Speaking of the hottest topic in the past two years, large AI models undoubtedly take the lead. As we all know, with the strong breakout of ChatGPT, a global AI craze has been ignited, and many companies both domestic and international have stepped on the gas to launch their own large model products. The continuous emergence of AI large model products has also staged a grand "battle of a hundred models."
With the successive launch of more and more AI large model products, the practical application of AI large models has also been put on the agenda, becoming a key focus for various industries. It is worth noting that in addition to general large models, the number of vertical large models targeting specific industries is also increasing day by day, with the medical field being a key area of attention. In fact, since 2023, many leading companies have been increasing their bets in the field of medical large models, which means that large models are gradually penetrating the medical industry.
Large models, a new accelerator for the medical industry?
In the past two years, the heat of AI large models has remained high, and various industries are actively embracing AI large models, hoping that they can bring new changes to the industry, and the medical industry is no exception. The wave of AI large models has already reached the medical industry, and a series of large model products and applications related to healthcare are emerging one after another.
For example, JD Health officially released the large model "Jingyi Qianxun" for the healthcare industry; Baidu officially released China's first "industrial-level" medical large model "Lingyi Large Model"; MedLink officially released its self-developed medical large language model MedGPT; Weining Health released the large model "WiNGPT" in the medical field. Behind the layout of medical large models by various players, there is also its own logic.
Advertisement
For companies, the introduction of AI large models helps to improve medical efficiency and reduce costs. AI large models are deep learning models, which means they have strong learning capabilities, capable of analyzing and processing massive amounts of data, and making inferences and predictions. Because of this, medical large models can analyze and organize medical data, and make corresponding diagnoses based on medical information, assisting doctors in diagnosis, which can to some extent promote the improvement of medical efficiency for healthcare workers. For example, medical large models can convert doctors' dictation into structured notes with conversational language, greatly saving the time doctors spend writing medical records and improving work efficiency.
For users, the launch of AI large model products helps to further enhance the patient's medical experience and provide comprehensive medical services. Unlike other needs, users' medical needs are long-term, but affected by various factors, patients often face a variety of problems, such as long waiting times, difficulty finding the appropriate department, etc. The various unexpected issues encountered during the medical process also affect the patient's medical experience. The emergence of medical large models is expected to solve the pain points that patients encounter during the medical process, bringing a more efficient and high-quality medical experience to patients. For example, in the pre-diagnosis phase, medical large models can recommend the appropriate department to the patient based on their medical needs, solving the problem of not knowing which department to register.
In addition, the emergence and application of AI large models can also accelerate the digital transformation of the medical industry to some extent. It is well known that digital transformation has long been one of the key directions for companies in various industries, and the medical industry is no exception. However, unlike other industries, the medical industry has great particularity and seriousness, coupled with different infrastructure construction situations in different regions and the coexistence of old and new systems, there is a significant "information silo" phenomenon in the medical industry, which is also an important reason for the slow pace of digital transformation in the medical industry. AI large models have a broad application prospect in the medical field, and as the combination of AI large models and medicine deepens, it may help to promote the development of the medical industry towards a more digital and intelligent direction.
Baidu, Tencent "water to the channel"In the wave of AI large model advancements, the frontrunners are undoubtedly the major internet companies. Companies like Baidu, Alibaba, and Tencent have successively launched their own AI large model products. For instance, Baidu has introduced the "Wenxin Yiyan" large model, Alibaba has presented the "Tongyi Qianwen" large model, and Tencent has unveiled the "Hunyuan" large model. As the medical industry is considered one of the best fields for the application of AI large models, it has also sparked a trend of large model enthusiasm, with many internet giants rolling out their own medical large model products. For example, Baidu has released the "Lingyi" large model, and Tencent has launched the "Tencent Medical Large Model." The ability of internet giants to be the first to introduce large model products aimed at the medical industry is closely related to their long-term accumulation.
On one hand, Baidu and Tencent have already launched their self-developed general large model products and have experience in developing large model products. This experience can provide references and insights for the launch of their specialized medical large model products. For a long time, internet giants like Baidu and Tencent have paid special attention to the field of artificial intelligence and have continuously deepened their efforts in this area, accumulating profound technical strength. All of this has laid a solid foundation for the launch of their AI large model products and has also promoted the emergence of their vertical large model products.
For example, internet giants like Baidu and Tencent have introduced professional medical data on the basis of their self-developed general large models, trained and fine-tuned the large models, and ultimately launched specialized large model products for the medical industry. Specifically, Tencent's medical large model is developed based on Tencent's full-link self-developed Hunyuan large model; the technical foundation of Baidu's "Lingyi" large model also comes from the previously mentioned Wenxin Yiyan large model.
