Jensen Huang to college grads: "Run. Don't walk" toward AI

https://www.axios.com/2026/05/...

Nvidia founder and CEO Jensen Huang told graduates at Carnegie Mellon University in Pittsburgh yesterday that demand for AI infrastructure is creating a "once-in-a-generation opportunity to reindustrialize America and restore the nation's capacity to build."

Why it matters: With many college grads fearing AI could obliterate their career dreams, Huang pointed to boundless opportunity as a "new industry is being born. A new era of science and discovery is beginning ... I cannot imagine a more exciting time to begin your life's work."

Nvidia, which makes AI chips, is the world's most valuable company. Huang told 5,800 recipients of undergraduate and graduate degrees that the AI buildout will require plumbers, electricians, ironworkers, and builders for chip factories, data centers and advanced manufacturing facilities.

"No generation has entered the world with more powerful tools â" or greater opportunities â" than you," he said. "We are all standing at the same starting line. This is your moment to help shape what comes next. So run. Don't walk."

"Every major technological revolution in history created fear alongside opportunity," Huang added. "When society engages technology openly, responsibly, and optimistically, we expand human potential far more than we diminish it."

Full speech: https://www.youtube.com/watch?...

It's not that. LLMs reproduce an output of consciousness, but they way they do so isn't fundamentally any different than a tape recorder or even a book. It's a deterministic process that we can fully reproduce by doing calculations on a piece of paper.

It's not that there's some "magic" in our brains, but there's obviously a very complex process at work that we don't understand. It's also true that the "neural networks" used to run LLMs have only the most superficial similarity to actual brains. Just because LLMs can produce similar reasoning it doesn't mean they're suddenly able to produce other second order effects.

Is it possible that LLMs reproduce this process? We can't authoritatively say no if we don't understand the process. But that's no different from saying a rock way also be conscious.

Extraordinary claims require extraordinary evidence, and Dawkins doesn't have any.

Nah, there's more than enough smoke to conclude that Altman's style rubs a lot of people the wrong way. Whether you think that's a problem or a virtue is another matter.

...like "Tell me about Tiananmen Square" or "Tell me about Xinjiang".

Is this what you want for the future?

My thoughts back when R1 came out:

https://www.afcea.org/signal-m...

Early users testing the service have touted competitive perks, including 3% cash back on eligible purchases

That's about the same as merchant fees, meaning the transactions themselves are probably a loss leader once you factor in all the stuff like fraud and dispute resolution.

I'm guessing Musk sees the real value as the dataset of shopping behaviours. Either that or that cash back rate is going to plummet once the service gets established.

Not really, the tech companies have been doing this for years.

They hire a bunch of folks at high salaries, but not all of those work out, and managers hate laying folks off.

So they make big across-the-board cuts and now everybody from top to bottom is forced to make a bunch of tough decisions about who to cut.

You don't get rid of the worst 10% of your work force, but on average, the 10% you lose is less valuable than the 90% you keep.

And then you go hire some more.

No, they're arguing there's ways to use their software to commit an illegal act, which is true of literally anything.

I can't imagine anyone making the argument that using AI tools to rewrite code in another language removes the copyright.

Not exactly, because the amount of stearates that came off the gloves would be fairly random, so there's no way to apply a general correction. You might not even know what kind of gloves they used in the experiment!

That doesn't mean you throw out the results, but you maybe mark those results and say there was potential factor unaccounted for and the results needs to be replicated.

The gold-standard tasteful way to handle stuff like this is cast a friend of the original actor. An example is Jon Lovitz replacing Phil Hartman on Newsradio.

I agree they could possibly get away with AI since it's being driven by his cast mates, but I don't think they'd try it.

It's an interesting journalistic debate. On the one hand their job is to report, not to help people stay anonymous.

But Banksy is part performance art, and his anonymity is part of that, by revealing his identity you arguably destroy the art work.

I feel like this expose kinda gets forgotten because Banksy was never completely anonymous, the reason he's not really known is that people recognize the anonymity is part of it and they don't want to know who he is.

Not in the slightest.

It just shows we have some levers to reduce consumption that we don't normally use.

It doesn't show that we can reasonably use those levers long term, not that those levers are actually sufficient to reduce fuel consumption enough to make up the difference.

And a douche bag of a president who drops bombs next to schools and kills 135 kids . Should resign on the spot for that.

Look up "human shields", the practice of siting military targets among (or in or under) large collections of non-military civilians, in order to deter strikes against them or produce propaganda claims of atrocities when they're attacked anyhow.

In such situations the fault for the "collateral damage" is assigned to the side that set up the arrangement, not the side that hit it.

Nevertheless, it should be noted that the US has been trying very hard to use precision munitions and extreme military intelligence to take out military targets with as little harm to the innocents they're embedded among as possible, with impressive success. Compare the amount of collateral damage in this war to any of those conducted in the 20th century.

He's doing the bare minimum sniff test of verifying that *you* are the guy whose name is on the bookings and not someone sneaking in on someone else's name who can't even pronounce the name on your fake id.

At least in the case of people claiming to be returning citizens I've been told that they're comparing your accent to your claimed residence (or residence history).

Different words are acquired at different ages, and many are pronounced with regional variations. An expert can talk to you for a few minutes and come up with a pretty good age-map of where you lived as you grew up. An agent with a modicum of training can detect a mismatch between how you pronounce certain words and your claimed residence and pass you through quickly or keep you around and drill more deeply. (If you now live in an area with a regional accent wildly different from where you grew up it can help to answer a where-do-you-reside question with "Footown, but I grew up in Barstate".)

I presume they are doing something similar, though no doubt with lower resolution, on the world-wide level for visitors from other countries.

That's why actual researchers did a study.
Researchers examined the year-to-year change of 33,000 populations in the northern hemisphere between 1993 and 2021, and isolated the effect of the decadal rate of seabed warming from short shifts such as marine heatwaves. They found the drop in biomass from chronic heating to be as high as 19.8% in a single year.

I mean the method they used to isolate the effects of temperature is literally in the first paragraph of the summary.

It the US planning a long term occupation of Iran?

Because without that they aren't getting regime change, certainly not a regime change they'd like.