Thomson Reuters CoCounsel Tests Custom LLM from OpenAI, Broadening its Multi-Model Product Strategy

Rhea-AI Summary

Thomson Reuters (TSX/NYSE: TRI) announced testing a custom version of their CoCounsel GenAI assistant built on OpenAI's o1-mini model. The custom model, developed in collaboration with OpenAI, leverages 'chain of thought' reasoning for enhanced analytical capabilities. CoCounsel, which saw a 1,400% increase in users in less than a year, is implementing a multi-model strategy, testing LLMs from various providers including Anthropic's Claude 3.5 and Google's Gemini 1.5 Pro. The company is also training its own models following the acquisition of Safe Sign Technologies and partnering with AWS and Microsoft for infrastructure and integration support.

Positive

• 1,400% increase in CoCounsel users in less than a year
• Strategic partnerships with major tech companies (OpenAI, Google, Anthropic, Microsoft, AWS)
• Expansion of AI capabilities through Safe Sign Technologies acquisition
• Integration with Microsoft 365 suite enhancing product accessibility

Negative

• Longer response times with new o1-mini model implementation

Global pioneer in professional-grade GenAI first to employ custom model built on OpenAI o1-mini.

• Custom model just the latest to be tested on the CoCounsel platform as part of company's multi-model approach to LLM-powered solutions 
• Thomson Reuters GenAI solutions for professionals, the only such products grounded in the world's leading content and expertise, provide uniquely sought-after testing ground for LLM technology leaders 

TORONTO, Nov. 25, 2024 /PRNewswire/ -- Thomson Reuters (TSX/NYSE: TRI), a content-driven technology company, today announced they are testing a version of their CoCounsel GenAI assistant that employs a custom model built on OpenAI o1-mini, the AI pioneer's latest large language model (LLM), notable for its enhanced reasoning skills. Because OpenAI o1-mini "thinks" before it responds—"chain of thought" reasoning—it enables CoCounsel to accurately perform certain highly analytical and data-driven tasks for the first time. While this means answers take longer, it also means they're more detailed, thorough, and accurate.  

Building on Thomson Reuters' proprietary datasets, the CoCounsel and OpenAI teams have collaborated to create this custom model, now in a proof-of-concept phase ahead of the new LLM's general availability. The original CoCounsel, launched in March 2023, was the first legal tech tool to publicly announce the use of OpenAI's GPT-4, and similarly was built in collaboration and early access with OpenAI, on its groundbreaking model. As CoCounsel has evolved and expanded, particularly since becoming the single GenAI assistant across Thomson Reuters products, so has the product team's strategy for employing LLMs to deliver the best possible performance for CoCounsel customers. Their current multi-model approach has set the standard for testing in-development models in partnership with LLM leaders.  

This is just the latest work between the two technology companies, as  CoCounsel was founded on a deep relationship with OpenAI, being the first legal tech tool to publicly announce the use of the GPT-4 large language model (LLM).  

"We're proud of our longstanding work with Thomson Reuters and learning how they use o1-mini to power new legal workflows," said Kevin Weil, Chief Product Officer at OpenAI. "Building a custom o1-mini model into CoCounsel equips legal professionals with advanced AI reasoning capabilities, supporting them in addressing challenging legal problems." 

Pioneering application of GenAI to professional-grade solutions 

As the first and leading solution of its kind—a professional-grade GenAI assistant—CoCounsel uniquely offers LLM creators a robust installed and daily active customer base, who are ready and willing to test the model evolutions. Many of these customers are early innovators in the GenAI revolution and have been users for more than 18 months. This has in part led to the 1,400% increase in CoCounsel users in less than a year. These users' multi-step use cases, such as legal workflows, offer particularly valuable testing ground to LLM development teams Additionally, CoCounsel offers LLM teams the opportunity to evaluate new models' performance across several products—including CoCounsel Core, Westlaw Precision, Practical Law Dynamic Tool Set, and CoCounsel Drafting—that tackle a wide range of sophisticated tasks.  

Thomson Reuters' LLM evaluation program includes in-product testing of new models from several AI providers, alongside current models. The goal is not just to determine which perform "better," but rather which—and in which combination—produce output that best meets customers' needs, for specific, real-life use cases. This process involves pushing the boundaries of what the models can do, giving valuable performance data to their creators—which Thomson Reuters is uniquely equipped to provide—and ultimately creating better technology and better results for customers who use the solutions built on it.  

