You call declaring LLMs conscious "safety"? That's just marketing, but believing it is psychosis.
Do you think it is a total and complete impossibility that we are in a simulation, or that it's completely impossible to simulate a mind?
Yes. And a simulated mind is not a mind.
AI safety in the context of the frontier labs is completely a joke, much like privacy to the previous iteration of big tech. They talk about safety but end up just releasing the next more powerful model. They only talk about “scaling back” when the models aren’t getting better to protect their valuations.
They release their models with safeguards HN whines endlessly about. They certainly do not "release their most powerful models"