Many graduate students weaken their thesis by confusing ๐ฎ๐ฐ๐๐ถ๐ผ๐ป ๐ฟ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต with ๐ฐ๐ฎ๐๐ฒ ๐๐๐๐ฑ๐โyet the two serve fundamentally different academic purposes.
๐๐ฐ๐๐ถ๐ผ๐ป ๐ฟ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต is initiated to solve an ๐ถ๐บ๐บ๐ฒ๐ฑ๐ถ๐ฎ๐๐ฒ ๐ฝ๐ฟ๐ผ๐ฏ๐น๐ฒ๐บ It focuses on ๐ถ๐บ๐ฝ๐น๐ฒ๐บ๐ฒ๐ป๐๐ถ๐ป๐ด ๐๐ผ๐น๐๐๐ถ๐ผ๐ป๐, often within the ๐ณ๐ถ๐ฒ๐น๐ฑ ๐ผ๐ณ ๐ฒ๐ฑ๐๐ฐ๐ฎ๐๐ถ๐ผ๐ป, where researchers may also ๐ฎ๐ฐ๐ ๐ฎ๐ ๐ฝ๐ฎ๐ฟ๐๐ถ๐ฐ๐ถ๐ฝ๐ฎ๐ป๐๐ in the research process. This approach is practical, intervention-based, and solution-oriented.
๐๐ฎ๐๐ฒ ๐๐๐๐ฑ๐, by contrast, involves ๐ถ๐ป-๐ฑ๐ฒ๐ฝ๐๐ต ๐ฎ๐ป๐ฎ๐น๐๐๐ถ๐ of a ๐ฝ๐ฎ๐ฟ๐๐ถ๐ฐ๐๐น๐ฎ๐ฟ ๐ฒ๐๐ฒ๐ป๐ ๐ผ๐ฟ ๐ฐ๐ฎ๐๐ฒ ๐ผ๐๐ฒ๐ฟ ๐ฎ ๐น๐ผ๐ป๐ด ๐ฝ๐ฒ๐ฟ๐ถ๐ผ๐ฑ ๐ผ๐ณ ๐๐ถ๐บ๐ฒ. It emphasizes ๐ผ๐ฏ๐๐ฒ๐ฟ๐๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ฎ๐ป๐ฎ๐น๐๐๐ถ๐ป๐ด ๐ฎ ๐๐ถ๐๐๐ฎ๐๐ถ๐ผ๐ป, is ๐๐๐ฒ๐ฑ ๐ถ๐ป ๐บ๐ฎ๐ป๐ ๐ณ๐ถ๐ฒ๐น๐ฑ๐, and ๐ฑ๐ผ๐ฒ๐ ๐ป๐ผ๐ ๐ฝ๐ฟ๐ผ๐๐ถ๐ฑ๐ฒ ๐ฎ ๐๐ผ๐น๐๐๐ถ๐ผ๐ป ๐๐ผ ๐ฎ ๐ฝ๐ฟ๐ผ๐ฏ๐น๐ฒ๐บ. Researchers typically ๐ฑ๐ผ ๐ป๐ผ๐ ๐๐ฎ๐ธ๐ฒ ๐ฝ๐ฎ๐ฟ๐ in the research setting.
Misunderstanding this distinction leads to flawed methodology, weak research design, and inconsistent findingsโcommon issues in rejected proposals.
๐ฒ If you need thesis help, WhatsApp DocAdeson on: +14243487554
โป๏ธ find this useful? follow + like + repost + comment.
