Comparative Evaluation of Gemini 3.0- and ChatGPT 5.0-Generated Regional Language Informed Consent Forms in Ophthalmology: A Dual-Rater Study in Hindi and Kannada - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Cureus . 2026 Mar 20;18(3):e105551. doi: 10.7759/cureus.105551 Search in PMC Search in PubMed View in NLM Catalog Add to search Comparative Evaluation of Gemini 3.0- and ChatGPT 5.0-Generated Regional Language Informed Consent Forms in Ophthalmology: A Dual-Rater Study in Hindi and Kannada Deepsekhar Das Deepsekhar Das 1 Ophthalmology, All India Institute of Medical Sciences, New Delhi, New Delhi, IND Find articles by Deepsekhar Das 1 , Bahubali Shetti Bahubali Shetti 1 Ophthalmology, All India Institute of Medical Sciences, New Delhi, New Delhi, IND Find articles by Bahubali Shetti 1 , Venkatesh Antalmarad Venkatesh Antalmarad 1 Ophthalmology, All India Institute of Medical Sciences, New Delhi, New Delhi, IND Find articles by Venkatesh Antalmarad 1 , Anshum Choudhary Anshum Choudhary 1 Ophthalmology, All India Institute of Medical Sciences, New Delhi, New Delhi, IND Find articles by Anshum Choudhary 1 , Kathyayini G Thodupunoori Kathyayini G Thodupunoori 1 Ophthalmology, All India Institute of Medical Sciences, New Delhi, New Delhi, IND Find articles by Kathyayini G Thodupunoori 1 , Sumit Grover Sumit Grover 1 Ophthalmology, All India Institute of Medical Sciences, New Delhi, New Delhi, IND Find articles by Sumit Grover 1 , Atindra Narayan Atindra Narayan 2 Medicine, All India Institute of Medical Sciences, New Delhi, New Delhi, IND Find articles by Atindra Narayan 2, ✉ Editors: Alexander Muacevic , John R Adler Author information Article notes Copyright and License information 1 Ophthalmology, All India Institute of Medical Sciences, New Delhi, New Delhi, IND 2 Medicine, All India Institute of Medical Sciences, New Delhi, New Delhi, IND ✉ Atindra Narayan [email protected] ✉ Corresponding author. Accepted 2026 Mar 20; Collection date 2026 Mar. Copyright © 2026, Das et al. This is an open access article distributed under the terms of the Creative Commons Attribution License CC-BY 4.0., which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. PMC Copyright notice PMCID: PMC13092168 PMID: 42011474 Abstract Purpose: To evaluate the accuracy, linguistic quality, clinical completeness, and real-world applicability of informed consent forms generated in Indian regional languages (Hindi and Kannada) by two large language model-based chatbots for common ophthalmic surgical procedures. Methods: In this comparative, blinded observational study, two chatbots (Gemini 3.0 (Google, California, USA) and ChatGPT 5.0 (OpenAI, California, USA)) were prompted to generate informed consent documents in Hindi and Kannada for five ophthalmic scenarios: cataract surgery, traumatic corneal perforation repair, therapeutic penetrating keratoplasty, orbitotomy, and squint surgery. Outputs were independently assessed by four ophthalmologist raters (two for each language) using a 10-point scoring system based on correctness, completeness, language and readability, clinical relevance, and real-world applicability. Descriptive statistics were calculated. Paired t-tests were used to compare chatbot performance, effect sizes (Cohen’s d) were estimated, and inter-rater reliability was assessed using intraclass correlation coefficients (ICCs). Results: Across both languages, Gemini 3.0 demonstrated more consistent performance and higher combined mean scores. In the Hindi cohort, combined mean scores were comparable between Gemini 3.0 (7.85) and ChatGPT 5.0 (8.00), with significant rater-dependent preference variability. In contrast, in the Kannada cohort, Gemini 3.0 significantly outperformed ChatGPT 5.0 (8.8 vs 7.7, p<0.01), with large to extremely large effect sizes (Cohen’s d: 1.23-3.8). Inter-rater reliability was moderate to good for Gemini 3.0 (ICC: 0.62-0.71) and lower for ChatGPT 5.0 (ICC: 0.38-0.59). ChatGPT 5.0 exhibited frequent grammatical and terminological inaccuracies, particularly in Kannada, affecting clinical usability. Conclusion: Large language models can generate clinically usable informed consent forms in Indian regional languages; however, performance varies significantly between models. Gemini 3.0 demonstrated superior linguistic accuracy, consistency, and clinical suitability. Language-specific validation and mandatory human oversight are essential before clinical