On January 27, 2025, the release of the new open-source large language model (LLM), DeepSeek, caused a global sensation. Humans have been working on developing artificial intelligence (AI) capable of natural language processing (NLP) and LLM is the biggest breakthrough to date. Even before the emergence of DeepSeek, LLMs had already demonstrated their vast potential in the medical field. The advantage of DeepSeek lies in its ability to achieve performance comparable to (or perhaps even superior to) top-tier closed-source LLMs like OpenAI at an extremely low cost—a level of performance once considered exclusive to proprietary LLMs. Given the long-standing advantages of open-source LLMs in terms of flexibility, cost-effectiveness, and transparency, the success of DeepSeek seems to signal that the medical community is one step closer to the “AI era”. Thoracic surgery is a discipline that has long been intertwined with AI. Twenty years ago, computer-aided diagnosis (CAD) was already being used in the diagnosis and treatment of pulmonary nodules (1). In this article, we will discuss the opportunities and challenges that thoracic surgery will face in the “DeepSeek era”.

In the “pre-DeepSeek era”, AI had already permeated the entire process of diagnosis and treatment in thoracic surgery, spanning preoperative, intraoperative, and postoperative stages. With the advancement of machine learning (ML), deep learning (DL), and convolutional neural networks (CNNs), AI has become deeply involved in the interpretation of imaging results in thoracic surgery (2). In cases where historical imaging data is insufficient, AI has even demonstrated greater accuracy than human doctors (3). AI has also gradually been applied to the interpretation of pathological results. While insufficient tissue samples often limit the practical application of immunohistochemistry and genetic testing, AI can make precise judgments even with limited samples (4). Furthermore, AI appears to outperform intraoperative frozen sections in identifying spread through air spaces, which directly impacts the surgical approach for early-stage non-small cell lung cancer (NSCLC) (5).

Meanwhile, AI has gradually become involved in the surgical procedures of thoracic surgery. By preoperatively delineating tumor boundaries, AI has reduced the time required to locate tumors during surgery while ensuring complete resection margins (R0 resection). Recently, the results of JCOG0802 and CALGB140503 have significantly elevated the status of segmentectomy as a curative surgery for early-stage NSCLC. Accurately identifying intersegmental planes during segmentectomy is a challenging task, but augmented reality (AR) and virtual reality (VR) technologies can clearly expose anatomical structures, greatly simplifying this process (6,7).

These achievements, however, do not mean that thoracic surgery has entered the “AI era”. First, the high economic burden has long been a concern for the adoption of AI. Second, many AI models are trained on small-scale, highly specialized datasets, which can lead to “overfitting” and diminish their performance in real-world applications. Most importantly, AI still cannot directly participate in medical decision-making. Medical decision-making is a complex reasoning task, where the reasoning process is as important as the outcome. For doctors, relying on an AI that only provides answers without transparency is unimaginable. Some closed-source LLMs (e.g., OpenAI) can provide reasoning processes, but the “black box” nature of these models still lacks transparency and persuasiveness at a technical level.

DeepSeek is set to change this. DeepSeek can construct a complete chain of thought for any response, demonstrating powerful reasoning capabilities. As an open-source LLM, DeepSeek’s reasoning abilities can be widely validated at a technical level. Therefore, doctors can confidently refer to DeepSeek’s recommendations during decision-making without worrying about the transparency or safety of the advice’s origin. Additionally, DeepSeek can extensively access large medical public databases and stay updated with the latest medical advancements, significantly enhancing the reliability of its recommendations. While DeepSeek cannot replace human doctors in decision-making, it can make decisions more accurate and faster. This is particularly valuable in critical situations, such as clinical decision-making for patients with severe conditions like heart valve rupture or aortic dissection, where time is of the essence.

DeepSeek can also assist in thoracic surgical procedures. By accurately interpreting imaging results and performing preoperative 3D reconstruction of the surgical field’s anatomical structures, AI can help surgeons plan surgeries more precisely before the operation (8). Beyond preoperative planning, DeepSeek can effectively assist in managing postoperative complications. Thoracic surgeries, especially complex cardiac surgeries, still have a high incidence of postoperative complications. DeepSeek can comprehensively assess a patient’s overall condition and various test indicators, providing early warnings of complication risks even before surgery (9). Moreover, DeepSeek’s NLP capabilities enable it to answer frequently asked questions in real time, greatly aiding patient education before and after surgery (10).

Using AI to process experimental data in medical research is no longer news (11). However, DeepSeek can do much more. DeepSeek can efficiently read literature, helping researchers overcome language barriers when reading non-native language publications. Novelty is a critical metric in evaluating medical research. By extensively reviewing literature, DeepSeek can quickly identify potential research hotspots and present them to researchers. Similarly, DeepSeek can shorten the lengthy preliminary processes required for writing meta-analyses and systematic reviews, such as literature collection and screening, generating forest plots, and conducting heterogeneity analyses. In summary, DeepSeek allows researchers to focus on the “research” itself rather than being overwhelmed by massive amounts of data and text. Multidisciplinary collaboration is the future trend in medical research, and achieving this requires a robust public platform. The success of the BioChatter platform (12) demonstrates that the open, diverse, and inclusive community environment of open-source LLMs has significantly contributed to the advancement of medical research. We believe DeepSeek will perform even better in the future.

Acute respiratory distress syndrome (ARDS) continues to be a tough nut to crack in critical care, taking lives despite years of research and better ventilator strategies. It is defined by acute hypoxemia, bilateral infiltrates on chest imaging, and non-cardiogenic pulmonary edema, and it remains a heterogeneous condition with mortality rates stubbornly close to 40% (1). Its complexity—spanning diverse etiologies, inflammatory profiles, and therapeutic responses—demands innovative solutions beyond traditional paradigms. In recent years, artificial intelligence (AI) has started making waves in medicine, and one standout is DeepSeek, an advanced large language model (LLM), employs deep learning techniques like transformer architectures to efficiently process and generate human-like text fast and efficiently. Compared to other LLMs, it boasts superior computational efficiency—delivering faster inference and requiring fewer resources—making it ideal for real-time clinical applications and cost-effective deployment (2). In medicine, DeepSeek’s precision and adaptability have shown promise, particularly for tackling complex and life-threatening conditions in emergency and critical care setting, such as cardiac arrest (3,4) and ARDS. This editorial explores how DeepSeek could illuminate these dark corners of ARDS management, urging the thoracic and critical care communities to embrace and evaluate its promise. This paper will explore the topic from several aspects, including diagnosis, classification, precision ventilation, immune-modulating therapies, prognosis, and the road ahead.