David Krueger

Discover the need for visibility and governance in AI agent operations.

Access levels in AI audits influence effectiveness and risk identification.

Exploring how to build safety cases for AI technologies.

Developers must prove AI systems are safe to manage risks effectively.

A new model concept shows how to test AI capabilities effectively.

Learn why unique IDs for AI systems enhance safety and trust.

Examining the difficulties of creating effective reward functions in reinforcement learning.

This article analyzes model performance across various tasks and datasets.

New method BaDLoss enhances protection against data poisoning in machine learning.

This article discusses methods to better understand neural networks through Sparse Autoencoders and Mutual Feature Regularization.

Exploring how transformers learn and the challenges they face against attacks.

Researchers develop a method for AI to coordinate without full information.

A study on two approaches to improve AI's performance in language tasks.

Learn how machines can forget unnecessary data for better privacy.