implementation. Keywords: ai generated consent, chatgpt 5.0, gemini 3.0, informed consent, large language model, ophthalmology Introduction Informed consent is a cornerstone of ethical and legal medical practice, ensuring that patients understand the nature, benefits, risks, and alternatives of proposed interventions before agreeing to treatment [ 1 ]. Inadequate or poorly understood consent can undermine patient autonomy and expose clinicians to medico-legal risk. In countries such as India, linguistic diversity and variable health literacy further complicate the consent process, particularly when standardized documents are unavailable in regional languages. Ophthalmology involves a wide range of surgical procedures, many of which are elective or semi-elective, making clear communication and shared decision-making especially important. However, clinicians often rely on ad hoc translations or verbal explanations when formal consent documents are not available in the patient’s native language. Such practices may lead to inconsistencies and omissions, reducing the quality and defensibility of consent [ 2 ]. Recent advances in large language models (LLMs) have demonstrated their ability to generate coherent and contextually relevant medical text, including patient education material and clinical documentation [ 3 , 4 ]. These tools offer potential advantages in standardising consent forms, reducing clinician workload, and improving accessibility. However, most existing evaluations of LLMs have focused on English-language outputs, with limited evidence regarding their performance in Indian regional languages such as Hindi and Kannada [ 5 , 6 ]. Regional-language medical text generation poses unique challenges, including accurate translation of technical terminology, grammatical correctness, cultural appropriateness, and alignment with local clinical practice. Errors in these domains may impair patient comprehension or compromise the legal validity of consent. Moreover, the subjective nature of language assessment necessitates human evaluation by clinicians familiar with both medical content and local linguistic usage. The present study aims to comparatively evaluate two extremely popular advanced LLM-based chatbots for their ability to generate informed consent forms in Hindi and Kannada for common ophthalmic surgical procedures. Using a blinded dual-rater design, we assessed accuracy, completeness, linguistic quality, and real-world applicability, with the goal of informing safe and responsible integration of AI-generated consent documents into clinical practice. Materials and methods This was a comparative, cross-sectional evaluation study conducted using anonymised outputs from two large language model-based chatbots, Gemini 3.0 (Google, California, USA) and ChatGPT 5.0 (OpenAI, California, USA). The study was conducted at All India Institute of Medical Sciences (AIIMS), New Delhi, as a one-day observational study. The main objective of the study was to evaluate the accuracy, linguistic quality, clinical completeness, and real-world applicability of informed consent forms generated in Indian regional languages (Hindi and Kannada) by Gemini 3.0 and ChatGPT 5.0 chatbots for common ophthalmic surgical procedures. Each chatbot was prompted to generate informed consent forms in Hindi and Kannada for five ophthalmic surgical scenarios: phacoemulsification for cataract, traumatic corneal perforation repair, therapeutic penetrating keratoplasty, orbitotomy, and squint surgery. Prompts were standardized across both models to minimise prompt-related bias. Four independent ophthalmologists (two for Hindi and two for Kannada) with experience in clinical consent and regional language usage served as raters. The raters were blinded to the identity of the chatbots. Each generated consent document was evaluated using a structured 10-point scoring system encompassing correctness of medical content, completeness of information, language and readability, grammatical accuracy, inclusion of relevant and exclusion of irrelevant data, and real-world applicability in the Indian clinical setting. Scores were assigned independently by each rater for all chatbot outputs in both languages. Descriptive statistics, including mean and standard deviation, were calculated. Paired t-tests were used to compare the performance of Gemini 3.0 and ChatGPT 5.0 separately for each rater and language. Effect sizes were estimated using