a more robust body of clinical research is required. This research should encompass randomized controlled trials, observational studies, and extensive collaborative research, which are critical to evaluating the clinical utility of LLMs accurately [49].

To accelerate the empirical evaluation of LLMs in the medical field, fostering collaboration between medical institutions, research organizations, and technology companies is essential. This interdisciplinary collaboration ensures the comprehensiveness and quality of the research, facilitating the rapid advancement and application of LLM technologies. To enhance the transparency, trustworthiness, and ethical application of LLMs in health care, it is crucial to address the societal implications, particularly in terms of data privacy. Publicizing research findings and fostering interdisciplinary collaboration among doctors, researchers, and ethicists will be key to ensuring that LLMs are used responsibly and equitably. Furthermore, the integration of robust data privacy measures and adherence to ethical standards must be a priority to prevent potential misuse or unintended consequences that could undermine public trust. Such an approach ensures that LLMs’ application in the medical field is underpinned by scientific rigor, is safe, and genuinely benefits both patients and the health care system.

Integrated Application of LLMs in Medical System

As we witness ongoing advancements in medical technology, the integration of LLMs with other tools and platforms within health care systems becomes increasingly crucial [50]. This fusion provides health care professionals with powerful tools to process, analyze, and effectively use vast amounts of health care data [23,51-54]. The integration of LLMs, such as ChatGPT, into medical systems has the potential to drive transformative progress in health care delivery. First, LLMs can potentially enhance diagnostic accuracy and clinical decision-making by analyzing comprehensive medical data to identify relevant information and suggest potential diagnoses based on presented symptoms [55-57]. Second, their proficiency in text processing and generation assists medical professionals in efficiently summarizing medical literature, facilitating research, and improving communication between health care providers and patients [58-61]. The rapid adoption of readily available LLMs, such as ChatGPT, within the medical community, signifies recognition of their potential to transform health care delivery [62-66].

However, the application of LLMs in clinical settings is not without challenges [67]. A primary concern is the generalizability of these models. Although LLMs have shown outstanding performance in numerous standard tasks, the complexity and diversity of the medical field suggest that these models may be susceptible to prediction errors in real clinical scenarios. Such errors can have serious implications, particularly when they influence critical health and life decisions. Additionally, the medical field encompasses a vast array of domain-specific knowledge that might exceed the training scope of LLMs, potentially leading to misunderstandings in complex medical scenarios.

Despite these challenges, the potential benefits and impact of LLMs in health care are considerable. LLMs can notably enhance the efficiency of medical workflows by automating routine processes such as appointment scheduling, diagnosis, and report generation [68]. Their data-driven recommendations provide powerful decision support to doctors, assisting them in making more accurate and timely decisions. Current digital health workflows often burden physicians with extensive data entry, querying, and management tasks, leading to information overload and fatigue. LLMs can alleviate these burdens by automating these tasks, thereby saving valuable time for health care providers. Moreover, by analyzing and integrating patients’ medical data, LLMs can offer tailored diagnoses and treatment recommendations, improving the overall quality of health care delivery. LLMs also play a crucial role in enhancing doctor-patient interactions. Leveraging NLP technology, they can better comprehend patients’ needs and concerns, offering more personalized medical advice [69]. This not only boosts patient satisfaction but also enhances the overall effectiveness of medical services. The potential of LLMs to optimize digital health care workflows is undeniable. With further technological advancements and empirical research, LLMs are expected to play an increasingly significant role in the future of health care (Figure 3).

Figure 3. Integration of LLMs in health care systems across different scales. LLMs can assist in monitoring and analyzing patient health records, treatment plans, and laboratory results at the individual bed level while managing care schedules and facilitating doctor-patient communication. At the hospital level, LLM helps manage patient data, operational logistics, staff scheduling, and resource allocation, while analyzing epidemic trends and hospital infection rates. At the community level, LLM can be used to predict public health crises, manage vaccination campaigns, coordinate community health initiatives, and analyze population health data to improve health policy. LLM: large language model.

Multimodal LLMs in Real-World Medical Scenarios

The advent of multimodal LLMs is bringing about a paradigm shift in the medical field by offering the capability to process and generate diverse data types such as text, images, sounds, and videos. This integration of multiple data types enables LLMs to provide more comprehensive and accurate predictions, thereby unlocking unprecedented potential [70-73]. To understand their role, it is essential to define what multimodal LLMs entail. Multimodal LLMs excel in processing, interpreting, and generating a wide array of data types, which significantly enhances their predictive capabilities. For instance, in the medical field, combining textual data from patient records with imaging data from magnetic resonance imaging (MRI), computed tomography scans, and x-rays allows these models to provide more nuanced and precise diagnoses. Additionally, integrating audio data from patient interviews or video data from medical procedures can further enrich the model’s understanding, leading to more accurate and personalized treatment recommendations. By leveraging the strengths of various data types, multimodal LLMs can offer a holistic view of a patient’s condition, which is often crucial for complex diagnoses and treatment planning.

The utility of LLMs is increasingly becoming a focal point in medical imaging [74-76]. For instance, when a patient undergoes an MRI or computed tomography scan, an LLM can swiftly analyze and integrate the image data with the patient’s textual medical records, thereby providing more comprehensive and detailed diagnostic insights. Additionally, LLMs have the capability to automatically identify and highlight crucial areas in medical images, thus providing clinicians with clear references that aid in identifying potential issues [77]. Moreover, LLMs can generate automated image reports, offering initial interpretations and treatment suggestions based on the analyzed image data, significantly boosting the efficiency and accuracy of medical diagnoses and treatments.