On the other hand, internet giants like Baidu and Tencent have laid out in the field of medical health and have accumulated industry data. Internet giants have been actively expanding their business territories, with the medical health field being one of the key areas they focus on. Taking Baidu as an example, Baidu has not only launched the AI medical brand "Lingyi Zhihui," which is deeply involved in the medical field, but also created a one-stop health management platform "Baidu Health" that can provide health science popularization, online consultation, and other services.
Because of this, internet giants like Baidu and Tencent have accumulated a large amount of medical data, making it possible for them to launch professional medical large model products. It is understood that during the model training process, Baidu's Health Care Group (HCG) has successively invested in its own accumulated data, including over 10 million high-quality medical Q&A data, over 20 million multilingual medical professional knowledge, over 200 million daily medical search data from users, and over 500 million authoritative health science content. Similarly, Tencent's medical large model has incorporated a knowledge graph and medical literature with more than 2.85 million medical entities, 12.5 million medical relationships, and over 98% medical knowledge.
"Micropulse and Weining Health 'Rise with the Trend'"
The heat of medical large models is high, and in addition to the frequent layout of internet giants, the participation of players in the industry is naturally indispensable. Internet medical companies are also continuously making moves. Specifically, the full-course management platform Micropulse has officially released its self-developed large language model application in the field of health management—CareGPT; Weining Health has officially launched the large model in the medical field—WiNGPT. Behind the active embrace of AI large models by internet medical companies, there is also a reason.
Firstly, Micropulse and Weining Health have rich medical industry experience and a vast amount of medical data, which can lay the foundation for training medical large models. The importance of data for AI large models can be imagined. The ability of AI large models to continuously evolve and improve is inseparable from the support of data. Especially for industry large models, industry data is particularly important. Both Micropulse and Weining Health have been deeply involved in the medical health field for a long time and have professional medical knowledge bases. Thanks to this, Micropulse and Weining Health have accumulated a vast amount of high-quality medical data. These data are undoubtedly high-quality training data sets for large model products, which can help the two train more accurate and reliable medical large model products.
It is understood that during the training phase of Micropulse's CareGPT, it was based on evidence-based medicine, using the latest version of clinical medical guidelines, disease diagnosis and treatment models, expert consensus, and other medical text data exceeding 1 billion, as well as millions of Micropulse case management data, to form a medical health knowledge base for specialty disease management, and more than 100 case managers participated in the RLHF supervision and debugging training. Another set of data shows that by May 2023, the training data volume of WiNGPT has reached 9,720 items of drug knowledge, more than 7,200 items of disease knowledge, more than 2,800 items of examination and test knowledge, and more than 1,100 guide documents, with a total training Token count of 3.7 billion.
Secondly, the business of Micropulse and Weining Health has a natural fit with the medical large model, which is easier to achieve the landing and application of the medical large model, and is expected to promote the further development of their business. In addition to the launch of large model products, the subsequent landing application is also a very important part. Since Micropulse and Weining Health are internet medical companies, their business development is also centered around medical health, and the medical large model has a high degree of fit with their business, which is easier to achieve landing.For instance, CareGPT can analyze patient chat content, assisting patients in conducting preliminary health screenings independently, achieving intelligent auxiliary triage and guidance, and enhancing the efficiency of patient management. As the capabilities of medical large models continue to evolve and are gradually applied to specific scenarios, these models can also empower the business of internet medical companies, potentially bringing new growth opportunities.
The journey of medical large models is still long.
Thanks to the rapid development of artificial intelligence technology and its gradual application in medical scenarios, AI in healthcare is becoming a reality, and the emergence of large medical models is expected to inject new momentum into the industry's development. Currently, the wave of AI large models is sweeping in, and both internet giants and internet medical companies are eager not to miss this opportunity, riding the wave. However, it must be said that while medical large models hold opportunities, they also have thresholds.
Firstly, medical large models have a low error tolerance, and manufacturers need to continuously refine their products. Unlike other industries, the medical field has a strong sense of seriousness and professionalism, which means it cannot afford mistakes. This implies that the error tolerance for medical large model products is extremely low, posing higher demands on manufacturers. To avoid such situations, manufacturers focusing on this area must maintain a prudent attitude and continuously refine the products themselves to enhance the capabilities of the large models.
Secondly, medical data is highly private and difficult to process, making the training of medical large models challenging. As is well known, medical data often involves the patients themselves, thus possessing high privacy. However, the training of large model capabilities requires a substantial amount of data for support, making data acquisition quite difficult for manufacturers developing medical large models. Moreover, due to the lack of uniform data standards, data processing in the medical industry is also challenging, requiring more effort from medical large model manufacturers.
At present, the birth of medical large model products may bring new opportunities for the development of the medical industry, helping companies reduce costs and increase efficiency, and providing patients with a higher quality medical experience. However, since medical large models are still in the early stages of development, there are many challenges that need to be overcome by manufacturers. In summary, the launch of medical large model products does not mean the end; instead, it is a new starting point. Both internet giants and internet medical companies must not be complacent, as the competition continues.