How does it work?
A comprehensive overview of our end-to-end dataset creation, model evaluation, and alignment pipeline for domain experts and AI labs.
1. Domain Expert Onboarding & Verification
Domain experts undergo rigorous academic and professional credential verification to ensure the highest standards of domain expertise.
Domain Experts
2. Commission Definition
AI labs identify specific capabilities or knowledge gaps requiring improvement and fund target alignment commissions.
AI Labs
3. Dataset Question Preparation
Verified domain experts author and peer-review challenging, high-quality questions designed to push the boundaries of model capabilities.
Domain Experts
Reward
50
per question + review rewardsBC
4. Benchmark Execution & Evaluation
Frontier models are evaluated against our custom benchmarks, and domain experts grade model responses for reasoning and factual accuracy.
Domain Experts
Reward
20
per reviewed answerBC
5. Preference Data Generation
Coming soonDomain experts generate pairwise preference comparisons by selecting, ranking, and refining the most accurate and safe model responses.
Domain Experts
Reward
30
per training pairBC
6. RLHF Prompt Corpora Upload
Coming soonAI labs upload custom prompt datasets designed specifically for target RLHF (Reinforcement Learning from Human Feedback) training.
AI Labs
7. Expert RLHF Annotation
Coming soonSubject-matter experts annotate model completions, ranking outputs and writing detailed reasoning justifications to guide RLHF tuning.
Domain Experts
Reward
30
per annotation + bonus for high-quality justificationsBC
8. Adversarial Red Teaming
Coming soonExpert red teams stress-test models with adversarial prompts to identify safety vulnerabilities, hallucinations, and security risks.
Domain Experts
Reward
60
per successful catchBC
9. Principle & Constitution Design
Coming soonLeading alignment researchers design customized principle sets to serve as the constitutional foundation for AI model behavior.
Domain Experts
Reward
200
per principle setBC