Dec 15, 2025 | AI and product innovation
From NeurIPS 2025: How Thomson Reuters Is Advancing the Frontier of AI Research
Last week, the Thomson Reuters Foundational Research team under TR Labs joined the global AI community in San Diego for NeurIPS 2025, one of the world’s most important gatherings in machine learning and artificial intelligence.
This was the second year that Thomson Reuters participated, reflecting how deeply we are investing in the science of AI and how seriously we take our role as a market-leading technology company shaping the future of trustworthy, professional-grade AI.
Building Connections and Sharing Breakthroughs
At our TR Labs booth, hundreds of researchers, practitioners, and students stopped by to learn about our latest foundational work in data mixture optimization, model efficiency, and responsible evaluation methods. Many were surprised to learn that Thomson Reuters not only applies AI across industries like law, tax, journalism, and compliance, but also actively publishes and contributes to open research.
As one of our researchers, Dr. Jonathan Schwarz, noted during his presentation, “We’re probably one of the few model training companies that openly publish our work.” That openness, paired with decades of domain-specific data and deep expertise, allows us to advance AI that is both state-of-the-art and built for real-world impact.
On Stage with Jonathan Schwarz
In one of the week’s featured sessions, Jonathan Schwarz, Head of Foundational Research, showcased two new publications that exemplify our commitment to scientific rigor and innovation.
He presented new data weighting techniques that improve model training efficiency by as much as 500 percent without architectural changes, underscoring the importance of high-quality, diverse data in model performance.
Jonathan also shared advances in efficient evaluation, which reduce the computational cost of benchmarking by up to 18 times. These projects demonstrate how Thomson Reuters blends academic-grade experimentation with industry-scale infrastructure, reaffirming our commitment to open science and responsible innovation.
Learning, Recruiting, and Engaging with the AI Community
Beyond our booth and presentations, our researchers spent the week exchanging ideas across thousands of poster sessions and keynotes covering everything from large language models to multi-agent reasoning.
As TR Labs researcher Wassim Seifeddine reflected, “Everyone here is working on their own corner of AI, and it’s inspiring to see how all those small ideas build toward something bigger.” Many attendees were surprised to discover the full scope of Thomson Reuters AI initiatives, a reminder that our work bridges deep research and enterprise-scale impact.
NeurIPS also provided an invaluable opportunity to meet exceptional talent. TR Labs connected with dozens of PhDs, postdocs, and engineers interested in joining our growing research hubs in London, Toronto, Zug and Minneapolis. Jonathan also announced our new academic partnership with Imperial College London, a collaboration that will further expand our ability to explore and publish foundational research.
Check out this video interview with Jonathan to hear more of his thoughts on NeurIPS 2025 and the skills he’s looking to hire for on his Foundational Research team.
Community, Connection, and Collaboration
We closed the week with a TR Labs happy hour, where researchers, collaborators, and new friends from across the AI ecosystem gathered to celebrate innovation and connection. The sunset over the ocean provided the perfect setting to reflect on the progress and possibilities that define this community.
Why NeurIPS Matters
For Thomson Reuters, NeurIPS is more than a conference. It represents our commitment to continuous learning and contribution. It is where ideas become experiments, and experiments become technologies that empower professionals around the world.
As The Neuron’s roundup of the best NeurIPS papers of 2025 highlighted, this year’s research centered on efficiency, interpretability, and trust. These themes align perfectly with our mission to deliver professional-grade generative AI that is both reliable and transformative.
By showing up, sharing our work, and engaging with the broader research community, we continue to demonstrate what it means to be a responsible leader in AI, one that advances science in service of society.