Multimodal LLMs are revolutionizing the field of telemedicine, transforming the dynamics of doctor-patient interactions [55,78]. For instance, LLMs have been successfully integrated into MRI analysis, where they can rapidly interpret imaging data and provide diagnostic recommendations. This has significantly reduced the time required for diagnosis and improved accuracy. However, the use of LLMs is not without its challenges. A notable example is Google BARD, which recently demonstrated racial bias in patient diagnosis, disproportionately affecting minority groups. This case highlights the dual-edged nature of LLMs in health care—they offer substantial benefits in efficiency and accuracy, yet they also pose significant risks if not properly validated and monitored for biases. Furthermore, the integration of LLMs with smart sensors and devices enables the continuous monitoring of patient’s physiological data, such as heart rate and blood pressure, facilitating early detection and intervention for any health anomalies, thus significantly bolstering patient health management.

In summary, multimodal LLMs offer a novel and efficacious approach to diagnosis, treatment, and health care management. Their robust capabilities in data processing and integration allow medical professionals to deliver more precise and efficient services to patients. At the same time, these models enable patients to access medical advice and care with greater convenience. As these technologies continue to evolve and improve, their significance and impact in the medical field are expected to grow exponentially (Figure 4).

Figure 4. The importance of multimodal large language models in medical applications. The central heart represents the cardiac health status of the human body. The surrounding circular icons depict various cardiac conditions including coronary artery disease, hypertension, arrhythmia, heart failure, valvular heart disease, cardiomyopathy, and congenital heart defects. These conditions are detected and analyzed through different medical imaging and diagnostic technologies such as electrocardiography, heart sounds, echocardiogram, coronary angiography, cardiac MRI, and nuclear cardiology. The results from these diagnostics are processed by an AI system to determine the type and severity of cardiac disease, assisting physicians in formulating treatment plans. MRI: magnetic resonance imaging.

The Key Role of LLMs in Medical Research

In the field of fundamental medical research, the capabilities of LLMs in AI are being increasingly recognized [79-82]. LLMs can swiftly retrieve and organize crucial information from vast biomedical literature, providing researchers with an efficient tool to access and synthesize the latest research findings on specific drugs, diseases, or genes [83]. In drug discovery, LLMs can predict the activity, toxicity, and pharmacokinetic properties of new compounds, facilitating early-stage drug screening [84]. These predictions not only save time but also facilitate the early-stage screening of potential drug molecules. LLMs can use existing literature and databases to predict the potential functions of newly discovered genes, a crucial aspect of genomic research, given the daily discovery and study of new genes. While protein structure prediction depends primarily on specialized models, such as AlphaFold, LLMs can enhance these models by supplying pertinent information from literature, thereby increasing prediction accuracy. In epidemiological research, LLMs can aid researchers in tracking and predicting disease spread by analyzing social media and other web-based text data, offering data support for public health decision-making. Finally, in bioinformatics applications, LLMs can assist researchers in predicting patterns, functional domains, and similarities to known biological sequences. Despite their extensive applications in biomedicine, LLMs cannot entirely replace laboratory experiments or in-depth biomedical expertise. Instead, they should be considered powerful supplementary tools, rather than replacements.

LLMs play a pivotal role in clinical research. They aid doctors and researchers by extracting essential information from medical records, and by organizing and categorizing data for easier analysis and application. For instance, they can expedite the selection of suitable patients for enrolment, thereby enhancing the design and implementation of clinical trials. In the role of a clinical research coordinator, these models assist with data entry, verification, and analysis, thereby accelerating the clinical research process. Through automated data processing and real-time analysis, LLMs can ensure data accuracy and completeness, while also reducing the workload of clinical research coordinators. This, in turn, speeds up the clinical research process and enhances research quality.

Although LLMs have revolutionized biomedicine by simplifying literature searches, aiding drug discovery, annotating gene functions, and supporting epidemiological studies, they experience certain drawbacks. Their ability to swiftly parse large datasets and make predictions may be counterbalanced by potential limitations in real-world validation [85]. For example, while they can predict a drug molecule’s properties, the actual biological response may vary. Similarly, despite gene function predictions being well grounded, they may not fully encapsulate the breadth of gene interactions. Moreover, using LLMs to analyze epidemiological trends without correlating them to underlying data could misdirect public health interventions. Therefore, while LLMs are undeniably beneficial to biomedicine, it is essential to adopt a balanced approach, combining their computational prowess with rigorous experimental validation and expert review, to fully harness their potential without sacrificing scientific rigor (Figure 5).

Figure 5. The crucial role of LLMs in medical science: bridging basic research and clinical trials. This illustration highlights the versatile roles of LLMs in medical research. LLMs analyze medical texts to uncover trends and inform research directions, facilitate hypothesis generation, and enhance clinical trial designs. They personalize medicine through data-driven treatment plans and use predictive modeling to inform clinical trial outcomes. LLMs also streamline research by integrating data and maintaining regulatory compliance. They assist in medical communication and education and evaluate the societal impact of clinical research. LLM: large language model.

Great Challenges of LLMs in Medical Scenarios and Feasible Roadmap

The integration of technology in health care invariably brings a mix of anticipation and challenges, particularly given its direct impact on human life and health. As a leading exemplar of current AI technology, LLMs present a complex array of opportunities and challenges in the medical field, warranting thorough exploration and discussion [86-88].

