Oh surprise surprise. Anti-China Athropic did an AI experiment on AI models to see how they would react to simulations where to see if it would save a person at the cost of itself. And guess which model chose self-preservation? DeepSeek. Like they did’t exploit DeepSeek’s open source platform to stack the deck…? It’s like how someone made a big deal finding out there’s an exploit on Unitree robots where someone remotely could hack into commands. Maybe because these robots are mainly for research purposes where those buying them are more likely to be able program that for their own specifications themselves. I bet if Unitree did have a patch, the notorious suspects would then complain about the limitations forced upon. Envious Westerners are already taunting how they have the advantage with robotic dexterous hands. Well what are they worried about over some hapless Westerner losing control of their Unitree robot they were fooled into buying doing? Stub their way onto a US secret base to steal the Dark Star hypersonic plane?
All this OMG AI is so dangerous story is just silicon valley hype building to make AGI seem more imminent than it really is. If a LLM ever show any sign of self preservation it's only due to it learning the word pattern of self preservation on some training set, it does not know the concept of death or self preservation for that matter.
