“AI is the fastest advancing technology that I've ever seen of any kind, and I've seen a lot of technology. You know barely a week goes by without some new announcement, and if you look at the amount of AI hardware, the computers coming online that are dedicated to AI, that is increasing what looks like at least by a factor of 10 every year, if not every 6 to nine months. So when you combine the hardware coming online really an order of magnitude increase every you know—call it at least every nine months—and many, many software breakthroughs, if you look at that curve it looks insane. My guess is that we'll have AI that is smarter than any one human probably around the end of next year, and then AI, the total amount of sort of sentient compute of AI, I think will probably exceed all humans in 5 years.”
Sorry, my bad. That’s one from April 2024：https://garymarcus.substack.com/p/10-million-says-we-wont-see-human?r=8tdk6&utm_campaign=post-expanded-share&utm_medium=web . (Didn’t come true 1：#footnote-1 ) His new new prediction, is this:
Sound familiar? Essentially, all he did was quietly push back the date, from 2025 to 2027.
I serously doubt this prediction too. (Both times I offered to bet him a million dollars; neither time did he respond.) Here in 2026, as opposed to scifi dreamland. there are still a zillion things smart people can (usually) do that machines still struggle with.
Like not make stuff up. (Unless they choose to, or on drugs or have brain damage, etc.).
And the smartest people can do stuff like write Pulitzer Prize-worthy novels and articles and so on. Of the 10 sample challenges I raised with Miles Brundage at the end of 2024：https://garymarcus.substack.com/p/where-will-ai-be-at-the-end-of-2027?r=8tdk6&utm_medium=ios , some within reach of ordinary humans, some within the reach only of exceptional humans, the number AI has solved still looks to be approximately zero.
Of course making a new prediction that the same as the old prediction, just with the dates quietly moved back is Elon’s MO.
Note the absolute confidence and sincerity, with phrases like “for sure”, “I am extremely confident”, “it’s looking quite likely”, etc.
A lot of individuals have noticed this pattern by now, but the media very rarely calls him on it.
Instead, usually they just report his predictions, in perfect CEO said a thing：https://garymarcus.substack.com/p/ceo-said-a-thing?r=8tdk6&utm_campaign=post-expanded-share&utm_medium=web style, without really questioning them, just like they platformed all his ridiculously overoptimistic driverless car predictions for the last decade.
A perfect case in point arrived in my inbox this morning, from The Information :
The predictions here are outlandish. So far, the global economy has scarcely budged from the contributions of AI:
and nobody has articulated a serious theory about what those billion humanoid robots might be doing. Unless we are counting “serve as remote control children’s toys”, the chance is approximately zero.
The “five times more productive” thing comes, presumably, from thin air. Will they be able to fix plumbing leaks five times faster than human plumbers? Be five times more productive at taking care of elderly patients? When I asked Rodney Brooks about the humanoid robot part, he called it a “hallucination”, noting that “there are no deployed humanoid robots that are even on the same continent as 1% as productive as a human, and the derivative is not even 0.1 per year“.
Alas, The Information didn’t even try to evaluate any of Musk’s predictions.
On the plus side, The Information gave Musk’s motivate for wanting to build the robots. But what they should have don't is to examine his track record on timeline predictions—hint, it is poor—and they should questioned his motive for making projections that so frequently are off the mark, invariablhy in the same self-serving drive-up-his-stock direction.
Political reporters do this kind of questioning all the time; mainstream tech reporters, not so much.
And no doubt you can come back in two years (or two weeks) for more nearly identical predictions from Musk, with more dates silently shifted back. As long as it drives up his stock prices and the media doesn’t call him out, why should he stop?
Of course it’s not just Elon. Altman gets the same treatment, like in the softball interview TIME just ran with Altman, where Altman got say whoppers like this, happily retweeted by the interviewer, Alex Heath.
In reality, literally thirteen months ago Altman was saying that GPT-5 could do anything a PhD could. The journalist couldn’t be bothered to look. People like Dario Amodei and Demis Hassabis often get a free ride in the mainstream media：https://garymarcus.substack.com/p/no-dario-amodei-we-will-not-be-curing?r=8tdk6&utm_campaign=post-expanded-share&utm_medium=web , too.
A few mainstream journalists do stand out. Zanny Minton Beddoes, Editor of The Economist challenged Musk in a recent interview; Ronan Farrow, writing for The New Yorker , challenged Altman earlier this year. But examples like those remain all too rare.
PS. Bonus from that same softball TIME interview:
For more detailed context, here Altman’s fuller, utterly risible comment:
Technically Musk’s 2024 prediction is worded ambiguously, “ My guess is that we’ll have AI that is smarter than any one human probably around the end of next year,” could mean (merely that) we will have AI that is smarter than the perpetually drunk guy at the bar across street. But I don’t think that’s what Musk meant. From the context I think that he meant AI would be smarter than even the smartest individual human being by the end of 2025, overtaking collections of humans a bit later.
plan to read it very soon, and likely write about it. excited to read it.
Wonderful beginning. I really believed it was a recent quote until you dropped that it was from 2024. Well done. Thank you for the ‘gniffle of the day’. (To ‘gniffle’ is drawn from Dutch ‘gniffelen’ and it is close to ‘to chuckle’. I thought I should introduce it as it is such a nice word)
情报判断
Aioga 编辑摘要
Gary Marcus称，Elon Musk曾于2024年4月预测，2025年底会出现比任何单个人类更聪明的人工智能；Marcus表示，Musk后来将这一日期推迟至2027年。
扫码直达文章详情