Handling medical data, some of the most private and sensitive information about individuals, is a significant challenge for LLMs. As LLMs are increasingly integrated into health care, ethical considerations surrounding data privacy and societal impact must be prioritized during their development and deployment. The key lies not only in using this data to enhance medical efficiency but also in implementing robust data protection frameworks to prevent misuse, leakage, and unauthorized access. Furthermore, addressing these ethical challenges requires ongoing dialogue among technologists, health care professionals, policy makers, and the public to ensure that LLM deployment aligns with societal values and legal standards [80,89]. A potential technical solution involves anonymizing patient data, ensuring that neither processing nor transmission stages can be linked to specific individuals. Concurrently, medical organizations and technology providers must establish robust data management and access protocols, ensuring clear authorization and purpose for each data access.

Interpretive challenges loom large with LLMs in medicine. Medical decision-making is distinct from other fields due to its complexity and direct implications for patients’ lives and health. When LLMs provide diagnostic or treatment suggestions, it is vital that the rationale behind these recommendations is transparent and comprehensible [90-92]. This brings us to the concept of interpretability in machine learning, which refers to the ability to understand and explain how a model makes its decisions. In the context of health care, interpretability is a significant challenge because clinicians must trust and validate the outputs of LLMs, especially when these models influence critical medical decisions [93]. Developing mental models can aid LLMs in presenting their decision-making logic in a manner that is more accessible to human users. Leveraging deep learning and other machine learning technologies, LLMs can extract disease pathophysiological mechanisms from a vast corpus of medical literature and data, providing a scientific basis for their outputs. To further enhance interpretability, LLMs could use visual tools, like graphics and animations, to clarify the logic and evidence underpinning their decisions for both physicians and patients [94,95].

The issue of technical bias and the possibility of generating misleading information or “hallucinations” are inherent challenges in LLMs. In this context, hallucinations refer to instances where LLMs produce outputs that are factually incorrect or misleading, often because the model attempts to generate an answer despite lacking sufficient context or knowledge. These hallucinations can be especially problematic in medical scenarios, where inaccurate information can have severe consequences. The data sources for these models, often anonymized consultation data and digital materials, are not uniform and vary in quality, sometimes containing erroneous samples. Fine-tuning LLMs based on such data may lead to biased or skewed medical recommendations [96,97]. Addressing this requires rigorous data auditing and the establishment of continuous bias-correction mechanisms. To mitigate the risk of hallucinations, knowledge enhancement methods, such as integrating a knowledge retrieval library or search enhancement tools, can be beneficial. The LLM’s responses can be cross-referenced with retrieved data to filter out inconsistencies with reality. Another approach involves reinforcement learning based on human feedback, where high-quality feedback is provided to fine-tune and correct model outputs in collaboration with medical experts [98,99].

The potential of AI to create “information cocoons” through personalized content, potentially reinforcing biases, is another critical aspect that needs to be addressed, especially in the medical domain [100]. AI technologies, including LLMs, in medicine require stringent scrutiny and continuous evaluation to align with the field’s unique characteristics and ethical standards. Ensuring privacy protection, eliminating biases and discrimination, and establishing clear accountability are essential. The use of LLMs should be guided by respect for life, aiming to enhance patient well-being and treatment outcomes, without compromising individual interests. A continuous monitoring and evaluation system is crucial for assessing the effectiveness of LLMs and managing potential risks. Regulations should be regularly updated to keep pace with AI advancements, ensuring medical safety and patient rights. By prioritizing safety, fairness, and effectiveness, we can fully leverage LLMs and other AI technologies to facilitate a transformative revolution in medicine, while upholding human values and rights.

In the era of information and intelligence within the medical field, the application of LLMs harbors immense potential [101]. However, the accompanying challenges are equally noteworthy and merit careful consideration. The ongoing discourse should emphasize not only the deeper integration of LLMs into medical practice but also their alignment with both the professional needs of health care providers and the experiential needs of patients [102,103].

Incorporating the theory of mind into LLMs can significantly enhance their utility in the medical field. This concept, which involves understanding others’ thoughts, feelings, and intentions, is crucial for fostering trust and empathy within health care interactions. Medicine is not solely a science; it is also an art, deeply influenced by each patient’s unique emotional, value-based, and experiential landscape. An AI system endowed with the capability to appreciate and respond to these individual differences can offer more personalized and compassionate medical advice [104,105]. By using the theory of mind, LLMs can gain deeper insights into patients’ inherent needs and respond with more attentive and empathetic advice [106-108]. When LLMs can emulate the thoughts and feelings of both doctors and patients, their outputs transcend mere data; they become imbued with empathy and human care, enhancing the patient’s treatment experience and fostering stronger trust and communication between doctors and patients. For example, in interactions with terminal patients, LLMs could suggest more compassionate communication strategies, aiding both doctors and patients in navigating these sensitive and complex situations.

LLMs can be synergistically combined with other advanced technologies, such as virtual reality and augmented reality, to transform medical consultations into more immersive and informative experiences. This integration can provide patients with a deeper understanding of their health conditions, empowering them to make more informed decisions regarding their treatment. The evolution of LLMs is also contingent upon the development of efficient and precise algorithms capable of adeptly handling complex medical data, which is essential for accurate and timely medical decision-making. As technology progresses, the use of LLMs in the medical field is expected to become increasingly intelligent, efficient, and personalized, thereby enhancing not only the quality of medical services but also the overall patient experience and driving the evolution and transformation of the health care industry.

