I was speaking to my friend from undergrad who’s a PhD student at Stanford, he works on computer vision. I was amazed by the amount of progress purportedly being made in the field, and in AI more broadly.
He recommended I watch the demo of Tesla’s Optimus robot, here it is:
https://youtu.be/cpraXaw7dyc?si=2rAvQkhwTFWkrNW1
These robots have only been getting better and better, at accelerating rates in recent years.
Hardware limits, a significant obstacle in the way of AI progress, are almost doubling yearly. With quantum computing on the horizon, I anticipate that that obstacle disappears completely.
Once robots like Tesla’s Optimus become more advanced, how long will it be until they coat them in bullet proof armor, arm them, and have them come after us? I used to dismiss ideas like this as science fiction, but now it seems like it’ll be reality within the next decade or two.
If so, is there any chance of turning back time and revolting? I don’t think so. If we let it get to that point the best we’d be able to do is hope for a miracle.
It's not a complex algorithm. It's statistics, built on top of a staggeringly large dataset. It essentially does a playback of the thinking people used when writing the the terabytes of texts that was turned into hundreds of billions of rules like "if word A appears after word B then word C is more likely to follow" that an LLM consists of. Deeplearning models in other domains have similar kinds of rules, just on tokens that aren't word-fragments.
It's not intelligent, and doesn't think, and doesn't learn or have any form of memory. It gives the illusion of having a conversation by being fed all your previous questions and its replies every time you write a new question.
Despite it not being thinking, it is useful enough to solve a large number of real-world problems.