> I think people kind of assumed it would take AGI taking off to end human civilization, but at this point I believe it's pretty clear that LLMs as a class of m

> I think people kind of assumed it would take AGI taking off to end human civilization, but at this point I believe it's pretty clear that LLMs as a class of models are up to the task already Absolutely, and in addition to that, people with zero grasp of the risks make decisions to trust AI with dangerous things -> money, vehicles, markets, justice, etc. > Which invites the question, if a lot of our cognition works this way, then just how much more complicated are the bits that don't? I this we could be mistaking quantity for the important. Maybe other aspects of thinking, the ones we perceive as mundane, are actually impossible for a computer. Consider mechanical tasks - a hundred years ago we had huge breakthroughs with engines - oh look, we can make crazy strong machines, we will soon automate all manual labour and blue collar jobs will be a thing of the past! Well, turns out power was the easy part. A robot arm can lift a car, but getting precision took decades. So now we have power and precision, but try to get a robot arm (under computer control) to undo buttons on your T-shirt without ripping the T-shirt or your face. If a robot arm touches a piece of fabric, or you skin, it has no idea. They routinely kill people. If you need non-repetitive dexterous movement with fine sensory input, that's basically beyond the cutting edge for today's robots. We can't automate sewing of clothes, cleaning of delicate items, etc.