Ever wondered how well Big AI follows the rules? LatticeFlow, a startup from ETH Zurich, is taking a stab at answering that question. They've created a way to measure if AI models stick to the European Union's new AI law. This law is like a set of rules for AI apps, and it's now in effect. But how do we check if AI models are following these rules? That's where LatticeFlow comes in.

They've made a new tool called Compl-AI. It's like a report card for AI models, checking if they're obeying the EU's rules. Big companies like Google, Meta, and OpenAI are on this report card. Some AI models do well, but others struggle with things like fairness and safety. LatticeFlow says this is because companies focus more on making AI models smart than safe.

This is just the beginning. The EU's rules are still being figured out, and LatticeFlow's tool is a work in progress. But it's a start. It's like trying to teach a naughty dog new tricks. It might take time, but it's important for the safety of everyone. LatticeFlow hopes other researchers and companies will help make their tool better.

AI Compliance: First Test of EU AI Act with LatticeFlow's LLM Framework

Actions