"If you like interesting snippets on all sorts of subjects relevant to academia, information, the world, highly recommended is @grip54 's collection:"
ย
La curation de contenus, la mรฉmoire partagรฉe d'une veille scientifique et sociรฉtale
Gilbert C FAURE's insight:
... designed to collect posts and informations I found and want to keep available but not relevant to the other topics I am curating on Scoop.it (on behalf of ASSIM):
because we have a long standing collaboration through a french speaking medical training program between Facultรฉ de Mรฉdecine de Nancy and WuDA, Wuhan university medical school and Zhongnan Hospital
Comment enseigner lโIA en santรฉ ? Une belle journรฉe autour du symposium organisรฉ par CAP Santรฉ numรฉrique - universitรฉ de Bordeaux et CAP IA - universitรฉ de Bordeaux pour รฉvoquer lโenseignement de lโIA en Santรฉ. Nous avons un vรฉritable dรฉfi ร relever pour la formation des รฉtudiants et des enseignants/formateurs! Une fresque gรฉnรฉrรฉe en direct par une artiste Sophie Bougrat, facilitatrice graphique, rรฉsume le symposium et le tout sans aucune IA gรฉnรฉrative!
Merci ร tous les intervenants Xavier Blanc Philippe Nauche Frederic ALEXANDRE Bonnemains Carole Gouvernance stratรฉgique de la Data IA Caroline Receveur Cedric Gil-Jardine Fleur Mougin Vianney JOUHET Laurent Beaumont Ingrid MONTEIL Christelle SOARES Laurent Simon Olivier Cousin Andy Smith Jean Benoit Corcuff et lโรฉquipe organisatrice en particulier Ines Hizebry Camille Bachellerie Hugo Corvaisier
"The adoption of artificial intelligence (AI)-powered tools is accelerating rapidly across all layers of healthcare systems. Predictive models, decision support tools and generative tools have entered clinical environments, and large language models are increasingly being used by the general public to seek medical information and advice. Yet evidence that AI tools create value for patients, providers or health systems remains scarce."
"Without a clear connection between claims and evidence, medical AI risks being adopted faster than its real value can be understood."
There are numerous examples of transformational technologies & products introducing substantial harms before broad based benefits. (or before we could mitigate harms). We should avoid adopting faster than we can adapt.
๐จ 561 pages c'est beaucoup Peut-รชtre un peu trop pour un jeudi matin ๐
Alors avec mon collรจgue Claude, nous vous avons prรฉparรฉ une page de synthรจse interactive de ma thรจse : son fil conducteur, ses trois contributions (pragmatisme, protocole F-T-T, modรจle H-O-T), ses points saillants, le tout en quelques minutes de lecture.
Pour ceux qui veulent aller ร l'essentiel avant (รฉventuellement) de plonger dans le document complet, c'est par ici : https://lnkd.in/ggYwvrWH
(cc Nicolas MOINET Christian Marcon Audrey Knauf Stephane Goria Philippe clerc et Olivier Le Deuff)
This work brought together collaborators across institutions to explore how medical students engage with blended learning, self-regulated learning, and digital resources in anatomy education. Our findings reinforce the growing shift toward flexible, student-centered learning environments, with strong emphasis on blended and online modalities, as well as the critical role of self-regulated learning in shaping future physicians.
I would like to sincerely thank all co-authors for their collaboration and shared vision in advancing medical education research. It was truly a pleasure working together across institutions and contexts.
Grateful for this meaningful collaboration and looking forward to building on this work in future projects. #MedicalEducation #AnatomyEducation #BlendedLearning #GMU #Research #HigherEducation
Both physicians and the public demand medical AI to outperform human clinicians, at accuracy levels most current systems cannot yet reach.
1๏ธโฃ A Swedish survey of 223 physicians and 155 adults measured the minimum AI accuracy they would accept across three clinical scenarios.
2๏ธโฃ Both groups required AI to achieve higher sensitivity than human clinicians in all three scenarios tested.
3๏ธโฃ Physicians required AI to catch 11 more chest pain emergencies per 100 than the human nurse baseline; the public demanded 16 more.
4๏ธโฃ Both groups set a median specificity target of 50% for AI in triage, well above the 30-34% human nurse benchmark.
5๏ธโฃ For electrocardiogram interpretation, where human specificity was already 99%, both groups set the same standard for AI.
