Japanese RHLF Evaluation & Creative Text Generation Project
I drove core SFT and RHLF activities to enhance the safety and performance of a Japanese language model. My responsibilities centered on creating high-quality SFT data, specifically by writing both prompt and model response pairs. I also conducted detailed, rubric-based evaluation and classification for RHLF tasks, including summarization and creative text generation. My contributions spanned multiple large-scale projects, including one global project involving approximately 4,000 contributors. I was repeatedly selected for high-tier reviewing and editing tasks. As a testament to my quality and linguistic precision, I was chosen as one of the elite 15 reviewers/editors within that 4,000-person project.