First off: What is Tegmark suggesting his colleagues should do to "prepare" for the notion of *every single human expert being replaced* in two years? Even if that is real (it's not), it's akin to "preparing" for the Earth to be hurtled into the Sun.
Second off: How is the media this bad? Why do the NYT, WSJ, Politico, Vox, The Guardian, the LA Times, WaPO, NONE OF THEM, ever push back on these ridiculous claims? Why is the follow up question to "it'll be smarter than Nobel Laureates by 2027" not "You sure? Because right now it can barely add two numbers together." These guys can say anything and get no pushback whatsoever. We live in an era where Dario Amodei can write a blog post claiming we'll have escaped the throes of mortality itself by 2030 and the prevailing narrative in the press is that he's a genius and a prophet instead of a charlatan or a lunatic.
The graph made me actually snort laugh. This hype is akin to the guy who invented the bicycle insisting he could have an interstellar spaceship within two years of the bicycle. I keep saying before we start talking about curing all diseases, let's start with literally more than zero.
Seriously, serving this anti-mRNA vax b*llsh*t in a blog that is mostly scientists and engineers?
Ms. Zetty - I worked with an ER group during Covid. The waves were like Omaha beach. People would come in, even whole families, and die before they could get them admitted, right there in the ER, on the floor one or 2 every few minutes.
One night at 2 AM my friend/colleague called. "A family of 7. One survivor. A little kid." I hear him sobbing. This is a guy who turned in his own father after watching him kill his mother when he was 12. He put himself through college and med school by grit and determination. "Mom, Dad, grandparents, siblings, all dead. Just one kid." He sobs more. I hear a voice in the background and sounds of people running metal noises. "Gotta go." click.
That doctor worked straight for a week at a time during covid. I made a vaccine and got it to him by March 2020. He and his group survived volunteering for the worst, highest mortality duty, which was intubations.
It took a year and a half before the big wave hit Idaho. People died in their cars down the road from here, waiting for help. Swamped.
All through covid, there were refrigerator trucks backed up where they stacked bodies like cordwood. After vaccines became available, doctors and nurses would hear patients demanding through their pneumonia coughing, "Please. Record me! I'm dying. Let me tell them to get the vaccine!" But it wasn't allowed. They died like flies. For nothing, because they believed some grifting sack of lying sh*t. They died for nothing.
I am talking about people JUST like you. People that came into the ER talking sh*t about vaccines being from Satan and god knows what else fanatic lies they had been told. People like you, dying from their decision, frantic in their last hours to tell the world, "NO! I WAS WRONG! GET THE VACCINE!"
I'll repeat then. The reason the press has failed is that from 2000 onward, "the press" was turned into a pay to play bribe festival, with journalists and editors benefiting. This is together with influence wielded across media platforms by venture capital with friends and direct ownership in media conglomerates wanting to hype its money into at least break even. The little essay comment I wrote a week or two back on "Dead fish and Dead Whales" applies. Uber (a dead whale) which never, ever, provided a written plan for profitability was nevertheless a success for VCs (at least those who sold out).
You see it across the board, from NYT to Matt Taibbi's blog where he tells the stories that his followers want to hear. Bari Weiss is doing it at The Free Press now. Feeds the kids, see. Sends the kids to college. Makes them lotsa moolah.
There are days I wonder if I could live with myself if I sold out. I could make a few million easy. But, that would be grifting. No. I can't. I just can't do it.
The press think that tech CEOs are the foremost experts on anything to do with AI. And this is doubly wrong: from hearing them talk, they don't seem to have that great an understanding of their own technology (or they do and they choose to misrepresent it). And they sure as hell don't know anything about how well their technology approximates human intelligence! Even if these guys were all experts on the technical aspects of deep learning models, they're out there making implicit claims about human minds. It is appalling that major media outlets aren't interested in talking to anyone with expertise in this area.
I just did a bit of checking on this, and the number of radiologists in the United States INCREASED from 37,482 to 49,070 (25% growth) from 2020 to 2023:
Love this. I’ve long been amazed that journalists just love to promote wild proclamations of the demise of human creativity and real science yet fail to fact check as you have here. Love it and I’d like to read more debunking along these lines. The robo cars example, for example, is partially being solved (not by Musk) — Waymo reports 200K rides a week, yet only in highly constrained and trained cities. But progress is being made there.