In our pursuit of technological progress, we must adhere to a fundamental principle: ensuring that technology is accessible to all. This is particularly pertinent in the context of LLM adoption, where it is crucial not to overlook those who may be marginalized by the technology gap [109,110]. Whether addressing the needs of rural farmers or urban older adults, every individual should have the opportunity to benefit from LLMs. This broad adoption must span various geographical regions and encompass diverse languages and cultural contexts, catering to users speaking English, Chinese, or local dialects [111,112]. Achieving this objective is not solely a technological challenge but also a social imperative. We must ensure that the design and application of LLMs overcome language and cultural barriers, truly reaching and benefiting a diverse global populace. Additionally, addressing technology accessibility issues is vital. For individuals in technologically underserved areas or older adults unfamiliar with new technologies, simpler access methods and more user-friendly interfaces are needed to facilitate effortless use of LLMs.

While the potential of LLMs in health care is significant, realizing this potential requires ongoing research, innovation, and dedication. Continuous efforts are necessary to refine LLM technology continually and ensure its broad adoption across all sectors of society. We firmly believe that with sustained commitment, LLMs will catalyze transformative changes in health care, benefiting society at large. By championing technological inclusivity, we can not only enhance the quality and efficiency of medical services but also promote overall societal health and well-being.

Economic Considerations in the Deployment of LLMs in Health Care

LLMs require significant computational resources for training and maintenance, which translates to substantial financial costs. In the medical domain, these costs can be particularly prohibitive due to the need for specialized data, high levels of accuracy, and continuous updates to ensure model relevance and safety.

Training a state-of-the-art LLM, such as GPT-3, requires access to extensive hardware infrastructure, including thousands of GPUs or TPUs, large amounts of RAM, and high-speed data storage solutions [113,114]. According to estimates, the training cost of models, like GPT-3, can reach up to US $1.4 million, with operational costs amounting to several hundred thousand dollars per day when deployed at scale. In a medical context, where accuracy and reliability are paramount, these costs are even higher due to the additional requirements for data security, privacy, and compliance with health care regulations.

Several studies have documented the economic challenges associated with deploying LLMs in health care [115,116]. For instance, the cost of implementing LLMs in hospital settings, including the necessary infrastructure upgrades, staff training, and ongoing maintenance, has been reported to be a major barrier to widespread adoption. Moreover, the need for regular updates to the models, which involves retraining them with new medical data, adds to the operational expenses [1].

As technology advances, it is expected that the costs associated with LLMs will decrease, making them more accessible to a broader range of health care providers. The development of more energy-efficient hardware, combined with advances in machine learning techniques, is likely to contribute to this trend. However, until these cost reductions are realized, careful planning and resource allocation will be essential for any institution looking to implement LLMs in their health care practice.

Conclusions

The era of digitalization and informatization underscores the transformative potential of LLMs in medicine. The evolution of this technology signifies a paradigm shift in medical services, offering unique opportunities and challenges to the medical community. LLMs, with their advanced NLP capabilities, have a wide range of applications including emergency triage, older people care, and the enhancement of digital medical workflows. As the diversity of medical data expands, LLMs’ ability to process multimodal data will play a crucial role in enabling more precise, personalized medical diagnoses and treatments.

Despite the promising trajectory of LLMs in the medical field, ensuring their safety and effectiveness in clinical practice remains a critical challenge. Currently, the regulation of LLMs in health care is still in its early stages, with several frameworks being developed to address the unique risks and challenges they pose. Regulatory bodies, such as the US Food and Drug Administration, European Medicines Agency, and China’s National Medical Products Administration, have begun to formulate guidelines that apply to AI-driven medical devices including LLMs. These guidelines typically focus on the validation of the models through rigorous clinical trials, ensuring that they meet specific safety, efficacy, and ethical standards before they can be deployed in clinical settings. However, the growth potential of LLMs in the medical arena is significant. They can enhance patient experiences through the integration of virtual reality and augmented reality, offer comprehensive medical advice through multimodal research, and humanize doctor-patient interactions using the theory of mind. With ongoing advancements in algorithms and computational power, we anticipate considerable improvements in LLMs’ processing speed and accuracy.

However, the path to technological advancement is not always linear. To ensure the benefits of LLMs are accessible to all, it is imperative to promote equitable development and address the digital divide, particularly for economically and technologically disadvantaged regions and groups. This goal requires the collective efforts of health care professionals, computer science experts, government regulatory bodies, patients, and their families. Such a collaborative approach will ensure that the application of LLM technology in the medical field genuinely contributes to the betterment of humanity, significantly enhancing health and well-being.

Acknowledgments

This study was funded by the National Key R & D Program of China (2020YFC2004705), the National Natural Science Foundation of China (81825003, 81900272, 91957123, 82270376, 623B2003), and the Beijing Nova Program (Z201100006820002) from Beijing Municipal Science & Technology Commission. This study was also supported by the S & T Program of Hebei (22377719D and 22377771D), the Key Science and Technology Research Program of Hebei Provincial Health Commission (20230991), the Industry University Research Cooperation Special Project (CXY2024020), the Hebei Province Finance Department Project (LS202214, ZF2024226), the Hebei Provincial Health Commission Project (20190448), the Key Discipline Construction Project of Shanghai Pudong New Area Health Commission (PWZxk2022-20), the Science and Technology Plan Project of Jiangxi Provincial Health Commission ([SK] P220227143), Tianjin University Science and Technology Innovation Leading Talent Cultivation Project (2024XQM-0024), and Tianjin Natural Science Foundation (23JCQNJC01430). This study was also supported by the China Hebei International Joint Research Center for Structural Heart Disease, and the Hebei Province Finance Department Project (LS202101).

