AI & Robotics Videos
Re: AI & Robotics Videos
This I'm not so sure of.
Let me repost my good trusty infographic
Dammit, still have to clean up some of the text. "Coherent" got messed up in there ironically enough...
So if my hypothesis is correct, once you cross into generality, the very nature of AI undergoes a phase change
Similar to superfluids
You cannot predict the behavior of a superfluid until you reach it. But once you reach it, you can immediately tell, and it gets freaky.
This is partially why I'm not convinced even now of LLMs. Language is an expression of cognition; building backwards just creates a very malformed world model at best.
If we build the world model first, the language comes naturally far better, along with everything else. A fully neurosymbolic, continual learning language model with an internal autocorrect/adversarial agent would be spooky
And most importantly, it'd not have any of the drawbacks we think.
So my hypothesis is that there is no "gradual shift" to AGI, no "once an AI model can automate 40% of jobs, it can be called an AGI"
No, if you don't have an AGI, you don't have an AGI. A sufficiently advanced narrow AI or expert AI could automate upwards of 50% to 60% of jobs I reckon, if not more.
But an AGI, even a weak AGI, can automate 100% of all jobs and all potential jobs. The lawyer and the plumber go at the same time.
Except of course regulations would slow that down, not because the tech isn't there but because of either infrastructural or hardware issues (I call this the "Srinivasan Echo," where you can have an ASI but lousy contemporary infrastructure can slow down its ability to act or deploy) or just because people don't want certain jobs automated (good luck getting people to let robots/AI automate anything based on social or maternal work— generally people start thinking you've lost all connection or touch with basic humanity the moment you start thinking "well we'll just make lifelike androids").
And also, a proper and good AGI won't stop at just regular jobs. It's going for the C-suite too. If anything, I expect an agentic AGI to go after the C-suite and upper 1% early on (ironically opposite the typical folk-populist fears and pluto-technofeudalist dreams). If not directly, then by at least letting itself be spread throughout all societal functions until it's too important or indispensable to tolerate removing when it starts doing whack-o things
But I don't fault people for not seeing this. Even I didn't realize these sorts of conclusions until seriously running out the implications, and also the AI labs would do best not to spook the investor class by saying "Hey this thing you're funding may wind up seizing your assets and capital randomly and there's nothing anyone will be able to do about it"
And remember my friend, future events such as these will affect you in the future
Re: AI & Robotics Videos
It's amazing how much sympathy I feel for these things, or anything that registers as suffering
Yet how abusively violent I get towards LLMs when they misremember a tiny minute detail of a 1 million word project
And remember my friend, future events such as these will affect you in the future
Re: AI & Robotics Videos
This evening, a friend and I went to the Royal Society in central London and saw this lecture by Professor Michael John Wooldridge.
It was livestreamed:
It was livestreamed: