I have had multiple encounters now with people interested or working in Machine Learning/AI Alignment, who wish they had more mathematical background. Most of them come from a Computer Science background, which is not surprising. Despite Machine Learning often being done in CS departments, the typical CS undergrad curriculum contains much less math than what one can find in some ML papers. In my experience, CS people struggle the most with probability and statistics, despite those being an integral part of modern ML (GPT-style models and Diffusion models are both inherently probabilistic, for example).

Continue reading