Thomson reuters develops in-house AI model using chinese open-source technology

Thomson Reuters, a Canadian data and legal services company, has launched its first in-house AI model, Thomson-1, which is based on a modified version of Alibaba's open-source Qwen model. The model, developed in collaboration with Imperial College London, was adapted over several months to ensure it is ethically and politically de-biased. The company's chief technology officer, Joel Hron, stated that the decision to build its own AI model was driven by the desire to reduce reliance on costly AI from Anthropic and to leverage the company's expertise in specific areas such as document review. Thomson-1 is not intended to replace existing AI tools like Claude but rather to complement them by focusing on areas where the company has specialized knowledge. The move reflects a growing trend among Western tech companies to adopt cheaper Chinese open-source AI models, which are seen as a potential boon in the face of high AI costs. Hron emphasized that having an in-house model allows Thomson Reuters to build on its own intellectual property rather than continuously paying for external AI services. He compared this approach to buying a house instead of renting, highlighting the long-term value of owning the asset. While some have raised concerns about the use of Chinese open-source models, including allegations of illicit data extraction by Chinese labs, Hron noted that Thomson Reuters has taken steps to ensure the model is safe and secure for use. The company continues to work with Anthropic on its AI legal assistant, CoCounsel, which still relies primarily on Claude. Anthropic and Alibaba did not respond to requests for comment on the matter.