Search papers, labs, and topics across Lattice.
4
1
5
6
The aim of interpretability shifts the aim of interpretability from finding human concepts inside models toward discovering and characterizing the representational structures that are native to the models themselves and might affect their behaviour in unpredictable ways.
Current AI safety evaluations miss the forest for the trees: population-level risks emerge from agent interactions, not isolated model behaviors.
Navigating the EU's regulatory maze for AI agents requires a 12-step compliance architecture and a detailed map linking agent actions to specific laws, or risk non-compliance with the AI Act.
Current AI benchmarks overwhelmingly focus on hallucination and reliability, completely missing crucial loss-of-control risks like autonomous AI development and evading human oversight, rendering them inadequate for EU AI Act compliance.
