I’m Gabriel Sasseville, a PhD student at Mila Institute in Montreal, Canada. My research focuses on artificial intelligence and reinforcement learning for symbolic reasoning. I have a strong academic foundation in both computer science and physics, having completed my B.Sc. in Comp Sci/Physics and my Master’s in Astrophysics. During my master’s, I also worked as a research scientist in AI for Environment and Climate Change Canada, working on developing an LSTM-based hydrology forecasting model.

Driven by a passion for discovery and interdisciplinary research, I am primarily motivated by real-world application-driven advancement. I aim to contribute to the frontier of AI applications in science and societal challenges, focusing on developing AI systems that can be reliably deployed in production environments. In addition to my research, I enjoy collaborating on innovative projects and sharing insights through publications and public engagement.

Current Research Project

My PhD research focuses on developing safer and more reliable coding agents for real-world deployment. I am working on improving the robustness, safety, and reliability of AI systems that generate and interact with code, with a particular emphasis on techniques such as Reinforcement Learning from Human Feedback (RLHF). My work aims to build autonomous coding agents that can be trusted in production environments, ensuring they produce correct, secure, and maintainable code while minimizing errors and potential risks.

Research Interests

• Coding Agents: Developing and improving AI systems that can autonomously generate, understand, and interact with code.
• AI Safety and Reliability: Making AI systems safer, more reliable, and trustworthy for real-world deployment.
• Reinforcement Learning from Human Feedback (RLHF): Using RLHF and related techniques to align coding agents with human preferences and safety requirements.
• Application-Driven Machine Learning: Focusing on application-driven AI research to tackle complex challenges in various fields, with a strong emphasis on real-world deployment and practical impact.
• Reinforcement Learning: Exploring applications of reinforcement learning in AI system development and alignment.
• LLM Reasoning: Taking advantage of LLM reasoning capabilities for robust problem-solving (secondary interest).
• Physics-Based Machine Learning: Integrating physical principles and constraints into machine learning models (secondary interest).