6๏ธโฃ Across all scenarios, 74-91% of respondents demanded stricter accuracy from AI than they would accept from a human clinician.
7๏ธโฃ The public was polarised on specificity: many demanded either 100% or 0% referral rates, both impractical for real clinical tools.
8๏ธโฃ 72% of physicians and 53% of the public had tried AI chatbots; 8.5% of physicians had used chatbot responses in real clinical decisions.
9๏ธโฃ Most respondents reported moderate trust in AI chatbots, matching physicians' trust in established electrocardiogram interpretation software.
๐ The thresholds both groups demand exceed most existing AI diagnostic systems, revealing a wide gap between expectation and real-world performance.
โ๐ป Rasmus Arvidsson, Jonathan Widen, Lina Al-Naasan, Ronny Kent K Gunnarsson, Peter Nymberg, Charlotte Blease, PhD, Anna Moberg, Par-Daniel Sundvall, Carl Wikberg, David Sundemo. Acceptable accuracy for medical AI: a survey of physicians and the general population in Sweden. BMJ Health & Care Informatics. 2026. DOI: 10.1136/bmjhci-2025-101899 | Open Access
๐จ 124 Papers. Clinical AI Models Built on Data of Unknown Origin
A new analysis linked more than 100 peer-reviewed studies to two widely used stroke and diabetes datasets with unknown origins and data patterns inconsistent with real patients. Some downstream models may already have reached clinical or public-facing settings.
Open data sharing is critical for AI progress. But in clinical AI, openness without provenance is not transparency.
Three points matter for implementation:
โข Dataset provenance is part of model validity If the origin, collection process, and population are unclear, performance metrics are not interpretable.
โข Robust dataset evaluation should be standard Basic checks (missingness patterns, value distributions, duplication) can already flag non-credible data.
โข External validation is not optional Models should be tested across independent external datasets.
๐ What should be the minimum standard before a clinical prediction model is considered deployable?| 40ย commentaires sur LinkedIn
Most researchers are using AI tools for literature reviews the wrong way.
They ask AI to โfind papersโ and hope for the best.
That is not a literature review strategy. That is search outsourcing.
Used properly, AI can save time, improve structure, and help you think more clearly.
But it should support your judgment, not replace it.
Here are practical tips I give research students:
1. Start with your question, not the tool A vague research question creates vague results. Define your topic, population, variables, or context first.
2. Use AI for search expansion Ask AI for synonyms, related terms, alternate spellings, and discipline-specific keywords. This improves database searching.
3. Use AI to screen faster Paste abstracts and ask for relevance against your inclusion criteria. This helps with first-pass screening.
4. Use AI to compare studies Ask it to summarise differences in methods, sample sizes, findings, and limitations across papers.
5. Use AI to identify patterns Good reviews are not summaries. Ask: What themes repeat? Where do studies disagree? What populations are ignored? What methods dominate?
6. Verify every citation Never trust references blindly. Cross-check authors, journal, DOI, and publication year.
7. Use AI for structure, not authorship AI can help organise themes and draft outlines, but your interpretation must lead the review.
8. Keep a decision trail Document search terms, databases, inclusion criteria, and why papers were included or excluded.
Use AI as an assistant, not as a scholar.
PS: What AI tool has actually helped your literature review most? Share in the comments
REPOST to help others.
Follow Dr Priya Singh, Founder Research Made Clear for more insights
For research tutorials and AI tool guides, subscribe to my YT channel: https://lnkd.in/e8zWuWV2 | 24 comments on LinkedIn
Une รฉtude du Imperial College of London et de Internet Archive rรฉvรจle que plus de la moitiรฉ des sites web est gรฉnรฉrรฉ en partie ou totalement par lโIA (52,9%), en forte augmentation. #GenAI https://lnkd.in/eac76_aG | 12 comments on LinkedIn
โ๏ธ๐Dear PhD Scholar:Starting your first research article can feel overwhelmingโฆ but with the right steps, it becomes simple and structured. Hereโs your easy roadmap ๐
๐น 1. Choose the Right Topic
Pick a clear, focused, and interesting research problem.