The radiologist example is a great one-they’re alive and well as a profession by most accounts.
I just did a bit of checking on radiologists, and the number of radiologists in the United States INCREASED from 37,482 to 49,070 (25% growth) from 2020 to 2023:
I just went on Indeed and searched for radiology job openings where I live (Birmingham, Alabama.) Within 15 miles of Birmingham there are about 30 job postings looking for radiologists.
I consult for law firms regarding AI, and it's no threat to attorney jobs. Maybe paralegals. It can't even effectively do legal research because so much key data is behind paywalls. Limited ability to analyze a case, and certainly flames out with any nuanced issues. They can't replace trial attorneys. Can't do the basic stuff us silly fleshbags can.
And the fact that lawyers cannot be replaced is truly terrible. Please lord, if there can be an exception to AI not being capable, let it be attorneys. Attorneys are so awful most of the time.
Hi Gary... omg. Such absurd statements show how dumb the utterers are.
"Science" is not about curve-fitting, "it turns out". ALL science is based on lived experience - we creatures, being of this world, are curious about how it (the world) works - our math, theories, equipment, instruments, measurements, hypotheses, are ALL about that, nothing less, nothing more. Our senses limit what we can perceive, so/but we build microscopes and telescopes, and sensors of every kind, to transcend direct experience [which is still the basis for it all].
Given the above, glorified dot product calculator algorithms will do... what again?
It's stunning that people who ostensibly understand how the LLM illusion of intelligence works under the hood can still fall for the illusion. It would be like watching a stage magician go though exactly how a magic trick is performed backstage, and then sitting in the audience and insisting it's really magic.
You're right, but in this case the trick is so complicated that one can't understand it in detail, and the results are so impressive, compared to anything computers have been able to do before, as all to make it very easy for people to see what they want to see.
That's a fair point. I'll push back mildly on the trick being far too complicated to understand... there's this "even the people who make the models don't understand how they work" talking point that I find pretty misleading and that adds to the hype, as though deep learning engineers are unaware that they're working on statistical next-token prediction models. But I do agree that a lay person is going understand Penn Jillette's demonstration for how he knows which card you picked far more easily than Steve Wolfram's demostration for how to statistically predict the next token.
AI needs to be able to run a lab first. Do these people know how science works? It is rarely one person working alone. It needs also, “a mind of its own”. Think about it.
That's the only reaction I'm left with after yet another such report. Sigh.
The sad thing is that this tech will 'convince' way beyond what it is actually capable of. So, it is doubtful the hype will crash as soon as it could. There is still a lot of life in this hype.
I had to LOL that the AI image generator hallucinated a second Elon Musk. So on brand for Gary Marcus to expose some AI slop ... (tin-foil hat on) ... Or is it?
Now, it all makes sense! I never could figure out how Elon has time to SpaceX and Tesla *and* hang out at the WH *and* post incessantly on X pushing his own political ravings. Never mind the 14 kids. who the h8ll has time to do all that?!? There's only one answer: The genius entrepreneur actually has his own real evil twin! (Hint: The evil one is the one w the chainsaw; that's movie villain 101.)
Truer words have not been said. I checked Grok3's capacity to draw a full glass of wine. It is about as resistant to this as it is to drawing a 2-D version of the Dalai Lama and putting that into a Dali flowing watch style painting. MOAR COMPUTE! Slower, for the same total lack of comprehension and repetitive wrong. I blame guardrails for the immunity to change.
Also, I thought that was Leroy Jethro Gibbs next to Elon Musk, but maybe it's just Hinton at Gibbs.
Can it draw a TV sitting on top of a cat? The Stable Diffusion models still can't. Explain it in as much detail as you like, it'll always have the cat sitting on top of the TV.
That's what singularitarians actually believe: one day there will be an AI, and the next day it will have improved itself by orders of magnitude, and within weeks, god-like super-AI will have solved all scientific and engineering problems in the world. For anybody who understands how the world actually works, it is extremely difficult to grasp this mindset, but that is really, actually what these guys believe in their cultish bubble.
