All Things Legal & AI
In Plain English: Fine-Tuning a Large Language Model on Your Legal Work Product (Part 3 - Model Results and Costs)
We benchmarked seven open-weight LLMs on a 500-item LegalBench subset, then priced the GPU hardware each one takes to serve. The leaderboard inverts: Qwen3.5-397B is the model a law firm should self-host and fine-tune.
June 28, 2026
In Plain English: Fine-Tuning a Large Language Model on Your Legal Work Product (Part 2 - Model Evals in Braintrust)
A plain-English walkthrough for setting up legal model evaluations in Braintrust, using LegalBench open and closed question datasets, LLM judges, and repeatable scorer workflows.
June 24, 2026
In Plain English: Using Your Existing Legal Work Product to Fine-Tune a Large Language Model (Part 1 - Your First Dataset)
Ten detailed example prompts covering litigation, M&A, employment, real estate, IP, tax, healthcare, bankruptcy, privacy, and contracts — use these as reference when writing prompts for your firm's best work product.
May 2, 2026
SDNY Ruling: Why 'Private' AI Isn't Enough for Privilege
A recent SDNY ruling found that using AI tools like Claude can waive attorney-client privilege. Learn why UnitizeAI's Zero Data Retention architecture protects you.
February 18, 2026
We Don't Just Say Your Data Is Isolated. We Prove It With Cryptography.
UnitizeAI's isolation audit is a cryptographically signed report that proves your data was processed in a fully isolated environment - and you can verify it yourself.
February 1, 2026
How to Split a Large PDF with AI for eDiscovery and Relativity
Learn how AI-powered document unitization splits large PDFs into individual documents, ready for eDiscovery review platforms like Relativity and Everlaw.
January 30, 2026