Too much, too fast, too faulty, that is AI in a nutshell. Nothing pushed out this fast has ever had good results. There needs to be stiff regulations, comprehensive standards, because they are not being transparent or honest about what they are doing. It seems like another fools gold epic rush to the bottom. I don’t want to be anywhere near any of this.
It really isn't that complicated. They are pushing new cars off an assembly line with no testing, no safety regulations, forcing everyone to pay the price of a 747 for each including stealing your electricitiy and water to power them, whether we want one or not, while we are left sitting at home with our old Chevettes wondering what the heck just happened.
Apparently in a recent advertisment- Jason Kelse has come up with a solution to the water demand issues....Thank you John for breaking this down and explaining it...a lot of politicians are changing course as the public outrage is growing against all the data centers being built in rural communities...even in Texas the Gov. has put a pause on things....
John you underestimate the power requirements of current AI build outs. They are in the gigawatt range, and that is how the centers are described, i.e. 3, 4, 5 gigawatt or larger. More power than some states consume at any one instant.
John, thank you! Your explanation of data centers ’ abilities and deficiencies is 5th grade simple and I learned a lot. However a great way to “grow a country” is to setup data centers in remote areas that are not developed or populated. In other words build small towns/cities around the center to support the center and stop stealing resources from established residential communities. I know the centers’ billionaire owners may feel theft equals increased investors’ profits but infinite lawsuits eventually decrease those profits especially as AI ‘new’ rapidly becomes 24 hour dinosaurs. But, perhaps I give billionaires too much pioneering intelligence or just too much intelligence, period.
Now regarding T, SCOTUS and mail voting ballots. I see a glaring deficiency in Trump’s SCOTUS/Robert’s Court rulings. The deficiency is the word “legitimacy/legality/legal.” Trump’s SCOTUS does not question the legal standing, per the Constitution, of Trump’s appeals. For the last 18 months the rulings give “limited power” and “for now power” to Trump. The Court simply rules in favor of T’s appeals and damned be the rest of America!
I realize we are facing growing pains as a country and that T was allowed by the King of Kings to be in the Oval for many reasons. I firmly believe the King of Kings always has one more move and we the people are that one more move. I also realize that T’s type has never held the Presidency; several have been close but no cigar. I hope America realizes that we may need another Constitutional Convention to push amendments that will severely limit the President’s power. That is unfortunate and I hope we do not resort to “knee jerk actions” that will quickly become regrettable responses.
Felix Salmon on Slate Money last weekend interviewed an AI "journalist(?)" who introduced me to a bunch of technical stuff about AI and its deployment and use. Worth listening to. The market for AI is largely companies. Their hesitation and ignorance often sub-optimize and slow down the useful features.
A related question that has been nagging me: Is there a fundamental innovation that has enabled all of this? What will be the modern equivalent of the steam turbine for Industrial revolution when we look back on the AI explosion?
All these building blocks and bottlenecks here are not new devices, we have had chips and processors for decades. Even the concept of a Large Language Model is an idea that has been around. Is it just the gradual improvement of processing chips that has made the scale needed for AI feasible? Or is there some math behind the networking or training that was the breakthrough? Could you have built a large language model on the computer chips of the 2000s if you had enough of them, albeit a slower one?
I suppose the reduction of technology to a breakthrough innovation is itself an invention of history. We had boilers and coal furnaces and the other components of steam turbines before we realized we could put them together to power all kinds of things. But still, the idea of a catalyzing invention is somehow satisfying, and AI for the most part, does not clearly have one.
Putting a key on a kite string was a research innovation. Halving processing chip size every two years or so is a research feature. If anything, the "breakthrough" may have been enlistment of the graphics chip in the fundamental processing.
AI remains an attempt to mimic thinking. Things it appears to fabricate derive from training. Things it correctly processes derive from training. Even if AI's technical development were to stop today, it cannot likely scale to general use absent energy and materials innovations that are beyond earth's ability to provide. While speculation continues that its further development relies on decimating the human race, the scale of such an effort is its own Achilles heel.
Thank you. Timely. wrote a letter last week to an AI advocate/trainer former colleague asking these questions! My concern is people like him are not also training around capacity. So if everyone buys in and push back wins, what then? Who gets use cut first ? No answer from him. This is so helpful!!!!
This column is a real public service. Thanks for doing the heavy lifting.
Great questions, John, and they all need to have positive answers before moving forward... not just some of them. Thank you for outlining the scope.
Glad you put this out for everyone to see and realize what it really means and takes. Wonderful insights and questions.
