Thomson Reuters AI: Fiduciary-Grade Model for $40M

Thomson Reuters has launched its first proprietary large language model, named Thomson. The company invested $40 million to develop this in-house model. What's interesting is that Thomson Reuters built this model on decades of its own proprietary content and expertise. They say this approach made it possible to create a "Fiduciary-Grade" model at a fraction of the typical cost. Joel Hron, the Chief Technology Officer, states that Thomson shows an alternative to simply building bigger models with more compute. The model uses state-of-the-art training techniques, drawing on content from sources like Westlaw and Reuters. Steve Hasker, CEO of Thomson Reuters, notes that early evaluations show Thomson performing on par with leading frontier models across various tasks. The model is already being used in CoCounsel Legal. This development matters because it suggests a new way to build powerful AI, focusing on specialized, efficient, and controlled intelligence.