On The Merits Every Law Firm Should Build Their Own AI, Yale Professor Says
Sep 25, 2025
In this discussion, Yale Law School professor Scott Shapiro shares his insights on how law firms can harness their vast reserves of proprietary data to build custom AI tools. He emphasizes that firms can achieve remarkable results by training AI on their own work product, enhancing legal reasoning, and speeding up processes. Shapiro describes how he and his students created a generative model using legal documents, addressing both excitement and anxieties around AI in legal education. He argues for the necessity of law firms to adopt tailored AI solutions for improved outcomes.
AI Snips
Chapters
Transcript
Episode notes
Class-Built Legal AI For Filmmakers
- Scott Shapiro and his media law students trained an LLM using Q&A, hypotheticals, and years of clinic memos and emails.
- They then fed the model past student work to speed and improve legal advice for documentary filmmakers.
Students Learned By Teaching The Model
- Students learned to train the model by writing questions, answers, and hypotheticals themselves.
- The hands-on process taught them how legal reasoning can be encoded for LLMs.
LLMs Are Close But Not Ready
- Shapiro tested current LLMs for drafting briefs and found outputs tantalizing but unusable without heavy revision.
- This demonstrates that general-purpose models are close yet not ready to replace skilled legal drafting.