Too much, too fast, too faulty, that is AI in a nutshell. Nothing pushed out this fast has ever had good results. There needs to be stiff regulations, comprehensive standards, because they are not being transparent or honest about what they are doing. It seems like another fools gold epic rush to the bottom. I don’t want to be anywhere near any of this.
And perhaps the most important thing not mentioned...REGULATION. There is none.
Way too complicated for me!
It really isn't that complicated. They are pushing new cars off an assembly line with no testing, no safety regulations, forcing everyone to pay the price of a 747 for each including stealing your electricitiy and water to power them, whether we want one or not, while we are left sitting at home with our old Chevettes wondering what the heck just happened.
Apparently in a recent advertisment- Jason Kelse has come up with a solution to the water demand issues....Thank you John for breaking this down and explaining it...a lot of politicians are changing course as the public outrage is growing against all the data centers being built in rural communities...even in Texas the Gov. has put a pause on things....
John you underestimate the power requirements of current AI build outs. They are in the gigawatt range, and that is how the centers are described, i.e. 3, 4, 5 gigawatt or larger. More power than some states consume at any one instant.
John, thank you! Your explanation of data centers ’ abilities and deficiencies is 5th grade simple and I learned a lot. However a great way to “grow a country” is to setup data centers in remote areas that are not developed or populated. In other words build small towns/cities around the center to support the center and stop stealing resources from established residential communities. I know the centers’ billionaire owners may feel theft equals increased investors’ profits but infinite lawsuits eventually decrease those profits especially as AI ‘new’ rapidly becomes 24 hour dinosaurs. But, perhaps I give billionaires too much pioneering intelligence or just too much intelligence, period.
Now regarding T, SCOTUS and mail voting ballots. I see a glaring deficiency in Trump’s SCOTUS/Robert’s Court rulings. The deficiency is the word “legitimacy/legality/legal.” Trump’s SCOTUS does not question the legal standing, per the Constitution, of Trump’s appeals. For the last 18 months the rulings give “limited power” and “for now power” to Trump. The Court simply rules in favor of T’s appeals and damned be the rest of America!
I realize we are facing growing pains as a country and that T was allowed by the King of Kings to be in the Oval for many reasons. I firmly believe the King of Kings always has one more move and we the people are that one more move. I also realize that T’s type has never held the Presidency; several have been close but no cigar. I hope America realizes that we may need another Constitutional Convention to push amendments that will severely limit the President’s power. That is unfortunate and I hope we do not resort to “knee jerk actions” that will quickly become regrettable responses.
Thanks. Now my BINGO card is full. Does anyone in the current administration understand any of this?
Felix Salmon on Slate Money last weekend interviewed an AI "journalist(?)" who introduced me to a bunch of technical stuff about AI and its deployment and use. Worth listening to. The market for AI is largely companies. Their hesitation and ignorance often sub-optimize and slow down the useful features.
A related question that has been nagging me: Is there a fundamental innovation that has enabled all of this? What will be the modern equivalent of the steam turbine for Industrial revolution when we look back on the AI explosion?
All these building blocks and bottlenecks here are not new devices, we have had chips and processors for decades. Even the concept of a Large Language Model is an idea that has been around. Is it just the gradual improvement of processing chips that has made the scale needed for AI feasible? Or is there some math behind the networking or training that was the breakthrough? Could you have built a large language model on the computer chips of the 2000s if you had enough of them, albeit a slower one?
I suppose the reduction of technology to a breakthrough innovation is itself an invention of history. We had boilers and coal furnaces and the other components of steam turbines before we realized we could put them together to power all kinds of things. But still, the idea of a catalyzing invention is somehow satisfying, and AI for the most part, does not clearly have one.
Putting a key on a kite string was a research innovation. Halving processing chip size every two years or so is a research feature. If anything, the "breakthrough" may have been enlistment of the graphics chip in the fundamental processing.
AI remains an attempt to mimic thinking. Things it appears to fabricate derive from training. Things it correctly processes derive from training. Even if AI's technical development were to stop today, it cannot likely scale to general use absent energy and materials innovations that are beyond earth's ability to provide. While speculation continues that its further development relies on decimating the human race, the scale of such an effort is its own Achilles heel.
Great piece. Appreciate all your work and explaining. It’s all moving too fast to soon
I suggest “Empire of AI” by Karen Hao.
Thank you. Timely. wrote a letter last week to an AI advocate/trainer former colleague asking these questions! My concern is people like him are not also training around capacity. So if everyone buys in and push back wins, what then? Who gets use cut first ? No answer from him. This is so helpful!!!!
No.