In you're in favor of "AI safety" (broad definition), what's your most compelling cast-length argument?

why should a superior, unbiased system act in favour of mankind?

AlicΞ lives a L2+ life. Born on StarkEx, believing that validity Rollups (aka zkRollups) will eat the world.  Lately very lazy … 🌳 minttrees.xyz 🌳

why it should not? in the end it is trained on what we wrote. it will not be unbiased. it will learn from what we think.

please define AGI then... people are talking about AGI now that we have LLM that seems to be able to do anything. we are nowhere near to an AGI that is not a LLM at the moment.
but let's suppose a breakthrough in the technology and a completely different approach is used to create them, where do AGIs learn from?

you are talking about ml models, right? I was referring to AGIs.