OPENAI, Baidu, Baichuan Intelligence... Big models are firing on all cylinders in the medical field

Large model companies, led by OpenAI, are fully exploring the medical market.

Not long ago, the soul of OpenAI, founder and CEO Sam Altman (Sam Altman) founded a medical AI company called Thrive AI Health (hereinafter referred to as Thrive), hoping to change the behavior of chronic disease patients through AI coaches. Not surprisingly, it will use OpenAI's generative AI technology.

According to public data, OpenAI has been in full force in the medical market since 2024. Not only OpenAI, but also domestic large model enterprises are quickly following up. According to media statistics, among the 120 financing events related to large models in the world in recent years, large models + medical care accounted for the first - up to 15% of large model financing enterprises are related to medical care.

The medical field is becoming the "jewel in the crown of the big model."

OpenAI,

All guns blazing in the medical field

In addition to Sam Altman, Thrive's other co-founder Arianna Huffington is also the founder of The Huffington Post and Thrive Global, a behavioral-change technology company. In addition, DeCarlos Love, a former Google executive, will also join the new company as CEO.

The company in the pipeline will also have deep partnerships with leading academic institutions and medical centers, including the Stanford Medical Research Institute, the Rockefeller Neuroscience Institute at West Virginia University, and the Walton School of Medicine.

It is more appropriate to describe it as the "rich second generation" born with a golden key.

Thrive will develop an "AI health coach" platform that will enable people with chronic diseases to receive professional guidance on key health issues in their daily lives, such as sleep, nutrition, fitness, stress management, and social interaction, with the aim of reducing the incidence of chronic diseases globally through behavioral change.

In the United States alone, 129 million people will have at least one chronic disease, such as cardiovascular disease, depression or diabetes, in 2023. These chronic diseases can actually be ameliorated through changes in daily behavior.

Thrive's AI health coach will leverage Thrive Global's legacy in healthcare and introduce OpenAI's latest advancements in AI, including long-term memory enhancements and domain-specific behavioral coaching models. This generative AI will be trained on the latest peer-reviewed scientific, biostatistical, laboratory and other medical data, combined with data on the user's personal preferences.

However, the founders say Thrive will not venture into medically-assisted diagnostics and will instead focus on providing health advice. It may also be to avoid sensitive medical data privacy and regulatory issues.

At the same time, the exact form of Thrive's product is not yet determined, it may be launched in the form of an app, it may be offered in various modes, and it may even be integrated into the workplace through an app like Microsoft Teams.

This isn't OpenAI's first foray into healthcare. A little statistics are not difficult to find that since entering 2024, OpenAI is in full swing in the medical field.

The biggest news is that in April of this year, well-known pharmaceutical company Moderna announced a partnership with OpenAI to deploy ChatGPT Enterprise across Moderna's entire company, and set a goal of making everyone with access to digital solutions fully proficient in using generated AI within six months.

In fact, back in early 2023, Moderna built an internal AI chatbot, mChat, based on OpenAI's API, and it was used by 80% of its employees and received positive reviews. After a comprehensive evaluation, Moderna finally decided to introduce ChatGPT Enterprise. Just two months after introducing the technology, 40 percent of Moderna's active users have created personalized chatbots, with a cumulative number of 750 and an average of 120 conversations per user per week.

Currently, with the support of OpenAI, Moderna has introduced generative AI in many everyday scenarios within the company.

For example, Moderna has used generative AI as a data analytics assistant for clinical research teams to review and analyze the potential of clinical data, and to integrate and visualize these large data sets to help enhance clinical judgment and decision making. For example, AI can assist in input detailed review, prioritize safety and optimize scenarios such as vaccines before further development in late-stage clinical trials.

Its regulatory team is also a big fan of generative AI. With the help of generative AI, regulators can quickly obtain clear and readable contract summaries, and in addition, generative AI can help employees quickly obtain internal policies, without the need to find a needle in a haystack as in the past, greatly improving work efficiency.

ChatGPT Enterprise is also being leveraged by Moderna's PR team. They built a generative AI chatbot to help create powerpoint presentations for quarterly earnings calls, and another to help translate rusty biotech jargon into approachable language to communicate with investors.