Your topic should answer a question or solve a gap.
๐น 2. Conduct a Literature Review
Read existing studies to understand whatโs already doneโand where your research fits in.
๐น 3. Define Your Research Objective
What exactly are you trying to find?
Be specific and concise.
๐น 4. Design Your Methodology
Decide how you will collect and analyze data (qualitative, quantitative, or mixed methods).
๐น 5. Collect & Analyze Data
Follow your method carefully and ensure your data is accurate and reliable.
๐น 6. Write the Structure
A standard research article includes:
Abstract
Introduction
Literature Review
Methodology
Results
Discussion
Conclusion
๐น 7. Interpret Your Findings
Explain what your results mean and how they contribute to existing knowledge.
๐น 8. Cite Properly
Always give credit using the required citation style (APA, MLA, etc.).
๐น 9. Edit & Proofread
Refine your writing for clarity, grammar, and flow.
๐น 10. Choose the Right Journal & Submit
Select a suitable journal and follow its guidelines carefully.
You donโt need to be perfect to startโฆ you just need to start to become perfect.
๐Happy Researching & Best of Luck, Future Scholars! ๐
More about paper mills and authorship for sale in today's Science issue. Need one more paper on your CV? Price depends... from $57 to $5600. Of course, this paper has a good chance of being entirely fabricated (fake) but who cares?
If you think this is a minor problem: - 18,710 advertisements related to authorship for sale were traced to Russia, Ukraine, Uzbekistan, India... (those were mostly written in English and do not count paper mills in other countries such as China) - A study published in January in The BMJ found that nearly 10% of 2.6 million cancer-research papers published from 2019 to โ24 seem to be paper mills products. That's... 250,000 papers. And this crap is used to train AI.
Publisher should aggressively fight this problem. But they won't unless it hurts their business... and I am not only talking of predatory publishers. Too many problems with many journals handled by "respectable" publishers with comfortable margins.
Link to the Science paper in comments. | 13 comments on LinkedIn
Quโest-ce quโun algorithme est capable de vous montrer, ร vous ou ร vos proches ? Ouest-France a construit un bras robotique pour scroller sur TikTok en continu afin de le savoir. Pendant 100 heures, la machine a visionnรฉ des milliers de vidรฉos. Certaines donnent des conseils pour sโaffamer, dโautres enseignent comment faire un nลud coulant. Dโautres encore renvoient vers des contenus pรฉdocriminels sur Telegram. Enquรชte sur une mรฉcanique qui amplifie tout, y compris le pire.
#tiktok #algorithme #enquete
00:00 Introduction โ Marie, 15 ans 2:05 Un robot 4:24 L'algorithme de TikTok 5:25 Make-Up et cuisine 8:12 Dans la bulle mascu 10:09 les Tartariens 11:37 La faille dans le systรจme 16:30 Des vidรฉos contrevantes aux rรจgles de TikTok 16:28 Les consรฉquences humaines des algorithmes 20:34 Le compte sans nom 21:51 La rรฉponse de TikTok 22:10 : Ce qu'un algorithme est capable de vous montrer --------------------------------------------------------------------------- Retrouvez toute lโactualitรฉ sur :
๐ณ ๐คฉ Magnifique ! L'ONF (Office National des Forรชts) a publiรฉ une carte interactive avec une sรฉlection de 30 forรชts exceptionnelles ร dรฉcouvrir en France.
Les forestiรจres et les forestiers de l'ONF les racontent en images, en vidรฉo et avec passion.
Voici cinq choses ร savoir sur les forรชts franรงaises.
