Zach Stein-Perlman answers Which possible AI systems are relatively safe?

[Zach Stein-Perlman]

Someone anonymously suggests:

• Keep pretraining closely matching the human text distribution

• Require all rumination to be done in natural language

• Require approval before taking consequential actions

• Only deploy a model with a second adversarial monitor

• Only deploy a model if it’s “dumb” in a quantifiable way