I have no frame of reference as to what 100,000 GPUs is capable of doing. Let alone 350,000 GPUs. Then what? A million? 10 Million? Is there an upper limit? At what point do humans become superfluous?
Yeah, that was pretty much my point.
This is going to get a bit long, so forgive me, BUT....
IMO, the article you linked is hyperbolic click-bait that throws around a lot of big numbers and insider jargon that most people don't really understand - in an attempt to get people all worked up.
GPUs are computer chips that are specifically designed and optimized to process images. Your laptop or desktop computer has one to enhance the speed and quality of images displayed on your screen.
The graphics power of a GPU can also be used to do the opposite - to "look" at and analyze graphics and images - in order to extract information from them. Facial recognition systems are one example of how this works.
As I understand what I read in the article you linked, it is basically saying that Musk is building a new supercomputer - built with a lot of GPUs. The idea is to create a computer that is focused on scanning and analyzing digital images, and the goal is for this new xAI to be able to learn to analyze, interpret, and extract data from the vast number of digital images that are already available online.
In other words, Elon Musk is trying to build an AI computer that can "learn" from looking at pictures. This is the next logical step beyond the current generation of AI computers that can only "read", analyze, and "learn" from WRITTEN materials online.
FWIW, as I see it, that article is just an attempt to get attention (and "clicks" and ad revenue) by crying WOLF regarding advances in AI.
Don't get me wrong, I'm just as leery of the potential pitfalls of a TRUE, self-aware, artificial intelligence as anyone. But IMO, what is now being called AI is just a faster, more powerful search engine with more data available to search. It is still a LONG way from being a self-aware, sentient, intelligence.
I recently read that ChatGP is returning completely nonsensical results because a part of how it is formulating its responses are based on tallying the number of "upvotes" (a.k.a. the number of "likes") for questions posted on sites like Quora and Google.
One example I read about was a question about the best ingredients to make marinara spaghetti sauce. When someone asked that question, someone else posted a completely nonsensical tongue-in-cheek response suggesting that gasoline would be a good ingredient to add to the sauce.
Then a lot of people "upvoted" the suggestion to add gasoline - because they found it amusing.
But ChatGP didn't recognize the humor that was being injected into the conversation. It took the "upvotes" at face value and because it was a popular answer, it began recommending gasoline as an ingredient in marinara spaghetti sauce.
Is that true intelligence - real or artificial? I don't think so. It is just another example of the GIGO principle. GIGO=Garbage In Garbage Out.
A computer "intelligence" is only as "smart" as the info it's programmers are feeding into it, or allowing it to access.
Adding the ability to search and interpret digital images (in addition to digital written data) may be a step towards a true artificial intelligence, but in reality it is just a more sophisticated form of search engine. True computer intelligence is still a long ways off - assuming it is even possible to begin with.
I kinda doubt that it is possible. How can the created be greater than the creator? Seems illogical to me. But it makes good Sci-Fi and creates fear - which is good for the 24-hour news cycle media circus.
JMO, and YMMV.
