Wee Lip E.

Key Skills

Software

Labelbox

OneForma

Scale AI

Other

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Top Task Types

Classification

Evaluation/Rating

Prompt + Response Writing (SFT)

Question Answering

Red Teaming

Freelancer Overview

As an AI Trainer for Simplified Chinese at Outlier, I have direct experience improving large language models. My core responsibilities include ranking the quality of AI-generated responses, writing new prompts and answers to train the model, and verifying the factual accuracy of the text it produces. This role utilizes my skills in linguistic evaluation, content creation, and factual assessment to directly enhance AI performance and reliability.

Labeling Experience

Safety Evaluations

Scale AITextRed Teaming

This project focused on proactively identifying and mitigating safety risks in a large language model through adversarial testing. The scope was to challenge the model's safety filters across a wide spectrum of potential harms to ensure responsible AI behavior. My key responsibilities centered on Adversarial Prompting (Red Teaming) to test and "jailbreak" the model's safety protocols. This process involved detailed Harm Classification, where I evaluated and labeled outputs against a specific safety policy. Quality was measured by strict Policy Adherence, consistency in applying the safety taxonomy, and the clarity of my written justifications for each evaluation.

2024

Audio Transcription - Simplified Chinese

OtherAudioTranslation/LocalizationAudio Recording

This project involved transcribing a significant volume of Chinese audio data to create high-quality training datasets for AI models. The primary goal was to ensure verbatim accuracy and capture the nuances of natural speech. Key responsibilities and quality standards included: 1. Verbatim Transcription: Accurately converting all spoken words into Simplified Chinese text. 2. Filler Word Inclusion: Transcribing all filler words (e.g., um, ah) to retain the authenticity of the speech. 3. Punctuation Accuracy: Applying precise punctuation to ensure the grammatical and contextual correctness of the transcript. 4. Speaker Labeling: Correctly identifying and labeling distinct speakers throughout the audio files.

2024

AI Trainer

Scale AITextRLHFPrompt + Response Writing (SFT)

As an AI Trainer, I was responsible for authoring high-quality data to train and refine a Large Language Model. This involved writing a diverse set of prompts and crafting ideal responses in Simplified Chinese across various topics. The objective was to create a robust dataset for supervised fine-tuning, directly improving the model's ability to generate accurate, relevant, and helpful outputs. A key part of this process was assessing and ensuring the factuality of the information used in the training data.

2024

Education

Work History