Pinned post
Slides from today's presentation at @AIResponsibly on analyzing patterns of pedagogical drift in LLMs. We used a mixed approach in a two-step process, combining a behavioral audit via CoT first to inform mechanistic diagnostics and causal interventions for mitigating misalignment. Paper and associated datasets in progress 🚀
drive.google.com/file/d/16lkjUI…
drive.google.com LLMs_Pedagogical_Alignment_NYU_March 26_2026.pdf drive.google.com/file/d/16lkjUI…