Singularitarians do not understand trade-offs, and they do not understand diminishing returns. They do not understand, or at least pretend not to understand, that most of our largest problems are political - we already know the technical solution but lack the will to implement it. They also never understand, or pretend not to understand, that science and engineering problems cannot be solved without experiments, prototype testing, and trials, which take place in real time and depend on laboratories, glass houses, field trips, telescopes, particle accelerators, reactors, and workshops. This is why even if god-like super-AI were available in 2027, all it could do was suggest various hypotheses and then request its share of limited research funding to test them over the following years.
And they're all shut ins who think in singularity world they'll get everything for free and not have to work, which tells you all you need to know about their grasp on reality
First off: What is Tegmark suggesting his colleagues should do to "prepare" for the notion of *every single human expert being replaced* in two years? Even if that is real (it's not), it's akin to "preparing" for the Earth to be hurtled into the Sun.
Second off: How is the media this bad? Why do the NYT, WSJ, Politico, Vox, The Guardian, the LA Times, WaPO, NONE OF THEM, ever push back on these ridiculous claims? Why is the follow up question to "it'll be smarter than Nobel Laureates by 2027" not "You sure? Because right now it can barely add two numbers together." These guys can say anything and get no pushback whatsoever. We live in an era where Dario Amodei can write a blog post claiming we'll have escaped the throes of mortality itself by 2030 and the prevailing narrative in the press is that he's a genius and a prophet instead of a charlatan or a lunatic.
yeah the press has really failed us here. that is absolutely true.
The graph made me actually snort laugh. This hype is akin to the guy who invented the bicycle insisting he could have an interstellar spaceship within two years of the bicycle. I keep saying before we start talking about curing all diseases, let's start with literally more than zero.
At which point, I would remind the great AI hypsters of the Polish proverb:
> "Never attempt to cure [or prevent] what you don't understand."
. . . . mRNA "vaccines" being a case in point.
mRNA vaccines work. So do DNA vaccines. So do component vaccines. So do killed virus vaccines.
> "Never attempt to bloviate about what you don't understand."
Metta whoever you are:
Seriously, serving this anti-mRNA vax b*llsh*t in a blog that is mostly scientists and engineers?
Ms. Zetty - I worked with an ER group during Covid. The waves were like Omaha beach. People would come in, even whole families, and die before they could get them admitted, right there in the ER, on the floor one or 2 every few minutes.
One night at 2 AM my friend/colleague called. "A family of 7. One survivor. A little kid." I hear him sobbing. This is a guy who turned in his own father after watching him kill his mother when he was 12. He put himself through college and med school by grit and determination. "Mom, Dad, grandparents, siblings, all dead. Just one kid." He sobs more. I hear a voice in the background and sounds of people running metal noises. "Gotta go." click.
That doctor worked straight for a week at a time during covid. I made a vaccine and got it to him by March 2020. He and his group survived volunteering for the worst, highest mortality duty, which was intubations.
It took a year and a half before the big wave hit Idaho. People died in their cars down the road from here, waiting for help. Swamped.
All through covid, there were refrigerator trucks backed up where they stacked bodies like cordwood. After vaccines became available, doctors and nurses would hear patients demanding through their pneumonia coughing, "Please. Record me! I'm dying. Let me tell them to get the vaccine!" But it wasn't allowed. They died like flies. For nothing, because they believed some grifting sack of lying sh*t. They died for nothing.
I am talking about people JUST like you. People that came into the ER talking sh*t about vaccines being from Satan and god knows what else fanatic lies they had been told. People like you, dying from their decision, frantic in their last hours to tell the world, "NO! I WAS WRONG! GET THE VACCINE!"
As if we don't understand viruses.
I'll repeat then. The reason the press has failed is that from 2000 onward, "the press" was turned into a pay to play bribe festival, with journalists and editors benefiting. This is together with influence wielded across media platforms by venture capital with friends and direct ownership in media conglomerates wanting to hype its money into at least break even. The little essay comment I wrote a week or two back on "Dead fish and Dead Whales" applies. Uber (a dead whale) which never, ever, provided a written plan for profitability was nevertheless a success for VCs (at least those who sold out).
