Conversaciones con investigadores destacados.

Early Signs of Steganographic Capabilities in Frontier LLMs

Thought Anchors: Which LLM Reasoning Steps Matter?

Reverse-engineering a neural network that plans: a mesa-optimizer model organism

Real-Time Detection of Hallucinated Entities in Long-Form Generation

Control Tax: The Price of Keeping AI in Check

The Theoretical Foundations of Reward Learning

AI in a vat: Fundamental limits of efficient world modelling

Safeguarded AI: a scalable workflow for safety-by-construction

Mitigating Goal Misgeneralization via Minimax Regret