Chinese Researchers Develop Military AI Using Meta’s Open-Source Llama Model — ChatBIT Reportedly Achieves 90% of the Performance of OpenAI's GPT-4 LLM

Create an image depicting a futuristic laboratory setting in China, where scientists are actively working on developing advanced military AI technology. Capture the scene with detailed computer screen

Martin Kouyoumdjian |

Development of ChatBIT

Chinese researchers have made significant strides in military AI development with the creation of an advanced AI tool named ChatBIT. This tool was developed using Meta's open-source Llama 2 13B large language model (LLM). The project involved institutions linked with the People’s Liberation Army (PLA), including the Academy of Military Science (AMS).

Capabilities and Performance

ChatBIT is specifically tailored for dialogue and question-answering tasks within the military sector. It reportedly achieves about 90% of the performance of OpenAI's ChatGPT-4. While the researchers have provided these figures, they have not specified how they benchmarked the performance or whether the model has been fully integrated into operational use.

Training and Data

To hone ChatBIT's capabilities, the model was fine-tuned using a dataset comprising 100,000 military dialogue records. This dataset size is relatively small compared to other large language models, which typically train on datasets containing trillions of tokens.

Potential Applications

The researchers foresee a broad range of military applications for ChatBIT. These include intelligence analysis, strategic planning, simulation training, and command decision-making. The tool is poised to enhance various aspects of military operations, underscoring its potential significance in future military strategies.

Use of Western-Developed AI

This development is part of a wider trend where Chinese institutions utilize Western-developed AI models across different sectors, including military and domestic security applications. The Llama model, for instance, has been employed to improve intelligence policing and police decision-making, illustrating the versatility of these AI tools in various domains.

Meta's Policy and Enforcement

Meta’s terms of service explicitly prohibit the use of its models in military, warfare, nuclear industries, or applications governed by U.S. defense export controls. However, given their open-source nature, there are limited enforcement mechanisms available to Meta. The company has stated that any use of their models by the PLA is unauthorized and in breach of their acceptable use policy.

Broader Implications

The employment of open-source AI models by Chinese researchers presents challenges regarding the control over the dissemination and utilization of advanced AI technologies. Analysts highlight that collaborations between Chinese and Western scientists complicate efforts to restrict Chinese institutions' access to these technologies.

Institutional Involvement

The research teams involved in developing ChatBIT hail from prominent institutions, such as the AMS’s Military Science Information Research Center, the National Innovation Institute of Defense Technology, the Beijing Institute of Technology, and Minzu University. Other organizations like the Aviation Industry Corporation of China (AVIC) have also played roles in applying Llama 2 for military-related purposes.

Logics Technology Online Shop