Looks like bad stuff is happening, really bad is a bit scary if you qualify that as threat to life or livelihood. Let's see what the next generation of models bring to this equation.
The next generation of the user is what I think will be the factor for AI-assisted loss of life, moreso than the models themselves. It's the trust we put in them.
The dinosaurs of the before-LLM age will increasingly struggle to convince young people e.g. to seek competent medical consultation, because ChatGPT can do that much better (it's always supportive, available at any hour, meets you where you are...) -- this aspect alone is bound to have a death toll associated with it.
Or imagine asking ChatGPT to map an Alpine climbing route, -- anything with a bad failure mode.