Bloom on all sides,

Generative AI is landing in multiple medical scenarios

In addition to Moderna, more and more healthcare companies are also working with OpenAI and introducing generative AI in many scenarios. Among them, AI health assistants have become one of the areas where generative AI is best at medical participation. In addition to Thrive, both Healthify and Whoop have partnered with OpenAI to leverage OpenAI's generative AI capabilities to build AI health assistants that analyze user data and assist humans in providing guidance on everything from diet to behavior.

At the same time, using generative AI's great advantages in natural language processing as an internal knowledge base or auxiliary word processing is also one of the current hot directions.

For example, Oscar Health became the first healthcare company to sign a business partnership agreement with OpenAI in 2024. With OpenAI's technology, Oscar Health built an AI assistant that effectively tracks claims and automatically answers questions about patient reimbursements. Not only does it reduce the time it takes to process claims by 50%, it also guarantees accuracy equal to or better than that of a human agent.

Lifespa will use generative AI to transform surgical consent forms from dense medical legalese into text that is easier for all patients to read, enabling more patients to understand the content of surgical consent forms. Summer Health, a pediatric medical practice, used generative AI to help doctors organize their office records, drastically reducing the average time doctors spent processing paperwork per office visit from 10 minutes to two minutes.

In the emerging field of "AI+ clinical", the advantages of generative AI are also very obvious.

At present, AI+ clinical in foreign countries has become a hot investment, only 2023 in the United States has about 200 AI-designed clinical trials. Generative AI can read and analyze vast amounts of disease phenotype and genetic information to make clinical indication selection, clinical endpoint optimization, and eligible subject identification. In addition, generative AI can also analyze electronic medical records, which can help pharmaceutical companies find which hospitals have more patients who are more eligible for enrollment requirements, and play a clinical monitoring role.

In the past, the challenge of enrolling in clinical trials has been to match and identify the right patients. Manual evaluation of a large number of patient records is not practical, which means that most clinical trials are done by patients close to the trial site, which can be subject to selection bias.

Paradigm uses OpenAI's GPT-4 model to extract and analyze cases, enabling it to match patients who are best matched in clinical trials. Practice has proved that the generative AI platform can evaluate hundreds of patients per minute, which is significantly more efficient than the previous manual assessment of only 50 patients per day, and the accuracy is also 10% higher than the manual, showing the broad prospect of generative AI+ clinical.

At the same time, OpenAI's generative AI also shows great advantages in assisting doctors to provide more personalized treatment plans for patients.

In June, for example, Color Health, which has been working with the American Cancer Society to help health plans and employers control cancer, announced it had chosen OpenAI as its AI solution provider. The company plans to use GPT-4o to integrate medical data with clinical knowledge to create a generative AI called Copilot to create personalized, comprehensive treatment plans for cancer patients.

Color Health's Copilot has already shown benefits in clinical trials. Copilot helps healthcare organizations identify four times more cases where lab, imaging, or biopsy and pathology results are missing. And it could dramatically reduce the time it takes clinicians to analyze patient records and identify gaps from weeks to an average of five minutes.

In the development of generative AI+ medical application scenarios, the domestic follow-up speed is very fast. Taking Baidu, which is directly related to OpenAI, as an example, in September 2023, Baidu Health released the first industrial-grade medical model - the spiritual medical model. After nearly a year of development, the model has now formed a three-layer technical architecture of MOE+ terminal components + agents, and has been widely used in popular science content, Internet hospital, smart hospital, smart diagnosis and treatment, and open platform products.

At the recent Baidu Health Industry Ecological Conference, Baidu Health focused on all aspects of the patient's medical journey, combined with the needs of doctors and patients, and issued four medical large model applications and one open platform, which effectively promoted the landing of this innovative technology.

The four medical large model applications are: vertical large model application AI health assistant for ordinary users, providing users with authoritative and convenient health science popularization information and medical consulting services; Copilot provides quality control and decision-making support in every link of Internet medical treatment, and then helps doctors to fully improve the effectiveness of online medical treatment in Internet hospital practice; AI intelligent clinic for hospitals, used to solve the conflicts of patients with wrong numbers, incorrect numbers and medical resources; And CDSS+LLM, which mainly focuses on the time-consuming scenarios commonly used by doctors such as case generation, patient briefing, intelligent question and answer.