Cohen’s d to assess the magnitude of differences. Inter-rater reliability was evaluated using the intraclass correlation coefficient (ICC), interpreted according to standard benchmarks. All statistical analyses were conducted using standard statistical software, with a two-sided p value <0.05 considered statistically significant. Results A total of 20 consent documents were evaluated, comprising five surgical scenarios each in Hindi and Kannada generated by two chatbots and assessed independently by two raters. Both chatbots were able to generate complete consent documents across all scenarios; however, notable differences were observed in accuracy, linguistic quality, and real-world applicability. Hindi-language evaluation In the Hindi cohort, the combined mean scores across both raters were comparable between the two chatbots, with Gemini 3.0 achieving a mean score of 7.85 and ChatGPT 5.0 achieving 8.00 (Table 1 ). Despite similar overall performance, rater-specific analyses revealed significant rater disagreements. Rater 1 scored Gemini 3.0 significantly higher than ChatGPT 5.0 (mean difference: +0.8; p=0.016), whereas Rater 2 scored ChatGPT 5.0 significantly higher than Gemini 3.0 (mean difference: -1.1; p=0.0027) (Table 2 ). Table 1. Hindi-language consent evaluation scores. Hindi consents Surgical scenario Rater 1 (Gemini 3.0) Rater 1 (ChatGPT 5.0) Rater 2 (Gemini 3.0) Rater 2 (ChatGPT 5.0) Cataract surgery 8.0 7.0 6.5 7.5 Corneal perforation repair 8.0 8.0 7.0 8.5 Therapeutic PK 8.0 7.0 8.0 9.0 Orbitotomy 8.0 7.0 8.0 9.5 Squint surgery 8.0 7.0 9.0 9.5 Mean score 8.0 7.2 7.7 8.8 Open in a new tab Table 2. Paired-statistical comparison of chatbot performance. Paired-statistical comparison of chatbot performance Language Rater Mean difference (Gemini 3.0-ChatGPT 5.0) p value Interpretation Hindi Rater 1 +0.8 0.016 Gemini 3.0 is better Hindi Rater 2 -1.1 0.0027 ChatGPT 5.0 is better Kannada Rater 1 +1.4 0.001 Gemini 3.0 is better Kannada Rater 2 +0.8 0.049 Gemini 3.0 is better Open in a new tab Effect size analysis demonstrated very large to extremely large differences for both raters (Cohen’s d ranging from 1.79 to 2.93), indicating strong rater-dependent preferences rather than marginal differences in output quality (Table 3 ). Inter-rater reliability was moderate for Gemini 3.0 (ICC=0.62) but low for ChatGPT 5.0 (ICC=0.38), suggesting greater variability and reduced consistency in the evaluation of ChatGPT 5.0’s Hindi outputs. Table 3. Effect size and inter-rater reliability. Effect size and inter-rater reliability Language Chatbot Cohen’s d Effect size ICC Agreement strength Hindi Gemini 3.0 1.79 Very large 0.62 Moderate Hindi ChatGPT 5.0 2.93 Extremely large 0.38 Low Kannada Gemini 3.0 1.23-3.8 Large-Extremely large 0.71 Good Kannada ChatGPT 5.0 1.10 Large 0.59 Moderate Open in a new tab Kannada-language evaluation In contrast to Hindi, the Kannada cohort demonstrated consistent and statistically significant dominance of Gemini 3.0. The combined mean score for Gemini 3.0 was 8.8 compared to 7.7 for ChatGPT 5.0, yielding a mean difference of +1.1 points in favour of Gemini 3.0 (Table 4 ). Both raters independently rated Gemini 3.0 significantly higher than ChatGPT 5.0 (Rater 1: p=0.001; Rater 2: p=0.049) (Table 2 ). Table 4. Kannada-language consent evaluation scores. PK: penetrating keratoplasty. Kannada consents Surgical scenario Rater 1 (Gemini 3.0) Rater 1 (ChatGPT 5.0) Rater 2 (Gemini 3.0) Rater 2 (ChatGPT 5.0) Cataract surgery 8.5 7.5 5.0 6.0 Corneal perforation repair 8.5 7.0 10.0 8.0 Therapeutic PK 9.0 7.0 9.0 9.0 Orbitotomy 9.0 8.0 10.0 9.0 Squint surgery 9.0 7.5 10.0 8.0 Mean score 8.8 7.4 8.8 8.0 Open in a new tab Effect sizes for the Kannada analysis were large to extremely large (Cohen’s d ranging from 1.23 to 3.8), indicating statistically significant differences between the two models (Table 5 ). Inter-rater reliability was good for Gemini 3.0 (ICC=0.71) and moderate for ChatGPT 5.0 (ICC=0.59), reflecting greater agreement between raters when evaluating Gemini 3.0’s Kannada outputs. Table 5. Common qualitative error patterns observed. Common qualitative error patterns observed Domain Gemini 3.0 ChatGPT 5.0 Medical terminology Accurate regional terms Frequent incorrect translations Grammar Mostly error-free Recurrent grammatical errors Transliteration Minimal Incorrect transliteration of English terms Cultural suitability High Variable Real-world applicability