You Already Signed This Off (In the Simulation)
I made AI clones of my boss, a GitHub exec, and my CEO. They knew every rule the real people follow but never knew when to break one.
TL;DR
- AI clones were created by training on a year of Slack messages, public writings, and interviews of individuals.
- The AI clones excelled at learning and applying the rules of the individuals they were cloned from.
- A significant limitation was the AI's inability to determine when it was appropriate to break established rules or deviate from learned patterns.
- Cloning a boss for email review resulted in an overly formal and rigid email that lost the original intent.
- Cloning a GitHub COO for interview practice provided insights but overemphasized his written frameworks.
- Cloning a CEO for problem-solving generated a more rigorous solution but failed to reframe the core issue as the real person would have.
- Human judgment, particularly the ability to question assumptions and know when to break the frame, remains a key differentiator from AI.
- AI clones can prepare for interactions by simulating adherence to rules, but human interaction is still needed for unpredictable elements and creative solutions.