"Our close cooperation with the technology leaders charting the course for AI allows us to deliver profoundly better CoCounsel outcomes for our customers, empowering them to better serve their clients," said Joel Hron, chief technology officer for Thomson Reuters. "Through early access to models from OpenAI, Anthropic, and Google, we've been able to evaluate each alone and in combination, right in product. And our learnings do more than help us continue pioneering professional-grade applications of GenAI, they help the teams behind these astonishing LLMs improve what the models can do. This benefits the products they power, creating a virtuous cycle of innovation to collectively push the boundaries of how AI can be used to transform professional work."  

Finding the right technology for each CoCounsel capability 

For Thomson Reuters products for tax professionals, beginning with Checkpoint, the CoCounsel team has been evaluating Anthropic's Claude 3.5 Haiku and Claude 3.5 Sonnet. The team is also looking at how prompt caching in Claude could be effective for CoCounsel Drafting.  

The integration of Google Gemini 1.5 Pro into the CoCounsel platform has been key to delivering the upgrades in CoCounsel 2.0. The models in this family offer a substantially longer "context window," enabling CoCounsel to increase processing throughput, better analyze complex patterns in legal documents, and retain the "memory" of legal matters across a series of tasks. 

In addition to collaborating with OpenAI, Anthropic, and Google, Thomson Reuters has begun training their own models, evidenced by their recent acquisition of UK-based startup Safe Sign Technologies, specializing in legal-specific LLMs. This future CoCounsel input will mark yet another opportunity for the team to evolve their multi-model strategy to better serve CoCounsel users. Key to delivering on this opportunity has been access to sufficient computational resources to train and run AI systems. Thomson Reuters has enabled this through close partnership with Amazon Web Services (AWS), becoming one of the early customers for AWS Sagemaker HyperPod. 

Thomson Reuters has also been working closely with Microsoft to bring another significant CoCounsel 2.0 upgrade to its customers: integration with M365. Just as important as ensuring CoCounsel can return the best possible output is empowering customers to streamline the complex work they do. By offering integration with the tools professionals use every day—Microsoft Outlook, Word, Teams, and SharePoint—the CoCounsel team continues its mission of creating a GenAI experience that doesn't just help professionals work faster, smarter, and better, but actually helps them transform what they spend their time on, what it's possible to accomplish, and ultimately what it means to work. 

Thomson Reuters 

Thomson Reuters (TSX/NYSE: TRI) ("TR") informs the way forward by bringing together the trusted content and technology that people and organizations need to make the right decisions. The company serves professionals across legal, tax, accounting, compliance, government, and media. Its products combine highly specialized software and insights to empower professionals with the data, intelligence, and solutions needed to make informed decisions, and to help institutions in their pursuit of justice, truth, and transparency. Reuters, part of Thomson Reuters, is a world-leading provider of trusted journalism and news. For more information, visit tr.com.  

Contact 
Ali Hughes +1.763.326.4421 ali.hughes@thomsonreuters.com   

View original content to download multimedia:https://www.prnewswire.com/news-releases/thomson-reuters-cocounsel-tests-custom-llm-from-openai-broadening-its-multi-model-product-strategy-302314877.html

SOURCE Thomson Reuters

FAQ

What is Thomson Reuters (TRI) testing with OpenAI's o1-mini model?

Thomson Reuters is testing a custom version of their CoCounsel GenAI assistant built on OpenAI's o1-mini model, which features enhanced reasoning capabilities through 'chain of thought' processing.

How much has Thomson Reuters (TRI) CoCounsel user base grown?

CoCounsel's user base has increased by 1,400% in less than a year.

Which AI models is Thomson Reuters (TRI) currently testing for CoCounsel?

Thomson Reuters is testing OpenAI's o1-mini, Anthropic's Claude 3.5 Haiku and Sonnet, and Google's Gemini 1.5 Pro models for various CoCounsel capabilities.

What recent acquisition has Thomson Reuters (TRI) made in the AI space?

Thomson Reuters has acquired UK-based Safe Sign Technologies, which specializes in legal-specific LLMs, to enhance their AI capabilities.