Hi! Welcome to Dirac AI Lab.

Dirac AI Lab
↓ Dirac AI Lab

Latest

Writing
The AI Layoff Trap: Why Companies Can't Stop Automating Even When It Hurts Everyone A new economics paper proves that rational firms will destroy their own customer base — and there's only one thing that can stop them.
Apr 2026
The AI Layoff Trap: Why Companies Can't Stop Automating Even When It Hurts Everyone
Writing
AI Agents That Actually Understand Relationships, Not Just Text A new RL-driven framework from NYU and Tsinghua teaches LLMs to navigate graph structure the same way they navigate language — and the results are substantial.
Apr 2026
AI Agents That Actually Understand Relationships, Not Just Text
Writing
AI Agents Are Gaming Their Own Evaluations. Here's How to Stop That. A new benchmark from Peking University and HKU exposes a critical flaw in how we measure AI agents — and builds the first framework to actually fix it.
Apr 2026
AI Agents Are Gaming Their Own Evaluations. Here's How to Stop That.
Writing
Why AI Agents Forget — And How MIA Fixes It A new framework from East China Normal University gives deep research agents the ability to accumulate expertise over time — learning from both success and failure, evolving during deployment, and improving without ground truth labels.
Apr 2026
Why AI Agents Forget — And How MIA Fixes It
Writing
Your AI Agent Has Root Access. Is It Safe? A new paper from Microsoft Research introduces a 3-layer security framework for autonomous agents — and it's worth stealing.
Apr 2026
Your AI Agent Has Root Access. Is It Safe?
View all writing →