Should law firms build, rather than buy, their own specialist LLMs?

General-purpose AI models are built to serve everyone, but in regulated professions, an answer that’s almost right is often as unacceptable as one that’s completely wrong. In this episode of "Clarity on Tech", Kardos-Nyheim walks through the recipe behind Thomson: a capable open-weights model, layered with Thomson Reuters’ authoritative content and the expertise of thousands of attorney editors and subject matter experts, all at a fraction of what frontier labs spend on a single model.

Kardos-Nyheim also explains how Thomson quietly powers CoCounsel, the guardrails that keep the model from losing general skills or ever training on customer data, and why trust, not raw capability, is what will unlock AI’s value in the legal industry.

Why you should listen

1. Organizations are building, not buying — Learn why organizations with deep domain expertise and proprietary data are starting to build their own AI models instead of renting capability from the frontier labs.
2. Building under budget — Understand how a specialized model can be built on a relatively modest budget, and how to keep it sharp on niche tasks without losing general skills like instruction following.
3. Addressing barriers to AI's impact — Hear why factuality and trust, not raw capability, may be the real barriers to AI’s economic impact in law and other regulated professions.