Bary Levy
I'm a Security and Machine Learning researcher,
currently focused on building infrastructure for AI Safety and alignment research in Israel.
I also teach modern deep learning and LLM internals at various independent groups.
Control setting for detecting deceptive vulnerability patches from LLMs. Awarded 2nd place globally at AI Control Hackathon organized by Apart and Redwood Research.
Loading user-supplied tensors through API endpoints triggered memory corruption in PyTorch
Jupyter-lsp, a plugin installed by default on JupyterLab, allowed users to write to arbitrary files on the system, leading to remote code execution.
Semantic search engine for internal company data