1 paper · 1 filter
C. Daniel Freeman, Laura Culp, Aaron Parisi +27
We introduce and study the problem of adversarial arithmetic, which provides a simple yet challenging testbed for language model alignment. This problem is comprised of arithmetic…