Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
@ohabryka @RyanGreenblatt @TomDavidsonX That does make it less obviously failure prone, and is part of what's proposed as plan A, but requires a far greater commitment than what is being proposed in almost any verification regime. And even then, it just means that a decisive advantage requires defeating those measures.
No, it mostly relies on intentionally designing things so that things get destroyed when the deal breaks down. Major costs get incurred like e.g. putting data centers into the ocean and outfitting them with self-destruction devices in order to ensure compute gets destroyed when things go south.
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.