Christian Schroeder de Witt

Towards Understanding Multimodal Fine-Tuning: A Case Study into Spatial Features

AI Models Can Provably Hide Arbitrary Capabilities

h1: Bootstrapping Models to Reason over Longer Horizons via Reinforcement Learning

Predicting Weak-to-Strong Generalization from Latent Representations

Efficient Dictionary Learning with Switch Sparse Autoencoders

MALT: Improving Reasoning with Multi-Agent LLM Training

Fundamental Limitations in Pointwise Defences of LLM Finetuning APIs

Mixture of Experts Made Intrinsically Interpretable

MAD-Sherlock: Multi-Agent Debates for Out-of-Context Misinformation Detection

Hidden in Plain Text: Emergence & Mitigation of Steganographic Collusion in LLMs

Mitigating Goal Misgeneralization via Minimax Regret

SAGE: Scalable Ground Truth Evaluations for Large Sparse Autoencoders

Toward Robust Real-World Audio Deepfake Detection: Closing the Explainability Gap