Preference Data
Pairs of AI outputs labeled "A is better than B" — the raw fuel for RLHF.
Definition
Preference data is the dataset of human-labeled comparisons used to train reward models. Workers see two AI-generated responses to the same prompt and pick the better one (or rank a list). This is the dominant gig type in 2026 AI training — high-volume, well-paid, requires careful judgment.
Examples
- You see two answers to "Explain quantum entanglement to a 12-year-old" and pick the clearer one
- You rank 4 code snippets from best to worst
Related terms
Reinforcement Learning from Human Feedback — the technique that made ChatGPT helpful instead of just plausible.
A small model that learns to predict which AI output a human would prefer — used to train the larger model.
A simpler alternative to RLHF that skips the reward model and trains directly from preference pairs.
Looking for actual gigs that use Preference Data? Browse all live listings or filter by $100+/hr.