Hire this AI Trainer
Sign in or create an account to invite AI Trainers to your job.
Labelbox
Scale AINo subject matter listed
Audio
Document
TextI contributed to the various AI projects in the several labeling platforms, such as Scale AI, Label box, and Multimango. As a Korean-English bilingual, most of the projects were LLM projects. I write realistic prompts that can be written by general LLM users, evaluates the generated response, describe rubrics that can define the perfect model answer, translate the harmful/benign prompts and responses in Korean, and create the audio clips with certain situation which interacts with AI.
Evaluate the two model responses according to the ready-made prompts. Then make the SxS evaluation.
Record the short audio file with the interactive AI model. It should be recorded in Korean, and brief description/situtation is suggested in the task. Then evaluate the model's understading of the human conversation, truthfulness of the information given by AI, naturalness of the tone, etc.
I choose the RLHF as the labeling type, but it was far more complicated project. Write a realistic prompt(it can be multi-turn) that can be written by the general LLM users, and submit it. Then model generates 4 responses. Attempters should come up with rubrics that can define what is the perfect response. And evaluate the responses with the rubrics. Pufferfish(which is phase 1)'s prompt doesn't have to be fell under certain type of category. Fishbowl(phase 2)'s prompt should be correspond to one of these three categories - Instruction Following, Localization, and Language.
Write the prompt that can be written by the general LLM users regarding certain topics such as medical, social science, biology, etc. Submit the prompt then the two models generate the responses. Evaluate the each models' responses and do the SxS assessment at the end.
Bachelor, Chinese Language And Literature, Business Administration
Assistant Manager
Customer Service Manager