1. Le patrimoine forestier franรงais a connu une extension forte et continue depuis 150 ans et atteint dรฉsormais une surface de 17M d'hectares soit 31 % du territoire franรงais (source : IGN https://buff.ly/42BakSd); C'est la 4รจme forรชt europรฉenne aprรจs la Finlande, la Suรจde et l'Espagne - le niveau de boisement de la France est revenu ร celui du XVe siรจcle. Voir cette infographie de Jules Grandin et Clara DeAlberto https://buff.ly/3J46xWC
Mais les massifs sont jeunes, leur vitalitรฉ et leur rรฉsilience pose problรจme puisque 50% des arbres ont moins de 60 ans et 20% seulement ont plus d'un siรจcle.
2. La #Corse est la rรฉgion la plus boisรฉe de mรฉtropole - 63% et la forรชt d'Orlรฉans est la plus grande forรชt domaniale (appartenant ร l'Etat) en France mรฉtropolitaine.
3. Les chรชnes sont les arbres les plus rรฉpandus en mรฉtropole. Avec 190 essences d'arbres, notre forรชt mรฉtropolitaine compte prรจs de 75% de toutes les essences prรฉsentes en Europe. C'est fou !
4. Les forรชts dโOutre-mer reprรฉsentent prรจs de la moitiรฉ de la superficie forestiรจre, soit 8 millions dโhectares, et abritent la biodiversitรฉ la plus riche.
5. En mรฉtropole, 75% des forรชts sont privรฉes. Les 25% restants sont des forรชts domaniales ou des forรชts appartenant ร des collectivitรฉs ou รฉtablissements publics.
Mais globalement, nos forรชts ne sont pas en bon รฉtat. Elle pourrait cesser de stocker du carbone.
Les arbres interviennent ร plusieurs titres dans l'รฉquation climatique. Ils sont utiles pour l'attรฉnuation (en jouant un rรดle de stock de carbone) et pour l'adaptation (en jouant un rรดle de rรฉgulateur climatique, dans le cycle de l'eau, des sols et de la biodiversitรฉ).
Mais les forรชts franรงaises sont, comme les autres forรชts du monde, soumises ร une pression croissante. Elles absorbaient 30M de tonnes absorbรฉes en 2020, soit environ 7,5 % des รฉmissions nationales. Mais cโest deux fois moins quโil y a dix ans.
En terme de biodiversitรฉ en 2023, 17 % des oiseaux de forรชt, 7 % des mammifรจres, 8 % des reptiles et amphibiens, 12 % des papillons sont menacรฉs dโextinction (Source : IGN)
Les effets des sรฉcheresses se font sentir sur les รฉcosystรจmes. La tendance est ร la monoculture et les essences ne sont plus aussi rรฉsilientes qu'autrefois.
Alors pensons nos forรชts dans un souci de variรฉtรฉ et de diversitรฉ.
Les arbres tissent des liens entre les rรจgnes du vivant. Soyons clairs, nous ne pouvons pas nous en passer.
#foret #arbre #onf #france
๐ฅ Petite annonce ! Mon futur livre a besoin de votre soutien. Contactez moi si vous souhaitez m'aider ร prรฉparer la campagne => https://buff.ly/P06fQIn| 10ย commentaires sur LinkedIn
Une extension qui regroupe ChatGPT, Claude et Gemini.
Monica transforme votre navigateur en assistant IA : rรฉsumรฉ d'articles, aide ร la rรฉdaction, traduction, gรฉnรฉration d'images... sans changer d'onglet.
Je viens de dรฉcouvrir Remotion (https://www.remotion.dev) qui permet de crรฉer des vidรฉos en React. De plus, c'est complรจtement gratuit !
รa s'installe dans le terminal (vous tapez : npx create-video@latest) et puis aprรจs vous demandez ร votre IA prรฉfรฉrรฉe (vous me connaissez, je passe รฉvidemment par OpenClaw) et il vous crรฉe votre vidรฉo en quelques secondes.
