I recently participated in ARENA, a technical AI safety accelerator. For my weeklong capstone project, I investigated current AI capabilities to reason in ways humans don’t understand. Read more on LessWrong.