Ruben Hassid · 11 min read · Gen AI Tools
777
Read the original on ruben.substack.com →
This article argues that AI models like Claude are inherently biased toward average, agreeable answers—always picking '7'—making them poor judges of creative decisions like naming, design, or content quality. The author advocates for leading with your own opinion first, then using AI to challenge it with data rather than generate opinions from scratch. The key insight is to use AI as an analytical tool, not a taste-maker.
Key Takeaways
- AI models are statistically biased toward the most average, likely answers (the '7'), making them unreliable for subjective creative judgments
- Never ask AI to evaluate its own output or choose between options—always bring your own opinion first and use AI to stress-test it
- Flip the prompt: give Claude your opinion and your data, then ask it to argue against you using only that material
- When AI-generated taste becomes universal, it loses differentiation—if everyone uses the same '7', nothing stands out
- Use AI to surface patterns and counterarguments, but keep humans (your audience, your own judgment) as the final decision-makers
Excerpt
Pick a number between 1 and 10. Got it? … You picked 7. Like one human out of 3. Now ask Claude. Or ChatGPT, Gemini, Grok. It’s always 7. Try it yourself: ask it to pick a number between 1 and 10. People are biased. But AI is much more biased. Funny with numbers. Less funny when you base your opinion on AI’s opinion. Because that’s what you do every day: “Is this post good?” “Which logo is better?” “What should I call my product?” And when everyone has the same 7 , it stops being tasteful: If everyone can…