In addition, Baidu Health also launched the open platform of spiritual medicine, and said that it will provide partners with medical service API and 10 million Token free quota on the open platform of spiritual medicine, as well as provide free co-construction support for 20 standardized scenarios. Jointly explore the intelligent service of the hospital, the efficiency of enterprise operation, post-diagnosis patient management, high-quality science popularization and creation, and the upgrade of Internet diagnosis and treatment, and finally promote the AI inclusive, and then reshape the intelligent and empowered medical service system.

In general, the development of generative AI giants such as OpenAI and Baidu in the medical scene is very representative, and will also provide a reference for other large model enterprises to enter the medical field.

Domestic generative AI unicorn,

Also at All-in Medical

In addition to Dachang, the unicorns in the field of generative AI are also running on the road of AI+ medical. Just recently, the well-known domestic large model unicorn Baichuan Intelligence announced the completion of A round of financing, the financing amount of up to 5 billion yuan, may become the largest financing in the domestic AI field in 2024.

Considering that Baichuan Intelligence was only founded about a year and a half ago, this pace of development is also the best portrayal of the importance of generative AI.

It is remarkable that Baichuan Intelligent is also a long-term focus on the medical field, and regards medical treatment as a dream domestic head large model enterprise.

Its founder, Wang Xiaochuan, who is also the founder of Sogou, chose the direction of biology to do interdisciplinary research when studying in the computer department of Tsinghua University, and has a strong feelings for life science. Since 2016, Wang Xiaochuan and his enterprises have invested in a number of medical technology enterprises such as Eagle Eye Technology, Deer Medical Museum, Encyclopedia Medicine, Feather medicine Kale, and Spring Rain Doctor.

In 2021, when Sogou was incorporated into Tencent, Wang Xiaochuan said in an open letter, "In the next 20 years, if you can make a contribution to the development of life health and medicine, and make a little contribution to public health, life will be more meaningful." Subsequently, Wang Xiaochuan founded Baichuan Intelligence in 2023, and firmly determined that "medical treatment is the pearl in the crown of the large model."

He believes that the mastery of knowledge, the mastery of experience, the multimodal capacity of the large model, the ability to remember and think, the ability of the large model to reduce hallucinations, to communicate empathy, all of these can be used in the medical field. The better the modeling ability, the higher the medical level, and the upper limit of this process is extremely high.

Because of this, since the date of its establishment, Baichuan Intelligent has written "health" into the company's vision, and then become the only large model unicorn focused on the medical and health field in China.

This route of building AI doctors through large models is highly consistent with the views of Jeffrey Hinton, one of the recognized giants of artificial intelligence. Hinton has always seen healthcare as an area where AI can make a difference and bring about great good. He believes that AI is already comparable to excellent experts in the interpretation of many types of medical images, and believes that AI is better at integrating large amounts of patient data, and the era of AI doctors is coming.

The relevant person in charge of Baichuan intelligence said when communicating with the artery network that medical treatment is the industry with the highest data density, and there is no one. Medical guidelines are the compression of medical records, and disease diagnosis is based on the prediction of the guidelines.

"The basic principles of medical research and clinical services and the big model are actually in the same paradigm," "In terms of doctor training, physicians are well-informed, surgeons practice makes perfect, and it is also highly consistent with the big model and embodied intelligence training," he told the Artery Network.

The person in charge further said that the goal of Baichuan intelligence is to create "AI doctor". This will lead to a medical supply-side revolution: "The pain point of the medical service industry is" unlimited demand and limited supply ", Internet medical benefits less but different from the medical model, it can achieve a thousand-fold increase in supply through the creation of AI doctors/assistants, so that the quality of medical services is better, more accessible, and lower costs."

He described to us the realization path of AI medical treatment in Baichuan's intelligent ideal, and compared AI medical treatment with the unmanned driving grade, giving a grade of L0-L5.

In the view of Baicuan Intelligence, the current large model technology has the ability to achieve L3 level (LLM) AI medical applications, that is, AI can assist in recommending treatment plans under specific conditions and be confirmed by doctors. Baichuan Intelligent will also take the lead in the near future in the form of health consultants and doctor assistants landing health consultation, auxiliary diagnosis and treatment and other medical scenes.

In the next phase, Hyakawa Intelligence will launch an impact to level 4 (AGI). At this stage, the ideal AI doctor would