Alexandre Drouin

DRBench: A Realistic Benchmark for Enterprise Deep Research

Deep Reflection Hinting: Leveraging Offline Knowledge for Improving LLM Agents Adaptation

Malice in Agentland: Down the Rabbit Hole of Backdoors in the AI Supply Chain

How to Train Your LLM Web Agent: A Statistical Diagnosis

Causal Differentiating Concepts: Interpreting LM Behavior via Causal Representation Learning

DoomArena: A framework for Testing AI Agents Against Evolving Security Threats

Context is Key: A Benchmark for Forecasting with Essential Textual Information

Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks

InsightBench: Evaluating Business Analytics Agents Through Multi-Step Insight Generation

Context is Key: A Benchmark for Forecasting with Essential Textual Information