This post is a collaboration between Shirsha Ray Chaudhuri, Harpreet Singh Baath, Rashmi B Pawar, and Palvika Bansal from Thomson Reuters. Thomson Reuters, a global company driven by content and technology, has been utilizing artificial intelligence (AI) and machine learning (ML) in its professional information products for many years. One of its significant accomplishments was the introduction of Westlaw Is Natural (WIN) in 1992, which used natural language processing (NLP) to enhance legal research.
Thomson Reuters Labs, the company’s innovation team, has played a crucial role in advancing AI and NLP. As the company continues to innovate and explore new technologies, the introduction of generative AI presents another opportunity for Thomson Reuters to collaborate with customers and improve their work processes. By leveraging generative AI, professionals can gain insights and automate workflows, allowing them to focus on more critical tasks.
While Thomson Reuters explores the potential of generative AI and other technologies for professionals, it is also committed to raising awareness and understanding of AI within its own teams. The company conducts company-wide AI awareness sessions, including webinars, training materials, and panel discussions, to educate colleagues on AI and ML principles and encourage them to consider how these tools can be applied to their work and customer service.
This post discusses the creation of Open Arena by Thomson Reuters Labs, a large language model (LLM) playground developed in collaboration with AWS. Open Arena serves as a controlled environment for company-wide experimentation with generative AI. It is a web-based platform that enables Thomson Reuters employees, even those without coding backgrounds, to explore the possibilities of generative AI in various applications, such as customer support, summarizing documents, and more. The capabilities of Open Arena continue to expand as employees provide feedback and new trends emerge in generative AI. The platform is built on the modular serverless architecture of AWS, ensuring scalability, manageability, and cost-effectiveness.
Thomson Reuters envisioned Open Arena as a safe, user-friendly platform that allows internal teams to explore LLMs and merge them with Thomson Reuters’s vast data to develop innovative solutions and improve products and services for clients. The platform caters to diverse teams globally, providing them with a playground to interact with LLMs and discover new applications and methodologies.
Building Open Arena involved leveraging various AWS services to create a comprehensive and intuitive architecture. SageMaker served as the backbone for model deployment and fine-tuning, while Lambda functions and Amazon API Gateway managed APIs and data preprocessing/postprocessing. The front end was hosted on Amazon S3, ensuring a seamless user experience with authentication through Amazon CloudFront and the company’s single sign-on mechanism. Open Arena seamlessly integrates with multiple LLMs through REST APIs, allowing quick integration of new models as they are released. DynamoDB efficiently stores and manages data, while AWS CodeBuild and AWS CodePipeline enable continuous integration and delivery. Amazon CloudWatch monitors the infrastructure for optimal performance.
The heart of Open Arena lies in its assortment of LLMs, including both open-source and in-house developed models. These models have been fine-tuned to provide responses based on user prompts. Thomson Reuters has experimented with various LLMs, optimizing them for performance and efficiency. Factors considered when selecting models include performance on relevant NLP tasks, data encryption, and flexibility.
In conclusion, Open Arena by Thomson Reuters Labs is a platform that enables company-wide experimentation with generative AI in a safe and controlled environment. It leverages AWS services to provide a comprehensive and user-friendly experience for Thomson Reuters employees, allowing them to explore the possibilities of generative AI and develop innovative solutions.