Why Open-Source LLMs are a Game Changer for Defence:

Apr 26

Industry 4.0 is well underway. Billions of dollars are being poured into open source Artificial Intelligence (AI) through the development of Large Language Models (LLM), which are freely available for organisations to download and use.

So why is this such a big deal for Defence and how do they address concerns about data security, reliability, and readiness in critical operations?

The Problem with the First Wave of Generative AI (GenAI)

Whilst many recognise the benefits that generative AI can deliver, there are a number of potential showstoppers when using them for Government & Defence applications:

You may have little control of your data once it’s been shared. It could be used to train the model or be compromised by a third party.
Intellectual Property (IP) defence means they can be a black box - there is little to no understanding about how they arrived at their response, making validation and assurance difficult.
Models may be updated, constrained or modified without warning, satisfying the business objectives of the supplier, for whom Defence may actually be a small customer.
Providers may change their policies, based on usage or who they wish to supply - resulting in the service stopping.
There is no guaranteed availability - you may find the service is stopped at a critical moment in your mission.

So how can we exploit the power of generative AI, while eliminating or mitigating these issues? The answer could be in Open Source LLMs.

The Advantages of the Second Wave of GenAI

For Government and Defence, the second wave of Generative AI, is much more compelling. The second wave has triggered both a massively competitive market, and delivered open source equivalents to first wave black boxes.

Open source can be likened to a collaborative workspace where the code is fully visible and accessible. This transparency means that experts from both within and outside government organisations can scrutinise and improve the software, leading to enhanced security and innovation. This collaborative approach is vital for addressing the unique challenges faced by the defence and government sectors.

Uncompromising Security

Security of data for Defence is obviously paramount. Whilst even black box LLMs can be hosted ‘on premises’, open-source LLMs offer other benefits:

Transparency: The open nature of these systems allows for thorough inspections and audits by experts, ensuring robust security measures.
Customisation: These models can be tailored to specific security requirements, providing a bespoke solution that proprietary models may not offer.
Control: Utilising open-source models ensures that data remains under the control of the agency, mitigating risks associated with third-party data management.

Reliability in Critical Operations

Self-hosting means that you control the availability of the technology in mission-critical operations. Open-source LLMs can also provide:

Broader Support Base: open source products are developed and maintained by a broad community of developers continuously enhancing the software, addressing bugs and potential vulnerabilities. These updates are more regular and responsive than in their black box counterparts.
Flexibility: The ability to modify the source code enables agencies to rapidly adapt the software to specific operational needs and conditions.
Interoperability: Open-source models are designed to integrate seamlessly with various platforms and software, reducing potential integration issues.

Ensuring Availability and Scalability

The capability to scale and remain available during unexpected demands is crucial, particularly during crises. Open-source LLMs are well-equipped for such challenges:

Community-Driven Improvements: There is an ongoing effort within the community to optimise performance and scalability, ensuring the software can handle demanding situations. The code base is usually more regularly refactored.
Cost-Effectiveness: The absence of licensing fees allows agencies to allocate more resources to infrastructure improvements, enhancing service availability.
Rapid Deployment: These models can be deployed swiftly across multiple departments and agencies, ensuring they are ready when needed.

Defence and government sectors are facing significant challenges when exploiting generative AI, and embracing open-source LLMs could be transformative. The move towards open-source technology can lead to improved security, heightened reliability, and unmatched operational readiness.

So why did I start with Industry 4.0? Isn’t that about manufacturing? Our mission is to unlock the potential of Defence’s two most important assets; people and data. In an unsettled decade, where the Defence insurance policy may well be called upon, we must produce competent warfighters at a rate never before required. We believe teaming current training systems and their processes with AI and machine learning (ML) is the only way to do that, and that open-source LLMs (deployed appropriately) provide a timely accelerant to the sovereign talent manufacturing line.