As Large Language Models (LLMs) such as ChatGPT have gained in popularity, the importance of ensuring the safety of such models has also received unprecedented attention. However, the frontline workers doing the safety work have often been left out of the public discourse. To better understand the experience of AI Safety Trainers, and specifically how it may be similar to or different from content moderator (CoMos) experience, my team conducted in-depth interviews with 10 Trainers from 2 different workflows. The study revealed 3 major differences as well as 4 main similarities between AI Safety Trainers’ and CoMos’ experience. Based on these findings, we proposed wellness best practices for AI Safety Trainers, specifically when they may need interventions above and beyond what is usually offered for content moderation. This talk will cover the major findings from the research study as well as our recommendations on how to best support our frontline ensuring AI safety.
165 Tanjong Pagar Road