tfw a minor tweak to your custom instruction sends you to the entirely humorless ancestor-expert... smdh
also... this is also still a mental stumbling-block for me wrt ASI doom... how are we gonna build a superintelligence w full internal alignment and yet not have discovered techniques that would also apply to humans & human-AI coexistence?
More notes on nearby ideas
Reply · Apr 3, 2023"even a superintelligent AI can face the alignment conundrum how to learn human values without misinterpreting them like a digital King Midas turning…Reply · Feb 22, 2023@conjurial 1 is full speed ahead to ASI, but inherently in the nature of superintelligece it is not xrisk-scale murderous or accidentally what we…Reply · Apr 3, 2023"two superintelligent heads are better than one"Reply · Aug 21, 2023@broductmommy this is why we need to build the Superintelligence asap!!!