AI Alignment
Ensuring AI systems pursue goals that match human intent and values — the hardest open problem in AI.
Definition
Alignment research asks: how do we make AI do what we actually want, not just what we literally said? Includes safety tuning, red teaming, interpretability. Many premium AI-training contracts feed alignment work.
Examples
- Constitutional AI training
- Red teaming
- Reward modeling for safety
Related terms
Reinforcement Learning from Human Feedback — the technique that made ChatGPT helpful instead of just plausible.
Adversarially probing AI for safety failures — getting it to produce harmful, biased, or dangerous outputs.
Training AI to refuse harmful requests, avoid bias, and stay within policy.
Anthropic's technique for self-supervised alignment — the AI critiques its own outputs against a written "constitution".
Looking for actual gigs that use AI Alignment? Browse all live listings or filter by $100+/hr.