Yarin Gal

8 papers University of Oxford

• Fundamental Limitations in Pointwise Defences of LLM Finetuning APIs
• MIP against Agent: Malicious Image Patches Hijacking Multimodal OS Agents
• Measuring what Matters: Construct Validity in Large Language Model Benchmarks
• Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning
• SECODEPLT: A Unified Benchmark for Evaluating the Security Risks and Capabilities of Code GenAI
• Scaling Up Active Testing to Large Language Models
• Security Challenges in AI Agent Deployment: Insights from a Large Scale Public Competition
• Temporal-Difference Variational Continual Learning