Skip to main content

Irys for Word is here. Install Free →

Research

Irys University

Research essays from the team building Irys on how legal AI should work. Writing on auditability, matter context, source-backed answers, and reliable work product for legal teams.

Reinforcement Learning Is the Wrong Bet: Four Structural Failures and What Works InsteadResearch

Reinforcement Learning Is the Wrong Bet: Four Structural Failures and What Works Instead

16 min read
Active Inference & The Free Energy Principle: From First Principles to Working SystemsResearch

Active Inference & The Free Energy Principle: From First Principles to Working Systems

14 min read
GLM 5.2: How Four Co-Designed Mechanisms Beat Raw ScaleResearch

GLM 5.2: How Four Co-Designed Mechanisms Beat Raw Scale

18 min read
What AlphaEvolve Teaches Legal AI About Evaluated WorkResearch

What AlphaEvolve Teaches Legal AI About Evaluated Work

6 min read
How to Train Graph Neural Networks: From First Principles to Planet ScaleResearch

How to Train Graph Neural Networks: From First Principles to Planet Scale

10 min read
Auditability Is The Product In Legal AIResearch

Auditability Is The Product In Legal AI

8 min read
Why Prompt-As-Memory Fails In Legal AIResearch

Why Prompt-As-Memory Fails In Legal AI

10 min read
Matter Memory Must Preserve ContradictionsResearch

Matter Memory Must Preserve Contradictions

3 min read
Prompt Instructions Are Not Control In Legal AIResearch

Prompt Instructions Are Not Control In Legal AI

3 min read
The Exception Leak In Legal AIResearch

The Exception Leak In Legal AI

3 min read
A Citation Is Not An Audit TrailResearch

A Citation Is Not An Audit Trail

3 min read
Rerunning Is Not Repair In Legal AIResearch

Rerunning Is Not Repair In Legal AI

3 min read