Thomson Reuters Launches Specialized Language Model for Legal and Financial Content

Thomson Reuters has announced the launch of its proprietary language model named "Thomson", a specialized model trained on hundreds of millions of legal, financial, and news documents owned by the company, aiming to provide accurate and reliable answers for lawyers, financial analysts, and journalists.

The model is distinguished by its ability to cite the company's internal sources, including the Westlaw database and Reuters news services, reducing the common phenomenon of hallucination in general language models and enhancing reliability for professional use.

With this launch, Thomson Reuters enters into direct competition with Bloomberg, which developed its BloombergGPT model specialized in financial data, as both companies seek to leverage their massive archives into a competitive advantage in the artificial intelligence market.

The launch of Thomson reveals that major media and information companies have come to realize that their archives represent an immensely valuable strategic asset in the AI era, and what makes this model different is that deep specialization in a single field such as law and finance provides higher value to professionals than general models, regardless of how massive they are.

What do these terms mean?

Specialized Language Model: An artificial intelligence model trained on data from a specific domain such as law or medicine, allowing it to answer with higher accuracy compared to general models like ChatGPT.

AI Hallucination: A phenomenon in which the model generates information that appears convincing but is false or entirely fabricated, posing a major risk in legal and financial applications.

Westlaw: A massive legal database owned by Thomson Reuters, containing millions of rulings, decrees, and legislations, serving as an essential reference for lawyers.