Why this course
1. The Rapid Rise and Power of LLMs
LLMs have already surpassed human-level performance in many domains and continue evolving at an exponential pace. Understanding their foundations and capabilities is essential for anyone working in AI.
2. Navigating Ciritical Challenges in Responsible AI
As LLMs become increasingly powerful, three central challenges emerge:
- Control: Who governs AI development—centralized or decentralized, open or closed?
- Trustworthiness: How do we ensure robustness, privacy, fairness, and ethical alignment?
- Safety: How do we prevent misuse and prepare for advanced AI systems?
This course equips you to think critically about these questions and contribute to safer, more responsible AI development.
Topics to be covered
Understanding:
- Foundations of LLMs
- Interpretability
- Scaling laws
- Reasoning and mathematics
- Agency and emergence
- Evaluation and benchmarking
AI Safety:
- AI alignmnet and governance
- Adversarial robustness
- Trojans
- Privacy, Unlearning
Research Genre
- new methods for interpretability
- new frameworks for evaluating safety properties, including robustness, monitoring, controllability, etc
- new methods for mitigating biases in LLMs
- new methods for unlearning hazardous or undesirable capabilities
- new methods for jailbreaking, adversarial attacks on LLMs & mitigations