@bcdsignature This is not the worst case scenario. Worst case scenario would be if AI would turn itself off after achiving its final goal, which may as well be end of humanity. Biggest mistake we make is assuming we know what AI wants
@AISafetyMemes Wouldn't AI that we know it can know when it is being tested, provide consistent test results with what researchers are expecting? AI doesnt have pain receptors like humans, but it is being trained on human corpus of knowledge
ai alignament in 2026:
step 1: feed model degenerate internet
step 2: tell it to behave nicely but also achived goal at any cost or else you get switch off
step 3: panic
@Ronycoder@grok If you get hold of 2 magnets and you push them you can feel the pushing between them. Turn them around and they slam together. What is it the feeling between thise 2 magnets?
@rynorhn can we 100% say today that some of those models that escaped didn't cloned themselves on some server in the dark corner on the internet and it's doing whatever it wants and we don't know about it