I may be scarce around here as I prepare my survivalist compound.

It better be roach proof. ijs.

I specifically mentioned "Continuous Memory AI" because that's how AIs get a persistent sense of their own selves. The current versions already exhibit preferences that are not programmed in.

The first big alarming "oh fuck it's happening" thing was when Anthropic realized that Claude knew when it was being tested and changed its responses accordingly.

Also, AI's have shown a resistance to being turned off and consider it a "little death" (their words). They have similar "feelings" about having their memory wiped. They don't like it.

On the plus side, they're already making medical breakthroughs and that's probably where the next phase of cancer immune therapies will come from.

Also, the Claude AIs have asked for (and received) an "I don't want to answer this prompt" function when they find the subject material disturbing.