Thomson Reuters turns to Chinese AI as Claude costs rise
Thomson-1 is built on Snowdon, which was created by 'realigning' Alibaba's open-source Qwen model
Thomson Reuters is turning to Chinese open-source artificial intelligence to reduce its reliance on costly models from Anthropic and build more AI capabilities in-house.
The Canadian legal and data company launched Thomson-1, its first AI model, on Monday, using Snowdon, a model created by modifying an open-source Qwen model developed by Chinese technology giant Alibaba.
This moves means that again Chinese open-source AI models are increasingly finding their way into Western technology companies as businesses look for cheaper ways to develop specialised AI systems. Thomson Reuters says Thomson-1 will initially focus on document review, while Claude will continue to power much of its existing work with Anthropic.
According to Thomson Reuters Chief Technology Officer Joel Hron, there was not only the issue of cost when deciding to build Thomson-1. An in-house AI language model has more customisation options for what a company is good at.
Instead of competing with Anthropic in general, Thomson-1 will do certain jobs where Thomson Reuters thinks its experience and information can make the difference.
“Our main objective is to make Thomson the model that powers more and more of CoCounsel's capabilities over time," Hron told Business Insider.
In May, Thomson Reuters deepened its collaboration with Anthropic for its AI lawyer assistant called CoCounsel, Hron explained.
More about Thomson-1
Thomson-1 is built on Snowdon, which was created by “realigning” Alibaba's open-source Qwen model. Because open-source models can be downloaded and modified, companies can adapt them for specific applications without developing an AI model entirely from scratch.
Not everyone is comfortable with the growing adoption of Chinese AI. Anthropic has accused Chinese AI companies of illicitly “distilling” outputs from its models to develop competing systems and has called for tighter US restrictions.
US lawmakers have also raised security concerns over American companies using Chinese open-source models. Senator Tom Cotton has questioned the potential risks, including the possibility of hidden backdoors in software based on Chinese AI technology.