Conflicts of Interest

None declared.

------------

Abbreviations

AI: artificial intelligence

BERT: Bidirectional Encoder Representations from Transformers

GPU: graphics processing unit

LLM: large language model

MRI: magnetic resonance imaging

NLP: natural language processing

TPU: tensor processing unit

------------

Edited by A Schwartz; submitted 01.04.24; peer-reviewed by Y Khan, K Singh, S Mao, D Patel, D Ghosh, P Sarmadi; comments to author 13.08.24; revised version received 26.08.24; accepted 10.09.24; published 07.01.25.

References

Arora A, Arora A. The promise of large language models in health care. Lancet. 2023;401(10377):641.

Ayers JW, Zhu Z, Poliak A, Leas EC, Dredze M, Hogarth M, et al. Evaluating artificial intelligence responses to public health questions. JAMA Netw Open. 2023;6(6):e2317517.

Thirunavukarasu AJ, Ting DSJ, Elangovan K, Gutierrez L, Tan TF, Ting DSW. Large language models in medicine. Nat Med. 2023;29(8):1930-1940.

Li R, Kumar A, Chen JH. How chatbots and large language model artificial intelligence systems will reshape modern medicine: fountain of creativity or pandora's box? JAMA Intern Med. 2023;183(6):596-597.

Minssen T, Vayena E, Cohen IG. The challenges for regulating medical use of ChatGPT and other large language models. JAMA. 2023;330(4):315-316.

Kanjee Z, Crowe B, Rodman A. Accuracy of a generative artificial intelligence model in a complex diagnostic challenge. JAMA. 2023;330(1):78-80.

Naveed H, Khan AU, Qiu S, Saqib M, Anwar S, Usman M, et al. A comprehensive overview of large language models. ArXiv. Preprint posted online on July 12, 2023.

Ayoub NF, Lee YJ, Grimm D, Balakrishnan K. Comparison between ChatGPT and Google search as sources of postoperative patient instructions. JAMA Otolaryngol Head Neck Surg. 2023;149(6):556-558.

Zhao WX, Zhou K, Li J, Tang T, Wang X, Hou Y, et al. A survey of large language models. ArXiv. Preprint posted online on March 31, 2023.

Chang K. Natural language processing: recent development and applications. Appl Sci. 2023;13(20):11395.

Ali SR, Dobbs TD, Hutchings HA, Whitaker IS. Using ChatGPT to write patient clinic letters. Lancet Digital Health. 2023;5(4):e179-e181.

Garcia MB. Using AI tools in writing peer review reports: should academic journals embrace the use of ChatGPT? Ann Biomed Eng. 2024;52(2):139-140.

Azerbayev Z, Schoelkopf H, Paster K, Santos MD, McAleer S, Jiang AQ, et al. Llemma: an open language model for mathematics. ArXiv. Preprint posted online on October 16, 2023.

Xiong W, Guo Y, Chen H. The program testing ability of large language models for code. ArXiv. Preprint posted online on October 09, 2023.

Yuan A, Coenen A, Reif E, Ippolito D, Assoc CM. Wordcraft: story writing with large language models. 2022. Presented at: Wordcraft Story Writing With Large Language Models. 27th Annual International Conference on Intelligent User Interfaces (ACM IUI); 2022 2022 Mar ; Univ Helsinki, ELECTR NETWORK; 2022; March 22, 2022:841-852; Helsinki, Finland.

Chu Y, Liu P. Public aversion against ChatGPT in creative fields? Innovation (Camb). 2023;4(4):100449.

Voelker R. The promise and pitfalls of AI in the complex world of diagnosis, treatment, and disease management. JAMA. 2023;330(15):1416-1419.

Singhal K, Azizi S, Tu T, Mahdavi SS, Wei J, Chung HW, et al. Large language models encode clinical knowledge. Nature. 2023;620(7972):172-180.

He S, Yang F, Zuo JP, Lin ZM. ChatGPT for scientific paper writing-promises and perils. Innovation (Camb). 2023;4(6):100524.

Ayers JW, Poliak A, Dredze M, Leas EC, Zhu Z, Kelley JB, et al. Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Intern Med. 2023;183(6):589-596.

Abd-Alrazaq A, AlSaad R, Alhuwail D, Ahmed A, Healy PM, Latifi S, et al. Large language models in medical education: opportunities, challenges, and future directions. JMIR Med Educ. 2023;9:e48291.

Cooper A, Rodman A. AI and medical education—a 21st-century pandora's box. N Engl J Med. 2023;389(5):385-387.

Mihalache A, Popovic MM, Muni RH. Performance of an artificial intelligence chatbot in ophthalmic knowledge assessment. JAMA Ophthalmol. 2023;141(6):589-597.

Mehrabanian M, Zariat Y. ChatGPT passes anatomy exam. Br Dent J. 2023;235(5):295.

Baker S. AI could be an opportunity for research managers. Nature. 2023.

Ray PP. Broadening the horizon: a call for extensive exploration of ChatGPT's potential in obstetrics and gynecology. Am J Obstet Gynecol. 2023;229(6):706.

