Financial services are facing a new reality: generative AI is transforming the industry. While closed-source models offer powerful capabilities, they can raise concerns about transparency and control. That’s where open-source LLMs come in. Now more advanced than ever, these models provide a flexible and secure foundation for building AI-powered solutions. With open-source, financial institutions can embrace innovation while maintaining control over their data and algorithms.
Gaurav Sharma, Client Partner, Financial Services at Fractal, spoke with AIM to explore this shift and explain how open-source LLMs are used in financial services.
Sharma emphasized the critical importance of data protection when deploying open-source LLMs. Businesses must ensure strict adherence to data privacy laws and regulations. This includes minimizing data collection, filtering content appropriately, and deploying models locally whenever possible. Techniques like anonymization, serialization, and differential privacy are essential tools for safeguarding sensitive information.
He elaborated on a framework for managing data privacy: detect, treat, and rehydrate. Detection involves identifying potential risks to personal or sensitive information. Treatment addresses these risks through processes and governance structures, while rehydration focuses on integrating findings into policy and governance.
Sharma emphasized the need for robust encryption protocols, data anonymisation, and comprehensive data governance policies to protect sensitive information
The regulatory landscape presents significant challenges when using open-source LLMs. Sharma identified four primary concerns: data privacy, bias and fairness, explainability, and scalability. “Data privacy involves handling sensitive information in accordance with regulations. Bias and fairness require addressing ethical considerations and ensuring equitable model outcomes,” Sharma noted.
Explainability is another critical area. “Regulations often require model decisions to be explainable in simple terms. The ability to articulate how models operate and make decisions is essential for compliance,” he added. Sharma also noted the need for scalability and efficiency, ensuring that LLMs can grow with organisational needs while maintaining performance standards.
Bias Mitigation and Model Explainability
Biases in open-source LLMs are a serious concern. “To mitigate them, use diverse datasets and employ robust bias detection and mitigation techniques throughout the model’s lifecycle,” shared Sharma.
“Transparency and accountability are key. Tools like LIME (Local Interpretable Model-agnostic Explanations) can help explain model decisions and enhance trust,” he said.
Explainability is essential for understanding and trusting open-source LLMs. Sharma emphasized using tools and techniques that allow users to explore model predictions and understand outcomes interactively. “Incorporating attention mechanisms and saliency maps can provide insights into model predictions, making it easier to explain decisions in natural language,” he explained.
Performance and Customisation
Regarding performance, Sharma acknowledged that while open-source LLMs might initially lag proprietary models, they are rapidly improving. “Open-source models like GPT-Neo, Mistral, and Llama are now performing at levels comparable to proprietary models. Companies realise the potential of these models, with usage shifting from 80-20 in favour of proprietary models to a more balanced 50-50,” he noted.
Customisation is a significant advantage with open-source LLMs. Sharma highlighted that open-source models offer flexibility for tailoring solutions to meet specific financial regulations and requirements. Customisation allows for fine-tuning models to address unique financial needs and compliance standards. While this requires resources, it provides the opportunity to develop highly specialised applications.
Hidden Costs and Safety Considerations
However, Sharma cautioned about the hidden costs associated with open-source LLMs. “The initial investment in computing resources, development, and maintenance is significant. Customisation demands expertise and ongoing effort. Unlike proprietary models, which offer out-of-the-box solutions, open-source LLMs require substantial in-house development,” he stated.
Other costs include ensuring regulatory compliance and integrating the models into existing systems. Sharma noted that open-source solutions require careful consideration of security measures, compliance with regulations, and integration into business processes. The need for continuous performance optimisation and support adds to the overall investment.
Moreover, hiring the right talent requires capital. After the initial investment in compute, companies need good talent to maintain the system’s pipeline throughout adoption. “You have to have the right expertise to take it forward.”
Preventing the misuse of open-source LLMs is critical. Sharma mentioned best practices for mitigating risks such as malware and harmful content. Exposing LLMs to adversarial examples during training, implementing robust input validation, and controlling access are essential steps.
“Ensuring proper security environments and feedback loops helps protect against malicious activities,” he said.
Adoption and Scalability
Sharma noted that when adopting open-source LLMs, companies develop cold feet, especially when the conversation around scalability gains pace. “While companies are talking about generative AI and all the fancy stuff, they also want to talk about POCs,” Sharma said.
Enterprises need to engage about taking LLMs forward, as every company has different needs and what works right for them. “I think companies need to figure out where they stand on several levels before taking it forward,” Sharma added, and said that when it comes to the sustainability of these open-source LLMs in the long run, it “learns and grows.”
Once a couple of companies start adopting a specific LLM, others start learning and growing along the way. “It requires much investment, but then, that investment will pay off.”
“Sustainability is crucial,” Sharma emphasized. “We’re currently in the awareness stage, but assimilation and adoption will follow.”
The post Open-Source LLMs in Financial Services appeared first on AIM.
