Imagine writing a policy for an AI company.
 

The CEO trusts you and will approve your policy, if you show it's based on 8 premises:

1. AI can be scaled in capability. It can offer benefits and power to those who wield it. Also it could become unsafe for all of humanity.
2. There are 'bad guys' scaling AI: people who are scaling on for temporary benefits or strategic power advantages, even if it leads to everyone dying.
3. There are also 'good guys': people who are developing AI to stay safe for humanity.
4. If we pause scaling because we think AI is becoming less safe, the other 'bad guys' will want to keep scaling regardless.
5. If we fall behind, we lose the power to try make the most advanced AI safer. 
6. If we keep scaling ahead of 'bad guys', we can try make and distribute a safer version of the most advanced AI than what those other 'bad guys' would make. 
7. But other 'bad guys' will learn from us and catch up, developing their own version of the most advanced AI faster than they would otherwise have. 
8. If ever, anyone builds an existentially unsafe version, everyone dies.
    

The CEO wants their company to be the 'good guys'. They want a policy to not only use, but also advocate/lobby for hard so other companies also start acting as the 'good guys'.

What policy do you write?