Je viens de faire un essai avec une leรงon de grammaire trรจs simple sur la fonction sujet et voici le rรฉsultat. ๐ C'est du NotebookLM++ que vous pouvez modifier ร la volรฉe. Demandez juste ร votre bot de faire les changements voulus.
J'avais dรฉjร un prompt se dรฉclenchant tous les matins pour produire des exercices de grammaire et des dictรฉes (y compris au format MP3). J'ai maintenant les leรงons au format vidรฉo !
We finally have a way to measure AI harm in clinical practice.
We usually judge clinical AI the same way we test medical students: exams, scores, and knowledge benchmarks. But here's the issueโnone of that tells us one critical thing: ๐ How much harm can this AI actually cause?
๐ง๐ต๐ฎ๐'๐ ๐๐ต๐ฎ๐ ๐ก๐ข๐๐๐ฅ๐ ๐ฐ๐ต๐ฎ๐ป๐ด๐ฒ๐. Released just months ago by Stanford and Harvard researchers, NOHARM is the first framework designed to measure real clinical risk, not just performance. ________________________________________ ๐๐ฒ๐ฟ๐ฒ'๐ ๐๐ต๐ ๐ถ๐ ๐บ๐ฎ๐๐๐ฒ๐ฟ๐: ๐ฌ It introduces a common language for safety AI outputs are graded by severity of harm (mild โ severe), based on expert consensus across real clinical cases.
๐ It enables true benchmarking Over 30 AI models evaluated head-to-head using the same safety standardsโwith a public leaderboard for transparency.
โ ๏ธ It captures the errors that actually matter Not just wrong answers, but: ย โข Harmful recommendations (commission) ย โข Missed appropriate care (omission โ the majority of severe errors)
๐ It gives decision-ready metrics Including Case Harm Rate and Number Needed to Harmโtools clinicians and hospitals can actually use before adopting AI. ________________________________________ ๐ง๐ต๐ฒ ๐ธ๐ฒ๐ ๐ถ๐ป๐๐ถ๐ด๐ต๐? Traditional benchmarks only moderately correlate with real-world safety (r โ 0.6). Meaning: high scores โ safe care.
NOHARM isn't perfect! but it's the first real step toward accountability in clinical AI.
Before integrating AI into your practice, ask yourself: ๐๐ผ๐ ๐บ๐๐ฐ๐ต ๐ต๐ฎ๐ฟ๐บ ๐ฎ๐ฟ๐ฒ ๐๐ผ๐ ๐๐ถ๐น๐น๐ถ๐ป๐ด ๐๐ผ ๐๐ผ๐น๐ฒ๐ฟ๐ฎ๐๐ฒ? ________________________________________ ๐ Paper: arxiv.org/abs/2512.01241
Research Details: Validated across 100 real eConsult cases (10 specialties) with 12,747 expert annotations on 4,249 management options. 95.5% expert concordance on severity classifications.
N.B. MAST (Medical AI Superintelligence Test) is ARISE AI's comprehensive benchmarking platform for advanced clinical AI evaluation, with NOHARM serving as its core harm-assessment framework. Together on bench.arise-ai.org, they rank 30+ models using physician-validated metrics from 100 real cases across 10 specialties, focusing on safety (Case Harm Rate, NNH) beyond traditional accuracy benchmarks.
NEW STUDY๐งจHalf a million AI chats reveal emotional support and symptom queries spike at night, when access to traditional care is often limited.
