Here is a three-paragraph summary of the blog post for mathopen.com:
Scott Aaronson is teaching a brand-new graduate course at UT Austin this semester called CS395T AI Alignment Theory. The course tackles one of the most pressing questions in modern AI research: do we truly understand how to align and control powerful AI systems so that they reliably do what we actually want them to do? As AI capabilities have advanced at a remarkable pace over the past decade, concerns about alignment and safety have grown alongside them.
The course takes a theoretical and mathematically rigorous approach to AI alignment, which sets it apart from more practical or policy-focused treatments of the subject. By grounding the material in formal theory, students are challenged to think carefully and precisely about the core problems in AI safety, exploring what we can and cannot prove about the behavior of AI systems.
This kind of course represents an exciting development for the research community, as it brings the tools of theoretical computer science and mathematics to bear on some of the most important open questions of our time. For students and researchers interested in both AI and rigorous formal reasoning, a course like this sits at a fascinating intersection of cutting-edge technology and deep mathematical thinking.