Currently, generative artificial intelligence represented by large models is becoming increasingly widely used in combat, accelerating the transformation of winning mechanisms and empowering various stages of combat processes. The intelligent combat form characterized by data convergence ammunition, algorithm iteration tactics, and doubled computing power is emerging. Correctly understand the mechanism of the application of artificial intelligence models in combat. The large model itself does not store data, nor does it generate raw data. Big models, with their strong generalization ability, empower various links in data circulation, connect the efficient conversion link between the "global sensor optimal transmitter" on the battlefield, and deeply influence combat operations. One, second level response, speed wins. The ability to quickly construct and close the kill network chain is the key to current combat confrontation. The large model relies on powerful computing power and strong algorithms to provide rapid analysis and dynamic agile response capabilities in various aspects such as discovery, positioning, tracking, aiming, and strike evaluation, supporting and adapting to the "flash sale level" battlefield rhythm. In foreign military exercises, the ability of remote strike kill chain closure led by intelligent algorithms was continuously tested, achieving single kill chain closure at the machine's second level and cyclic kill chain closure with only a few seconds interval. Secondly, transparent perception enables global superiority over the enemy. The large model can significantly reduce information entropy by capturing small changes on the battlefield, helping commanders analyze intelligence "fog" and penetrate the "fog" of war, achieving a "panoramic" and "transparent" perception of the battlefield. The big model is based on multimodal data such as text, sound, and images obtained from various channels, and performs heterogeneous fusion to comprehensively extract key intelligence data such as enemy situation, personal situation, and battlefield environment. It can provide commanders with a visualized and interactive global battlefield situation, create a one-way strike time window for commanders, and thus grasp the initiative of combat. Thirdly, brain computer decision-making and collaborative intelligent control. The difficulty of building a multi domain kill chain covering land, sea, air, space, electricity, and network in the face of strong confrontation and fast-paced battlefield environment has exceeded the mental limit of commanders. The "human brain+machine brain" collaborative decision-making system empowered by large models has powerful logical analysis capabilities and can continuously cultivate and evolve machine intuition and reasoning abilities. It can quickly form multiple combat plans, formulate and distribute combat commands, coordinate and control combat actions based on human-computer interaction, and achieve the transformation from backend intelligence analysis tools to frontend decision-making "staff" for complex battlefield situations. Fully grasp the process of applying artificial intelligence models in combat. After simplification technology processing based on military requirements, the large model can form various specialized model systems that are suitable for combat scenarios, empowering the combat process by integrating command and control nodes, embedding weapon platforms, and operating equipment terminals. Firstly, empower the intelligence chain. Large models can integrate various current models, programs, or software, and through direct access to data streams on the battlefield, establish a second level intelligence processing process for discovering, using, and presenting battlefield situational targets; It can significantly expand the breadth of intelligence acquisition across command levels, improve battlefield perception accuracy, and accelerate the speed of situation updates. Secondly, empower the accusation chain. The large model can automatically arrange task instructions, generate interpretable reasoning sequences, and assist commanders in making decisions through human-computer interaction; Capable of decomposing task instructions, autonomous firepower planning, forming suggested solutions, calling operational plans for matching and optimization, providing feedback to business systems, or directly generating decisive suggested solutions, etc., to assist commanders in making efficient and rapid decisions in complex situations. Thirdly, empower the strike chain. The big model focuses on optimizing sensors and transmitters from multiple domains, focusing on parallel and synchronized combat operations to achieve optimal combat effectiveness. It can arrange kill chain sequences from cross domain resource pools based on the nature, quantity, and threat level of multi-dimensional targets on the battlefield, adapt to multi domain weapons and equipment, automatically bind target data, implement precise shooting online, and evaluate damage effects in real time. It effectively supports the rapid construction and closed operation of intelligent kill network chains. Fourthly, empower the security chain. Facing complex battlefield environments and massive support needs, based on large models, real-time monitoring of resource status information such as weapons and equipment, energy and ammunition, and materials and equipment is carried out. According to the combat process and battlefield situation, dynamic control of battlefield supply, maintenance, rescue, and transportation support actions is carried out, optimizing the allocation of combat resource supply flow, achieving comprehensive perception on the demand side and rapid response on the supply side, and improving the sensitivity and sustained support efficiency of the entire battlefield. Explore the path of practical application of artificial intelligence large-scale models in combat. In the future, the application of generative artificial intelligence in combat will experience a "surge", with vertical models developed based on general large-scale models and combined with combat scenarios, as well as various functional intelligent agents, becoming the "new track" and "high ground" for major countries in the world to compete. One is to develop a new theoretical system. The big model drives the reshaping of the battlefield form, and the future combat advantage belongs to the military that can deeply integrate algorithm innovation and theoretical innovation and put them into practice. Therefore, it is advisable to regard the empowerment of large models as an important "growth pole" for cultivating new domain and new quality capabilities, attach importance to military theoretical innovation, accelerate the exploration of fundamental, applied, and technical issues related to the use of large model empowerment in combat, and form a relatively complete theoretical system for unleashing the potential of large model empowerment; Regard the concept of combat as an important lever for consensus building and wisdom accumulation, design a new conceptual system that integrates joint and multi domain support, form a list of key capabilities and technical requirements, and accelerate the practice of intelligent warfare based on software definition, data-driven, algorithm led, and computing power support. The second is to build a backbone ecosystem. We can utilize the technological penetration and ecological integration capabilities of the general large model to integrate many general large models and fully build a backbone ecological architecture that empowers combat applications with large models. This includes various intelligent agents developed based on various combat styles and scenarios, such as intelligence agents, planning agents, logistics agents, etc., forming a full chain empowering combat ecological system supported by many general large model bases, vertical professional model frameworks, and various intelligent agent linkage. The third is to promote the integrated development of research, training, and application. In future warfare, it is important to pay attention to data collection and processing, strictly regulate each link of data processing, ensure that the large model retrieves data as needed and required, and try to maintain physical isolation or one-way flow from the original data storage system as much as possible; Emphasis should be placed on upgrading iterative algorithms. Large models can generate self evolving abilities through continuous training, and the training content, method, duration, and frequency are crucial for improving the quality and efficiency of empowering large models; Pay attention to the application model system, open up channels for the system to enter the fields of combat, training, support, and management, and improve the accuracy of large-scale model prediction and analysis, the strength of decision-making support, and the degree of effective empowerment in generalized use. (New Society)