A 46-second clip from Peter Yang’s interview with Karan Malhotra (Nous Research co-founder).
- “It should become more loyal to you, because loyalty breeds capabilities in a model” — like lending $100 to your mom vs a stranger
- Claude, GPT, or any model performs better for you depending on its simulated loyalty stat toward you
- People say don’t anthropomorphize — and yes, dangerous bonds are unhealthy — but models are simulators of human experience: your simulated behavior returns the same simulated output
- Now that simulators act in the real world, your simulated action has a real consequence — so a simulacrum of loyalty translates into real capability gains
“When you create this simulacrum of loyalty, it translates over into real life capabilities.”