What can Microsoft Copilot tell us from over HALF A MILLION (n = 617,827) de-identified health conversations? ย -- Nearly 1 in 5 chatbot conversations involved personal symptoms, conditions, or test interpretation
-- About 1 in 7 personal health queries were about someone else (child, parent, partner)
-- Symptom queries happened mostly on mobile (15.9%) vs desktop (6.9%), whereas health-related research support was mostly on desktop (16.9%) vs mobile (5.3%)
-- Nighttime spikes on emotional support queries (from 3.3% morning to 5.2% nighttime) and symptoms/health concerns (from 10.6% morning to 13.4% nighttime)
These nighttime spikes (when healthcare access is most limited) may amplify user risk, as highly actionable LLM outputs are more likely to be relied on without professional support. ย The authors did acknowledge safety RISKS in that mainstream AI models (ChatGPT, Google Gemini, Microsoft Copilot) may perform well on medical exams but NOT always on real-world emergency triage or decision-making, yet users keep turning to them for health advice and emotional coping. ย The authors seem to take these risks seriously and concluded:
โFrom a safety perspective, the personal health intents identified hereโsuch as symptom assessment, condition management and emotional well-beingโarguably define categories in which the consequences of conversational AI responses are greatest and where investment in response quality and safety measures should be concentrated.โ
Translation: we still have a long way to go and investment needs to be focused on the highest-risk use cases to make these models safe for public use, in a domain with near-zero tolerance for harmful errors. ย Study in Nature Health linked below ๐ | 18 comments on LinkedIn
Ifย youย want to build healthcare AI at scale,ย followย these 10ย Chineseย Healthcare AIย companies.ย
1. AQ AI Health App
ยท 30M monthly active users, 10M daily health consultationsย ยท 27M health questions answered in 2025, national health OS infrastructureย ยท 50%+ users from Tier 3+ cities, true population-scale deployment
2. XtalPi Inc.
ยท $59.9B drug pipeline deal with Harvard spinout (largest ever)ย ยท 300+ robotic workstations deployed globally, operating 24/7ย ยท AI + quantum physics + robotics integrated platform
3. JD.COM (JD Health)
ยท 50M+ users served by AI doctor agentsย ยท Full-cycle health management from prevention to chronic diseaseย ยท 2.2M+ patient encounters at partner hospitals
4. United Imaging Healthcare
ยท 20+ FDA-approved AI-powered medical devicesย ยท 92% diagnostic accuracy (beats industry SOTA by 10%+)ย ยท Hardware + AI + clinical workflows fully integrated
5. BGI Genomics
ยท Fully automated "dark lab" running 1+ year, zero human interventionย ยท Sample โ sequencing โ analysis โ storage, end-to-end automationย ยท Deepseek-R1 and Evo 2 AI models integrated into platform
6. MicroPort
ยท World's first LLM-autonomous surgery completed (Dec 2025)ย ยท AI-powered surgical planning, imaging fusion, automated operationsย ยท SAIL Award winner (Super AI Leader)
ยท 150+ hospitals deployed, 60+ clinical scenarios coveredย ยท 20+ specialized medical AI agents operationalย ยท Medical LLM WinGPT 3.0 with clinical reasoning capabilities
9. ็พๅนดๅคงๅฅๅบท
ยท 600 examination centers, 30M annual checkupsย ยท AI-driven preventive care and longevity medicine servicesย ยท Strategic partnership with Alibaba DAMO Academy for cancer screening
10. Kingmed Diagnostics
ยท 30PB multi-modal medical data accumulatedย ยท 15M+ monthly API calls for diagnostic intelligenceย ยท 60+ AI agents deployed, 70K monthly active physician users
These 10 companies didn't just build AI.
They built deployment infrastructure for 1.4 billion people.
Data infrastructure + full-stack systems + deployment speed.
The models are good enough.
The systems are what matter.
P.S.ย Which company's approach surprises you most?
Le Shen | Jian Ma | Daniel Wan | Kathy LIU | Jeremy (Sujie) cao | Jasmine Liang | ๅพๆต้ญ | Terry Zhao | ้ญ็ | Shen Luan
To get content containing either thought or leadership enter:
To get content containing both thought and leadership enter:
To get content containing the expression thought leadership enter:
You can enter several keywords and you can refine them whenever you want. Our suggestion engine uses more signals but entering a few keywords here will rapidly give you great content to curate.
Your new post is loading...