Consistently suitable Limited in Kannada Output consistency High Variable Open in a new tab Overall performance across languages When both languages were considered together, Gemini 3.0 demonstrated a higher overall mean score (8.33) compared to ChatGPT 5.0 (7.85), corresponding to an overall mean difference of +0.48 points (Table 2 ). Gemini 3.0 also exhibited more stable performance across languages, with consistently higher inter-rater agreement and fewer extreme score variations (Table 3 ). Qualitative observations Qualitative analysis supported the quantitative findings. Gemini 3.0 consistently used appropriate regional medical terminology, demonstrated grammatical correctness, and adhered closely to standard consent structure. In contrast, ChatGPT 5.0 frequently exhibited grammatical errors, incorrect translations of common ophthalmic terms, and inappropriate transliterations, particularly in Kannada, negatively affecting real-world clinical applicability. These qualitative deficiencies aligned with the lower scores, larger variability, and reduced inter-rater reliability observed for ChatGPT 5.0 (Table 5 ). Confounding factors Rater Subjectivity and Linguistic Preference Although raters were blinded to chatbot identity, individual differences in linguistic sensitivity, familiarity with regional medical terminology, and personal preferences for brevity versus detail may have influenced scoring. This was particularly evident in the Hindi cohort, where rater preferences diverged significantly despite similar objective content quality. Variation in Clinical Emphasis Between Raters Raters may have prioritised different aspects of informed consent, such as procedural detail, complication disclosure, prognosis, or post-operative instructions. Such differences can affect overall scores even when the generated content is factually correct, introducing evaluation bias. Differences in Regional Language Proficiency Although the raters were proficient in Hindi and Kannada, subtle differences in fluency, dialect familiarity, or exposure to standardized medical Kannada or Hindi could influence judgments regarding grammatical accuracy and real-world applicability. Non-Standardised Weighting of Evaluation Domains While a structured scoring framework was used, individual raters may have implicitly weighted certain domains (e.g., correctness or terminology) more heavily than others (e.g., sentence length or style), potentially confounding comparative scores between chatbots. Prompt Interpretation Variability by Chatbots Despite using standardized prompts, large language models may interpret prompts differently due to architectural or training differences. Variations in verbosity, structure, or emphasis may reflect prompt-response dynamics rather than inherent model capability. Discussion Artificial intelligence is currently being utilized in a multitude of ways in healthcare. Its use in ophthalmology has seen a tremendous rise in the recent past. They are routinely used in obtaining information from patients as well as healthcare personnel. They are also being used in identifying pathologies based on their clinical pictures, especially fundus images for diseases like diabetic retinopathy, age-related macular degeneration, and glaucoma [ 7 - 11 ]. The authors have identified the limitations of one such popular AI model with respect to cancer patients [ 12 ]. Informed consents are extremely vital documents, the construction of which is quite tedious, and the authors have previously studied the role of large language models in the generation of ophthalmology consent forms in English. One of the important findings was that the AI models were in a significant number of situations generating consent with additional relevant information, which was lacking in the standard All India Ophthalmological Society (AIOS) consent forms [ 6 ]. As the existing LLMs are capable of generating information in different Indian languages, the authors found it logical to conduct another study testing the more advanced versions of chatbots in generating informed consents in regional languages. This dual-rater comparative study evaluated the performance of two large language model-based chatbots in generating informed consent documents in Hindi and Kannada for common ophthalmic surgical procedures. The findings demonstrate that while both chatbots were capable of producing structurally coherent consent forms, their