This is one of my favorite pieces of internet writing, up there with the SR-71 speed check and the Story of Mel. Every time I see it, I have to read it again and end up giggling through the whole thing.
> Marco Pierre White passionately defends chefs using microwaves. White dubbed microwaves “sensational things” and revealed he thinks they’re far better at preparing kippers than any other technique, like boiling or grilling
> José Andrés, a renowned Michelin-starred chef, New York Times bestselling author and internationally recognized humanitarian. He listed the microwave omelet as his number one foolproof dish and called it the “best fluffy omelet in the history of mankind!”
A more accurate analogy is Charles and Henry Greene using tech to construct an intricate rig to fasten the joints in a delicate jewelry box to fit inside the Gamble House. Yes, they could make the rig by hand, but time is a precious resource to people with so much to build.
What Tao and other artists of his caliber are demonstrating is that the tech is capable of building the rig. And the machine makers are incrementally demonstrating that the machine can make not only the jewelry box rig, but rigs to build rig-making machines.
Yep, I'm fairly certain that general learning algorithms acting upon an ANN (which is fairly general too, see the universal approximation theorem) can reach and surpass performance of the human brain. As we have approximately zero evidence that the human brain contains "magic," that is something that can't be modeled by an ANN of a practically feasible size. (I know about chaotic processes that can't be modeled precisely. The "magic" here would be the brain using such a process to make useful decisions. "Useful" not in a sense of a mixed strategiy. You can roll dice to do that.)
But, no, it's not "any day now." The required size and structure of the ANN is to be determined.
I think it's a giant leap of faith to go from "there's no evidence that the human brain contains magic" to "ANNs will achieve AGI and we just need to monkey with the parameters to get there." You've ruled out the supernatural, but don't account for mundane limitations like energy, training data, memory or the architecture. That theorem you referenced tells us a sufficiently wide network can (quoting from Wikipedia) "approximate any continuous function to any desired degree of accuracy" – but it doesn't give us any bound on the sample complexity or the energy required. That is to say, just because the theorem says it's technically possible doesn't mean it's literally possible.
From my POV it's "we haven't found physical processes that allow to do information processing, which is relevant to learning, exponentially faster than digital computers and we aren't likely to find them in the brain."
Or, at least, I don't tend to postulate existence of such mechanisms in the brain after hearing about every empirical ANN deficiency.
BTW, I'm not talking about supernatural. The "magic" could be unknown physical processes, unknown quantum algorithms, unknown physics.
I thought the narrative a few months ago was that we had AGI[1]? What happened? It's not like these trillion dollar companies would have an interest in using AGI to hype up their technologies that they would need to replace a percentage of the global workforce over the next few years to justify their current valuations?