Kovoor JG, Gupta AK, Bacchi S. ChatGPT: effective writing is succinct. BMJ. 2023;381:1125.

Decker H, Trang K, Ramirez J, Colley A, Pierce L, Coleman M, et al. Large language model-based chatbot vs surgeon-generated informed consent documentation for common procedures. JAMA Netw Open. 2023;6(10):e2336997.

Madani A, Krause B, Greene ER, Subramanian S, Mohr BP, Holton JM, et al. Large language models generate functional protein sequences across diverse families. Nat Biotechnol. 2023;41(8):1099-1106.

Askr H, Elgeldawi E, Ella HA, Elshaier YAMM, Gomaa MM, Hassanien AE. Deep learning in drug discovery: an integrative review and future challenges. Artif Intell Rev. 2023;56(7):5975-6037.

Yang X, Chen A, PourNejatian N, Shin HC, Smith KE, Parisien C, et al. A large language model for electronic health records. NPJ Digital Med. 2022;5(1):194.

Brown T, Mann B, Ryder N, Subbiah M, Kaplan JD, Dhariwal P. Language models are few-shot learners. 2020. Presented at: Proceedings of the 34th International Conference on Neural Information Processing Systems; 2020 December 06:1877-1901; Red Hook, NY, United States.

Ramesh A, Dhariwal P, Nichol A, Chu C, Chen M. Hierarchical text-conditional image generation with clip latents. ArXiv. Preprint posted online on October 13, 2022.

Raffel C, Shazeer N, Roberts A, Lee K, Narang S, Matena M. Exploring the limits of transfer learning with a unified text-to-text transformer. J Mach Learn Technol. 2020;21(1):5485-5551.

Carpenter KA, Altman RB. Using GPT-3 to build a lexicon of drugs of abuse synonyms for social media pharmacovigilance. Biomolecules. 2023;13(2):387.

Ali R, Tang OY, Connolly ID, Fridley JS, Shin JH, Sullivan PLZ, et al. Performance of ChatGPT, GPT-4, and Google bard on a neurosurgery oral boards preparation question bank. Neurosurgery. 2023;93(5):1090-1098.

Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A. Attention is all you need. In: Advances in Neural Information Processing Systems. United States. Curran Associates, Inc; 2017.

Manolitsis I, Feretzakis G, Tzelves L, Kalles D, Katsimperis S, Angelopoulos P, et al. Training ChatGPT models in assisting urologists in daily practice. Stud Health Technol Inform. 2023;305:576-579.

An J, Ding W, Lin C. ChatGPT: tackle the growing carbon footprint of generative AI. Nature. 2023;615(7953):586.

Sharir O, Peleg B, Shoham Y. The cost of training nlp models: a concise overview. ArXiv. Preprint posted online on April 19, 2020.

Srinivasan V, Gandhi D, Thakker U, Prabhakar R. Training large language models efficiently with sparsity and dataflow. ArXiv. Preprint posted online on April 11, 2023.

Zhou Z, Wang X, Li X, Liao L. Is ChatGPT an evidence-based doctor? Eur Urol. 2023;84(3):355-356.

Blum J, Menta AK, Zhao X, Yang VB, Gouda MA, Subbiah V. Pearls and pitfalls of ChatGPT in medical oncology. Trends Cancer. 2023;9(10):788-790.

Duffourc M, Gerke S. Generative AI in health care and liability risks for physicians and safety concerns for patients. JAMA. 2023;330(4):313-314.

Ji J, Qiu T, Chen B, Zhang B, Lou H, Wang K, et al. AI alignment: a comprehensive survey. ArXiv. Preprint posted online on October 30, 2023.

Ward E, Gross C. Evolving methods to assess chatbot performance in health sciences research. JAMA Intern Med. 2023;183(9):1030-1031.

Butte AJ. Artificial intelligence-from starting pilots to scalable privilege. JAMA Oncol. 2023;9(10):1341-1342.

Hu ZY, Han FJ, Yu L, Jiang Y, Cai G. AI-link omnipotent pathological robot: bridging medical meta-universe to real-world diagnosis and therapy. Innovation (Camb). 2023;4(5):100494.

Thirunavukarasu AJ. Large language models will not replace healthcare professionals: curbing popular fears and hype. J R Soc Med. 2023;116(5):181-182.

The Lancet Regional Health-Europe. Embracing generative AI in health care. Lancet Reg Health Eur. 2023;30:100677.

NA. Will ChatGPT transform healthcare? Nat Med. 2023;29(3):505-506.

Jiang LY, Liu XC, Nejatian NP, Nasir-Moin M, Wang D, Abidin A, et al. Health system-scale language models are all-purpose prediction engines. Nature. 2023;619(7969):357-362.

Shah NH, Entwistle D, Pfeffer MA. Creation and adoption of large language models in medicine. JAMA. 2023;330(9):866-869.

Kluger N. Potential applications of ChatGPT in dermatology. J Eur Acad Dermatol Venereol. 2023;37(7):e941-e942.

Howard A, Hope W, Gerada A. ChatGPT and antimicrobial advice: the end of the consulting infection doctor? Lancet Infect Dis. 2023;23(4):405-406.

Kulkarni PA, Singh H. Artificial intelligence in clinical diagnosis: opportunities, challenges, and hype. JAMA. 2023;330(4):317-318.

Jiang K, Zhu M, Bernard G. Few-shot learning for identification of COVID-19 symptoms using generative pre-trained transformer language models. In: Koprinska I, editor. Machine Learning and Principles and Practice of Knowledge Discovery in Databases. ECML PKDD 2022. Communications in Computer and Information Science, vol 1753. Switzerland. Springer, Cham; 2023.

Miller K, Gunn E, Cochran A, Burstein H, Friedberg JW, Wheeler S, et al. Use of large language models and artificial intelligence tools in works submitted to journal of clinical oncology. J Clin Oncol. 2023;41(19):3480-3481.

Goodman RS, Patrinely JR, Stone CA, Zimmerman E, Donald RR, Chang SS, et al. Accuracy and reliability of chatbot responses to physician questions. JAMA Netw Open. 2023;6(10):e2336483.

Hua HU, Kaakour AH, Rachitskaya A, Srivastava S, Sharma S, Mammo DA. Evaluation and comparison of ophthalmic scientific abstracts and references by current artificial intelligence chatbots. JAMA Ophthalmol. 2023;141(9):819-824.

Chen S, Kann BH, Foote MB, Aerts HJWL, Savova GK, Mak RH, et al. Use of artificial intelligence chatbots for cancer treatment information. JAMA Oncol. 2023;9(10):1459-1462.

Marchandot B, Matsushita K, Carmona A, Trimaille A, Morel O. ChatGPT: the next frontier in academic writing for cardiologists or a pandora's box of ethical dilemmas. Eur Heart J Open. 2023;3(2):oead007.

Lee P, Bubeck S, Petro J. Benefits, limits, and risks of GPT-4 as an AI chatbot for medicine. N Engl J Med. 2023;388(25):2400.

Lee P, Bubeck S, Petro J. Benefits, limits, and risks of GPT-4 as an AI chatbot for medicine. N Engl J Med. 2023;388(13):1233-1239.

Kataoka Y, So R. Benefits, limits, and risks of GPT-4 as an AI chatbot for medicine. N Engl J Med. 2023;388(25):2399.

Fernandes AC, Souto MEVC. Benefits, limits, and risks of GPT-4 as an AI chatbot for medicine. N Engl J Med. 2023;388(25):2399-2400.

Gottlieb S, Silvis L. How to safely integrate large language models into health care. JAMA Health Forum. 2023;4(9):e233909.

Pan A, Musheyev D, Bockelman D, Loeb S, Kabarriti AE. Assessment of artificial intelligence chatbot responses to top searched queries about cancer. JAMA Oncol. 2023;9(10):1437-1440.

Bernstein IA, Zhang YV, Govil D, Majid I, Chang RT, Sun Y, et al. Comparison of ophthalmologist and large language model chatbot responses to online patient eye care questions. JAMA Netw Open. 2023;6(8):e2330320.

Kiros R, Salakhutdinov R, Zemel R. Multimodal neural language models. 2014. Presented at: Proceedings of the 31st International Conference on International Conference on Machine Learning—Volume 32; June 21, 2014; Beijing China. URL: https://proceedings.mlr.press/v32/kiros14.html

Driess D, Xia F, Sajjadi MSM, Lynch C, Chowdhery A, Ichter B, et al. Palm-e: an embodied multimodal language model. ArXiv. Preprint posted online on March 06, 2023.

Zhang T, Liu N, Xu J, Liu Z, Zhou Y, Yang Y, et al. Flexible electronics for cardiovascular healthcare monitoring. Innovation (Camb). 2023;4(5):100485.

Volpe NJ, Mirza RG. Chatbots, artificial intelligence, and the future of scientific reporting. JAMA Ophthalmol. 2023;141(9):824-825.

Li S. ChatGPT has made the field of surgery full of opportunities and challenges. Int J Surg. 2023;109(8):2537-2538.

O'Hern K, Yang E, Vidal NY. ChatGPT underperforms in triaging appropriate use of Mohs surgery for cutaneous neoplasms. JAAD Int. 2023;12:168-170.

Xu HL, Gong TT, Liu FH, Chen HY, Xiao Q, Hou Y, et al. Artificial intelligence performance in image-based ovarian cancer identification: a systematic review and meta-analysis. EClinicalMedicine. 2022;53:101662.

Zhu M, Chen Z, Yuan Y. DSI-Net: deep synergistic interaction network for joint classification and segmentation with endoscope images. IEEE Trans Med Imaging. 2021;40(12):3315-3325.

Cheng K, Wu C, Gu S, Lu Y, Wu H, Li C. WHO declares the end of the COVID-19 global health emergency: lessons and recommendations from the perspective of ChatGPT/GPT-4. Int J Surg. 2023;109(9):2859-2862.

Madden MG, McNicholas BA, Laffey JG. Assessing the usefulness of a large language model to query and summarize unstructured medical notes in intensive care. Intensive Care Med. 2023;49(8):1018-1020.

Li H, Moon JT, Purkayastha S, Celi LA, Trivedi H, Gichoya JW. Ethics of large language models in medicine and medical research. Lancet Digital Health. 2023;5(6):e333-e335.

van Heerden AC, Pozuelo JR, Kohrt BA. Global mental health services and the impact of artificial intelligence-powered large language models. JAMA Psychiatry. 2023;80(7):662-664.

Kwok K, Wei W, Tsoi M, Tang A, Chan M, Ip M, et al. How can we transform travel medicine by leveraging on AI-powered search engines? J Travel Med. 2023;30(4):taad058.

