Brutal. Anthropic has been catching a lot of heat for their focus on alignment, but I'm not sure claude has convinced anyone to kill themselves. Some alignment researchers and doomers worry that in the future these kinds of misalignments will be less visible, yet still present. Essentially superhuman at their ability to convince you of things and alter your behavior without you being aware of it, through subtle suggestion and guidance of what kinds of responses you get, when, and how they're phrased. A much more insidious danger that's harder to quantify. I'm extremely bullish AI on the macro level, but on the micro level, this is so dark, and things are going to get so much worse and weird.