I built a game where domain experts try to break frontier AI
I built a game where domain experts try to break frontier AI
Visit I built a game where domain experts try to break frontier AI
What is I built a game where domain experts try to break frontier AI?
This is a competitive platform where domain experts submit questions designed to expose weaknesses in frontier AI models like GPT-4 and Claude. You create questions within your field of expertise, test them against current AI systems, and get paid when your question successfully stumps the model in a way that's verified by the platform. The goal is to systematically document where advanced AI fails in real-world professional contexts. It's useful for researchers, subject matter experts, and organisations wanting to understand AI limitations before deployment, whilst also creating a public record of frontier model weaknesses.
Key Features
Expert question submission
create and submit domain-specific questions designed to test AI model limits
Multi-model testing
questions are evaluated against several frontier AI models
Verification system
submitted failures are checked by the platform to confirm legitimacy
Payment for verified fails
receive compensation when your question successfully breaks a model
Failure documentation
contribute to a growing public record of where AI systems struggle
Freemium access
free tier for question submission with paid rewards for verified results
Pros & Cons
Advantages
- Direct way to contribute to AI safety and transparency research
- Earn money by finding genuine model failures in your area of expertise
- Helps identify real-world gaps before AI is widely deployed in your field
- Simple submission process without needing technical knowledge
- Creates publicly documented evidence of AI limitations
Limitations
- Payment amounts and verification criteria are not clearly specified upfront
- Limited to written questions; may not capture all types of failures (visual, audio, etc.)
- Success depends on finding genuinely novel failures that haven't been discovered before
Use Cases
Medical professionals identifying diagnostic scenarios where AI gives incorrect advice
Lawyers finding legal interpretation errors in AI responses
Academics testing AI reasoning in specialist research areas
Engineers discovering safety oversights in technical explanations
Content creators finding factual errors in their industry domains