Neural-network-based AI has been around since the 50's, and pretty much the entire time it's existed, people have been positing about an AI that could match or outpace humans. It's gone under various terms, like post-singularity AI, strong AI, and AGI.
Over the decades, various generations of AI have advanced enough to fill various niches, but each time the hype far outpaces the abilities, and the goals get diluted down until they are more practical. The term AGI is general enough that it doesn't mean anything, so whatever we achieve with this generation of AI infrastructure we can call AGI.
We're still really far away from getting a neural network that's anywhere near the capability of a reasonably capable vertebrate. The way LLM training works is more like selective breeding than learning. To get a neural network that behaves the same way a your pet dog or bird thinks, each back-and-forth in a conversation would require the equivalent of retraining a foundational model.
It's not something we'll never get. Modern desktop workstations have the computing power of a supercomputer from a decade or two ago, so within a lifetime we could individually have access to something worth calling intelligent.
Then it isn’t AGI, unless you agree that most people’s thinking skills “just suck.” For which there is an argument. As long as we can try and try and try again, especially in context of pre-existing working examples, then sure GI and therefor AGI may just suck yet average out just fine!
AGI is whatever you want it to be.
Neural-network-based AI has been around since the 50's, and pretty much the entire time it's existed, people have been positing about an AI that could match or outpace humans. It's gone under various terms, like post-singularity AI, strong AI, and AGI.
Over the decades, various generations of AI have advanced enough to fill various niches, but each time the hype far outpaces the abilities, and the goals get diluted down until they are more practical. The term AGI is general enough that it doesn't mean anything, so whatever we achieve with this generation of AI infrastructure we can call AGI.
We're still really far away from getting a neural network that's anywhere near the capability of a reasonably capable vertebrate. The way LLM training works is more like selective breeding than learning. To get a neural network that behaves the same way a your pet dog or bird thinks, each back-and-forth in a conversation would require the equivalent of retraining a foundational model.
It's not something we'll never get. Modern desktop workstations have the computing power of a supercomputer from a decade or two ago, so within a lifetime we could individually have access to something worth calling intelligent.
If it "sucks" at enough important tasks, does calling it AGI actually change anything? Capability matters more than the label.
Then it isn’t AGI, unless you agree that most people’s thinking skills “just suck.” For which there is an argument. As long as we can try and try and try again, especially in context of pre-existing working examples, then sure GI and therefor AGI may just suck yet average out just fine!
We are still talking about a narrative that never arrived.
"AGI" is meaningless.
So just arguing over semantics?
Then you want R (real) AGI?
I mean it was AI before and now AGI...