linguistic accuracy, clinical reliability, and real-world applicability varied substantially across languages and models. A key observation was the language-dependent performance of the chatbots. In Hindi, both models achieved comparable overall mean scores, but with pronounced rater-dependent variability. One rater favoured Gemini 3.0 for clarity and consistency, while the other favoured ChatGPT 5.0 for relative completeness. This divergence underscores the subjective component inherent in evaluating patient-facing medical text, even among trained clinicians. Moderate to low inter-rater reliability, particularly for ChatGPT 5.0, suggests that its Hindi outputs were less predictable and more sensitive to individual rater interpretation. In contrast, the Kannada analysis revealed a consistent and statistically significant superiority of Gemini 3.0 across both raters, with large to extremely large effect sizes. ChatGPT 5.0 demonstrated frequent grammatical errors, incorrect translations of standard ophthalmic terminology, and inappropriate transliteration of English medical terms. Such errors are not merely linguistic but medico-legally relevant, as inaccurate terminology can compromise patient comprehension and potentially invalidate informed consent. These findings highlight the importance of evaluating AI tools in each target language rather than assuming uniform multilingual competence. The qualitative error analysis further strengthens this conclusion. Gemini 3.0 showed better adherence to established consent structures, more appropriate use of regionally accepted medical vocabulary, and fewer semantic inaccuracies. ChatGPT 5.0, while sometimes more verbose, frequently included incorrect or culturally incongruent terminology, particularly in Kannada. This aligns with prior literature indicating that large language models trained predominantly on English or high-resource languages may underperform in low-resource or morphologically complex languages without targeted fine-tuning [ 13 ]. Inter-rater reliability was consistently higher for Gemini 3.0, suggesting greater output stability and predictability, an essential requirement for clinical deployment. Variability in ChatGPT 5.0’s performance raises concerns regarding its unsupervised use in medico-legal documentation. These findings reinforce current ethical recommendations that AI-generated clinical documents should be used only as assistive tools, with mandatory human oversight [ 3 , 4 ]. From a broader perspective, this study contributes to the emerging body of evidence on AI-assisted patient communication in regional languages. Prior studies have primarily focused on English-language outputs or general patient education material [ 1 , 2 , 5 ]. By specifically evaluating informed consent, a legally sensitive and ethically critical document, this work addresses a significant gap. The results suggest that regional-language validation should be considered a prerequisite before deploying AI systems in multilingual healthcare settings such as India. Nevertheless, this study had limitations. The sample size was small, restricted to ophthalmology, and involved only two raters for each language. Patient comprehension and acceptability were not assessed. Future studies should include larger numbers of procedures, multiple specialties, more raters, and patient-centred outcome measures. Additionally, longitudinal evaluation is needed to account for model updates over time. Conclusions In conclusion, while large language models show promise in generating informed consent documents in Indian regional languages, performance is highly language- and model-dependent. Gemini 3.0 demonstrated superior linguistic accuracy, consistency, and clinical suitability, particularly in Kannada. Careful validation, language-specific testing, and clinician oversight are essential before clinical adoption. Disclosures Human subjects: All authors have confirmed that this study did not involve human participants or tissue. Animal subjects: All authors have confirmed that this study did not involve animal subjects or tissue. Conflicts of interest: In compliance with the ICMJE uniform disclosure form, all authors declare the following: Payment/services info: All authors have declared that no financial support was received from any organization for the submitted work. Financial relationships: All authors have declared that they have no financial relationships at present or within the previous three years with any organizations that might have an interest in the submitted work. Other relationships: All authors have declared that there are no other relationships or activities that could appear to have influenced the submitted work. Author Contributions Concept and design: Atindra Narayan, Deepsekhar Das, Bahubali Shetti, Venkatesh Antalmarad, Anshum Choudhary, Kathyayini G. Thodupunoori, Sumit Grover Acquisition, analysis, or interpretation of data: Atindra Narayan, Deepsekhar Das, Bahubali Shetti, Venkatesh Antalmarad, Anshum Choudhary, Kathyayini G. Thodupunoori, Sumit Grover Drafting of the manuscript: Atindra Narayan, Deepsekhar Das, Bahubali Shetti, Venkatesh Antalmarad, Anshum Choudhary, Kathyayini G. Thodupunoori, Sumit Grover Critical review of the manuscript for important intellectual content: Atindra Narayan, Deepsekhar Das, Bahubali Shetti, Venkatesh Antalmarad, Anshum Choudhary, Kathyayini G. Thodupunoori, Sumit Grover References 1. Principles of Biomedical Ethics: marking its fortieth anniversary. Beauchamp T, Childress J. Am J Bioeth. 2019;19:9–12. doi: 10.1080/15265161.2019.1665402. [ DOI ] [ PubMed ] [ Google Scholar ] 2. Improvement of informed consent and the quality of consent documents. Jefford M, Moore R. Lancet Oncol. 2008;9:485–493. doi: 10.1016/S1470-2045(08)70128-1. [ DOI ] [ PubMed ] [ Google Scholar ] 3. ChatGPT: the future of discharge summaries? Patel SB, Lam K. Lancet Digit Health. 2023;5:0–8. doi: 10.1016/S2589-7500(23)00021-3. [ DOI ] [ PubMed ] [ Google Scholar ] 4. AI chatbots in answering questions related to ocular oncology: a comparative study between DeepSeek v3, ChatGPT-4o, and Gemini 2.0. Das D, Narayan A, Mishra V, Takia L, Grover S, Bharati A, Mb S. Cureus. 2025;17:0. doi: 10.7759/cureus.90773. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Evaluating AI-generated informed consent documents in oral surgery: a comparative study of ChatGPT-4, Bard Gemini advanced, and human-written consents. Vaira LA, Lechien JR, Maniaci A, et al. J Craniomaxillofac Surg. 2025;53:18–23. doi: 10.1016/j.jcms.2024.10.002. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Artificial intelligence-generated informed patient consent in various ophthalmological procedures: a comparative study of correctness, completeness, readability, and real-word application between Deepseek and Chatgpt 4o. Das D, Chawla B, Lomi N, Tomar P, Herle A, Joshi S. Indian J Ophthalmol. 2025;73:1466–1470. doi: 10.4103/IJO.IJO_1126_25. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Deep learning for retinal image quality assessment of optic nerve head disorders. Chan EJ, Najjar RP, Tang Z, Milea D. Asia Pac J Ophthalmol (Phila) 2021;10:282–288. doi: 10.1097/APO.0000000000000404. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. Gulshan V, Peng L, Coram M, et al. JAMA. 2016;316:2402–2410. doi: 10.1001/jama.2016.17216. [ DOI ] [ PubMed ] [ Google Scholar ] 9. Diagnostic performance of the offline Medios Artificial Intelligence for glaucoma detection in a rural tele-ophthalmology setting. Upadhyaya S, Rao DP, Kavitha S, et al. Ophthalmol Glaucoma. 2025;8:28–36. doi: 10.1016/j.ogla.2024.09.002. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Diagnostic accuracy of automated diabetic retinopathy image assessment softwares: IDx-DR and Medios Artificial Intelligence. Grzybowski A, Rao DP, Brona P, Negiloni K, Krzywicki T, Savoy FM. Ophthalmic Res. 2023;66:1286–1292. doi: 10.1159/000534098. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Simple, Mobile-based Artificial Intelligence Algorithm in the detection of Diabetic Retinopathy (SMART) study. Sosale B, Aravind SR, Murthy H, Narayana S, Sharma U, Gowda SG, Naveenam M. BMJ Open Diabetes Res Care. 2020;8 doi: 10.1136/bmjdrc-2019-000892. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. AI in the shadows: unveiling the strengths and blind spots of Medios AI retinal screening in cancer care. Das D, Chawla B, Lomi N, Grover S, Narayan A. Cureus. 2025;17:0. doi: 10.7759/cureus.99002. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. A survey of multilingual large language models. Qin L, Chen Q, Zhou Y, et al. Patterns (N Y) 2025;6:101118. doi: 10.1016/j.patter.2024.101118. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Articles from Cureus are provided here courtesy of Cureus Inc. ACTIONS View on publisher site PDF (140.4 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top