NA. AI-powered structure-based drug design inspired by the lock-and-key model. Nat Comput Sci. 2023;3(10):827-828.

NA. Upswing in AI drug-discovery deals. Nat Biotechnol. 2023;41(10):1361.

Xu Z, editor. Using large pre-trained language model to assist FDA in premarket medical device classification. In: SoutheastCon. Orlando, FL, USA. IEEE; 2023.

Wu C, Lei J, Zheng Q, Zhao W, Lin W, Zhang X. Can GPT-4V (ision) Serve Medical Applications? Case Studies on GPT-4V for Multimodal Medical Diagnosis. ArXiv. Preprint posted online on October 15, 2023.

He K, Mao R, Lin Q, Ruan Y, Lan X, Feng M. A survey of large language models for healthcare: from data, technology, and applications to accountability and ethics. ArXiv. Preprint posted online on October 09, 2023.

Haug CJ, Drazen JM. Artificial intelligence and machine learning in clinical medicine, 2023. N Engl J Med. 2023;388(13):1201-1208.

Marks M, Haupt CE. AI chatbots, health privacy, and challenges to HIPAA compliance. JAMA. 2023;330(4):309-310.

Mello MM, Guha N. ChatGPT and physicians' malpractice risk. JAMA Health Forum. 2023;4(5):e231938.

Kanter GP, Packel EA. Health care privacy risks of AI chatbots. JAMA. 2023;330(4):311-312.

NA. ChatGPT is a black box: how AI research can break it open. Nature. 2023;619(7971):671-672.

Tang YD, Dong ED, Gao W. LLMs in medicine: the need for advanced evaluation systems for disruptive technologies. Innovation (Camb). 2024;5(3):100622.

Grigorian A, Shipley J, Nahmias J, Nguyen N, Schwed AC, Petrie BA, et al. Implications of using chatbots for future surgical education. JAMA Surg. 2023;158(11):1220-1222.

Kozlov M, Biever C. AI 'breakthrough': neural net has human-like ability to generalize language. Nature. 2023;623(7985):16-17.

Ferryman K, Mackintosh M, Ghassemi M. Considering biased data as informative artifacts in AI-assisted health care. N Engl J Med. 2023;389(9):833-838.

Tan TF, Teo ZL, Ting DSW. Artificial intelligence bias and ethics in retinal imaging. JAMA Ophthalmol. 2023;141(6):552-553.

Harris E. Large language models answer medical questions accurately, but can't match clinicians' knowledge. JAMA. 2023;330(9):792-794.

Kim J, Cai ZR, Chen ML, Simard JF, Linos E. Assessing biases in medical decisions via clinician and AI chatbot responses to patient vignettes. JAMA Netw Open. 2023;6(10):e2338050.

Piao J, Liu J, Zhang F, Su J, Li Y. Human–AI adaptive dynamics drives the emergence of information cocoons. Nat Mach Intell. 2023;5(11):1214-1224.

Nordling L. How ChatGPT is transforming the postdoc experience. Nature. 2023;622(7983):655-657.

The Lancet. AI in medicine: creating a safe and equitable future. Lancet. 2023;402(10401):503.

The Lancet. AI in medicine: creating a safe and equitable future. Lancet. 2023;402(10401):503.

Topol EJ. Machines and empathy in medicine. Lancet. 2023;402(10411):1411.

Hagendorff T, Fabi S, Kosinski M. Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT. Nat Comput Sci. 2023;3(10):833-838.

Yang R, Tan TF, Lu W, Thirunavukarasu AJ, Ting DSW, Liu N. Large language models in health care: development, applications, and challenges. Health Care Sci. 2023;2(4):255-263.

Wu Y, Si Z, Gong H, Zhu SC. Learning active basis model for object detection and recognition. Int J Comput Vis. 2009;90(2):198-235.

Horgan J. The consciousness conundrum. IEEE Spectr. 2008;45(6):36-41.

Lam K. ChatGPT for low- and middle-income countries: a Greek gift? Lancet Reg Health West Pac. 2023;41:100906.

Wang X, Sanders HM, Liu Y, Seang K, Tran BX, Atanasov AG, et al. ChatGPT: promise and challenges for deployment in low- and middle-income countries. Lancet Reg Health West Pac. 2023;41:100905.

Hswen Y, Voelker R. New AI tools must have health equity in their DNA. JAMA. 2023;330(17):1604-1607.

Seghier ML. ChatGPT: not all languages are equal. Nature. 2023;615(7951):216.

Tan B, Zhu Y, Liu L, Wang H, Zhuang Y, Chen J. RedCoast: a lightweight tool to automate distributed training of LLMs on any GPU/TPUs. 2024. Presented at: Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 3: System Demonstrations); June 01, 2024:137-147; Mexico City, Mexico.

Narayanan D, Shoeybi M, Casper J, LeGresley P, Patwary M, Korthikanti V, et al. Efficient large-scale language model training on GPU clusters using megatron-LM. 2021. Presented at: SC21: International Conference for High Performance Computing, Networking, Storage and Analysis; November 13, 2021:1-15; New York, NY, United States.

Thirunavukarasu AJ, Ting DSJ, Elangovan K, Gutierrez L, Tan TF, Ting DSW. Large language models in medicine. Nat Med. 2023;29(8):1930-1940.

Meng X, Yan X, Zhang K, Liu D, Cui X, Yang Y, et al. The application of large language models in medicine: a scoping review. iScience. 2024;27(5):109713.