You see it across the board, from NYT to Matt Taibbi's blog where he tells the stories that his followers want to hear. Bari Weiss is doing it at The Free Press now. Feeds the kids, see. Sends the kids to college. Makes them lotsa moolah.
There are days I wonder if I could live with myself if I sold out. I could make a few million easy. But, that would be grifting. No. I can't. I just can't do it.
So true. Are we really surprised that the paper owned by Jeff Bezos, for instance, isn't telling the truth about corporate hype?
The press think that tech CEOs are the foremost experts on anything to do with AI. And this is doubly wrong: from hearing them talk, they don't seem to have that great an understanding of their own technology (or they do and they choose to misrepresent it). And they sure as hell don't know anything about how well their technology approximates human intelligence! Even if these guys were all experts on the technical aspects of deep learning models, they're out there making implicit claims about human minds. It is appalling that major media outlets aren't interested in talking to anyone with expertise in this area.
And to think that twenty years ago, Tegmark was a good scientist.
"Detection of the Baryon Acoustic Peak in the Large-Scale Correlation Function of SDSS Luminous Red Galaxies"
https://iopscience.iop.org/article/10.1086/466512
As a radiologist, their comment disgusts me. As a retired radiologist doing editing of radiology papers, their comment disgusts me again.
I just did a bit of checking on this, and the number of radiologists in the United States INCREASED from 37,482 to 49,070 (25% growth) from 2020 to 2023:
https://medicalxpress.com/news/2025-02-current-radiologist-shortage-persist.html?utm_source=perplexity
Rightly so, Libby!
We are needed!
As someone recovering from interstitial pneumonia, allow me to say "THANK YOU!" Radiologists rock. And save lives.
Love this. I’ve long been amazed that journalists just love to promote wild proclamations of the demise of human creativity and real science yet fail to fact check as you have here. Love it and I’d like to read more debunking along these lines. The robo cars example, for example, is partially being solved (not by Musk) — Waymo reports 200K rides a week, yet only in highly constrained and trained cities. But progress is being made there.
The radiologist example is a great one-they’re alive and well as a profession by most accounts.
I just did a bit of checking on radiologists, and the number of radiologists in the United States INCREASED from 37,482 to 49,070 (25% growth) from 2020 to 2023:
https://medicalxpress.com/news/2025-02-current-radiologist-shortage-persist.html?utm_source=perplexity
I just went on Indeed and searched for radiology job openings where I live (Birmingham, Alabama.) Within 15 miles of Birmingham there are about 30 job postings looking for radiologists.
HUMAN radiologists? Haha.
It is staggering to see how a field initially aimed at understanding 'intelligence' has descended into ignorance and irrationality.
100% ~ AI is dumb:
> https://bra.in/4pRx4x
I consult for law firms regarding AI, and it's no threat to attorney jobs. Maybe paralegals. It can't even effectively do legal research because so much key data is behind paywalls. Limited ability to analyze a case, and certainly flames out with any nuanced issues. They can't replace trial attorneys. Can't do the basic stuff us silly fleshbags can.
Plus, anyone familiar with the legal field knows it's one of the most conservative and slow-to-adapt professions out there.
And the fact that lawyers cannot be replaced is truly terrible. Please lord, if there can be an exception to AI not being capable, let it be attorneys. Attorneys are so awful most of the time.
You’re having fun again - rightfully so… 👍🏿
Every enterprise suffers with hucksterism, and AI is proving in turn its lack of immunity...plus ça change...etc.
The Hinton and AI hype must switch to another AI model they don't even believe exists
http://aicyc.org/2025/02/16/yes-prof-hinton-there-is-a-symbolic-ai/
True, although Cyc was the poster child of symbolic AI hype - bogosity is measured in microLenats after all.
The CYC hype began 40 years ago as a response to Japanese AI threat. The failure was trying to code knowledge by hand. A fools errand.
Hi Gary... omg. Such absurd statements show how dumb the utterers are.
"Science" is not about curve-fitting, "it turns out". ALL science is based on lived experience - we creatures, being of this world, are curious about how it (the world) works - our math, theories, equipment, instruments, measurements, hypotheses, are ALL about that, nothing less, nothing more. Our senses limit what we can perceive, so/but we build microscopes and telescopes, and sensors of every kind, to transcend direct experience [which is still the basis for it all].
