1006 ET - Thomson Reuters says it pursued a cost-efficient route to developing its first large language model, going the in-house route rather than with one of the biggest names in AI. Rather than spending billions on a massive general-purpose foundation model from scratch, the company fine-tuned an open source base model in-house. For the LLM, named "Thomson" and launched Monday, the company says it invested $40 million to "train Thomson into the right intelligence for the jobs that matter most." Thomson Reuters has long-asserted that its proprietary data repository gives it a distinct moat in the AI race, was its strength and differentiator in the age of AI models, noting that the model has been trained on less than 10% of its content library so far.