Given the above, glorified dot product calculator algorithms will do... what again?
It's stunning that people who ostensibly understand how the LLM illusion of intelligence works under the hood can still fall for the illusion. It would be like watching a stage magician go though exactly how a magic trick is performed backstage, and then sitting in the audience and insisting it's really magic.
You're right, but in this case the trick is so complicated that one can't understand it in detail, and the results are so impressive, compared to anything computers have been able to do before, as all to make it very easy for people to see what they want to see.
They know there is a trick involved, even though they don’t understand precisely how the trick works.
I would liken the case of LLMs to a card “trick” that appears to involve magic but actually works out by pure mathematics.
Not incidentally, in science, there is another word we use for “trick”.
That's a fair point. I'll push back mildly on the trick being far too complicated to understand... there's this "even the people who make the models don't understand how they work" talking point that I find pretty misleading and that adds to the hype, as though deep learning engineers are unaware that they're working on statistical next-token prediction models. But I do agree that a lay person is going understand Penn Jillette's demonstration for how he knows which card you picked far more easily than Steve Wolfram's demostration for how to statistically predict the next token.
AI needs to be able to run a lab first. Do these people know how science works? It is rarely one person working alone. It needs also, “a mind of its own”. Think about it.
Sigh.
That's the only reaction I'm left with after yet another such report. Sigh.
The sad thing is that this tech will 'convince' way beyond what it is actually capable of. So, it is doubtful the hype will crash as soon as it could. There is still a lot of life in this hype.
As a software engineer you are really getting used to the fact that the last 5% in any project always taken the longest.
I had to LOL that the AI image generator hallucinated a second Elon Musk. So on brand for Gary Marcus to expose some AI slop ... (tin-foil hat on) ... Or is it?
Now, it all makes sense! I never could figure out how Elon has time to SpaceX and Tesla *and* hang out at the WH *and* post incessantly on X pushing his own political ravings. Never mind the 14 kids. who the h8ll has time to do all that?!? There's only one answer: The genius entrepreneur actually has his own real evil twin! (Hint: The evil one is the one w the chainsaw; that's movie villain 101.)
No one has time to do all of those things *well*. To do a half-assed, slipshod job at them, however, doesn't require much time at all.
Dontcha worry Gary. You're safe from the AI Hall of Shame.
Truer words have not been said. I checked Grok3's capacity to draw a full glass of wine. It is about as resistant to this as it is to drawing a 2-D version of the Dalai Lama and putting that into a Dali flowing watch style painting. MOAR COMPUTE! Slower, for the same total lack of comprehension and repetitive wrong. I blame guardrails for the immunity to change.
Also, I thought that was Leroy Jethro Gibbs next to Elon Musk, but maybe it's just Hinton at Gibbs.
Can it draw a TV sitting on top of a cat? The Stable Diffusion models still can't. Explain it in as much detail as you like, it'll always have the cat sitting on top of the TV.
Nope. Cat sitting in front of. Lying near. Cat sitting on a table in front of a TV with what appears to be a squashed cat rug under the tabletop.
That's what singularitarians actually believe: one day there will be an AI, and the next day it will have improved itself by orders of magnitude, and within weeks, god-like super-AI will have solved all scientific and engineering problems in the world. For anybody who understands how the world actually works, it is extremely difficult to grasp this mindset, but that is really, actually what these guys believe in their cultish bubble.
Singularitarians do not understand trade-offs, and they do not understand diminishing returns. They do not understand, or at least pretend not to understand, that most of our largest problems are political - we already know the technical solution but lack the will to implement it. They also never understand, or pretend not to understand, that science and engineering problems cannot be solved without experiments, prototype testing, and trials, which take place in real time and depend on laboratories, glass houses, field trips, telescopes, particle accelerators, reactors, and workshops. This is why even if god-like super-AI were available in 2027, all it could do was suggest various hypotheses and then request its share of limited research funding to test them over the following years.
Quite unlike the AGIs that they believe in, the singularitans are only capable of singular thought
And they're all shut ins who think in singularity world they'll get everything for free and not have to work, which tells you all you need to know about their grasp on reality