93 Comments
User's avatar
E.R. Flynn's avatar

Ahh, the supreme irony of CEOs who don't trust their own dangerous half-assed creations. Mary Shelley wrote a book that was a parable about such things, but I guess few have read it.

Marc Schluper's avatar

You seem to be as clueless about risk management as you are about A.I.

The chance that A.I. fails at tasks they have been thoroughly tested for is small, but not zero. The impact of a failure is unprecedented, though.

So the risk is high. And their product is not at all a "dangerous half-assed creation".

E.R. Flynn's avatar

Since you seem to be so sure about trusting not only a group of unhinged, dangerous imbeciles to use AI without a shred of common sense, but also an AI, that when tested in various war game scenarios has chosen “nuclear attack” more than a few times, then let me suggest that you purchase a good Geiger counter so that you might navigate through all those good intentions that will be paving the path to our hell.

mcswell's avatar

ER and Marc, is it possible that you are on the same side, and just not understanding each other? ER, you seem to be saying (esp. in your reply) that AI should not be trusted, particularly with nukes. Marc seems to be saying the same--the risk is (too) high.

E.R. Flynn's avatar

Interesting take on the varying degrees of disagreement. However, the point I was making was regarding a parable on the careful consideration of consequences. Marc seemed not to understand that but instead opted for derogatory insults without a measure (at least in my reading of his response) of the sensible distrust needed towards this new technology. But I guess it is in way, two sides of the same coin. Although one side comes with the risk of needing a lead-lined bunker.

Purnima Gauthron's avatar

Generative AI in every single AI company is based on the same brute-force Transformer-based AI models, and flawed not just in Math and probabilities but in basic computer science fundamentals mixing instructions with data in the model. Needless to say this leads to a host of security issues and hallucinations and cannot be trusted. And where do you even begin to design these systems for autonomous safety in military applications?

So yes, from a pure technology and math assessment this is a half-assed creation.

From an economic standpoint the same AI companies have hyped and sold the public, corporations, and world governments a bad bill of goods costing a whopping $600 billion per year — money that could have been invested to solve thousands of critical issues on our planet. On top we are burning through valuable energy resources, and even polluting our own minds.

Seriously the value of these Generative AI products is worth no more than $10 billion — if they had done it right.

Do we humans not have standards anymore?

Notorious P.A.T.'s avatar

"The chance that A.I. fails at tasks they have been thoroughly tested for is small"

Since when?

Harry Adams's avatar

Depends how you define fail. 95% hitting the red button sounds like as big a fail as is conceivable, at least by any sane standards. And yeah certain tasks it’ll fail 100% of the time I’m sure, no matter how many tests you’ve run.

Sam Litvin's avatar

Sounds like the AI bros didn’t get the President they thought they paid for after all. Who would have thought?

Bonnie Blodgett's avatar

Thanks for writing about this. The biggest threat to the world isn't artificial intelligence but authentic stupidity. I am referring not only to Trump, but also to the entire elite political class that "governs" the U.S.

Plunders U.S. taxpayers is more like it.

BF's avatar

The Epstein Class core values: GREED, KIDS AS SEX SLAVES, & ABSOLUTE POWER.

Alex Tolley's avatar

Not just the US. pa;antir is in deep with the UK government, with their tentacles in the NHS and Ministry of Defence. There is concern about this, but the government seems oblivious...so far.

Joy in HK fiFP's avatar

"the entire elite political class that "governs" the U.S." AKA Pathocracy. What are we going to do to stop them?

Purnima Gauthron's avatar

Trump never owns up to anything. He will blame "the RADICAL Woke LIBERAL LEFT for creating a DANGEROUS CRAPPY product under Sleepy JOE. A RIPOFF!!!" for sending a swarm of LLM drones to Palm Beach when they were planned to target Playa Carib (in Venezuela).

David Hsing's avatar

Ham-fisted moves are par for the course for that guy

p.s. Trump won't "own" for anything. He's doesn't take responsibility for squat.

mcswell's avatar

President Harry S. Truman had a sign on his desk "The buck stops here." If Donald had a sign on his desk, it would say "The buck comes here."

Catherine Blanche King's avatar

What a bright light in a dark time. The Anthropic people . . . . (Later edit: I omitted "and Sam Altman" out of this post in the light of recent Altman naivete: he apparently believes the Trump administration will stand by their promises--hahahahha) . . . may have just "set the doomsday clock back" and saved the world. (I said "may.") I hope others will follow and not cave. There is NOTHING sane or reasonable about either of the Trump/Hegseth intrusions on the whole idea of safety.

But Hegseth either knows nothing about, or pays no attention to, or gives a hoot about the U.S. Constitution or the oath he took to it (goes without saying for Trump. Again, "lawfulness" in an authoritarian state is not even close to the lawfulness of a democratic state. So even THAT reference is a colossal mischaracterization (deliberate or from ignorance) and sign of contempt as is the entire Trump diatribe.

Just to be clear, in authoritarian states, the authoritarian (fascist) IS the only source of the law (in this case, Trump is the guy, and he says openly that's what he wants) and the pseudo-laws can be generated by a whim, a personal bias, or a lack of sleep, or have no reason at all. It's already not sane to even want such a thing, much less to carry it out..

On the other hand, look at the procedures and rules, and the network of thought-through protocols, checks and balances, habeas corpus, rules of inference, the courts, and accountability for awarded power in a democratic state; and where the ideal is that everyone is the same under the law. Though not always perfect, it's based on well-developed argumentation with lots of input from intelligent people who, in most cases, seriously care about such things as the First Amendment.

The alternative is to say, yeah, let's go ahead and give the racist/misogynist in chief the power to surveil anyone he wishes, and to go golfing while our non-human tech takes care of things at the Pentagon and also the fascist's perceived enemies.

That lefty woke thing, applied to reasonable people trying to do the safe, sane, and reasonable thing with their tech for "the people" of the United States et al, is so telling of a manipulative character. The nicest thing I can say about that is that's it's juvenile and reminds me of that kid in middle school who stood a sharp pencil up on the seat in front of him for the geeky guy to sit on the point of the pencil and then laughed about it. Someone, please: take away his keys.

Olivia's avatar

I’ll just leave this here ….AI Models Consistently Escalate to Nuclear War in Simulated Military Scenarios

https://oecd.ai/en/incidents/2026-02-25-2d63

Jack Phillip's avatar

This all a kabuki to get in Grok?

User's avatar
Comment removed
Feb 27
Comment removed
Oaktown's avatar

The US govt doesn't appear to know that Elon is a nut; they keep giving him millions of dollars in DOD contracts despite the fact he's an admitted drug abuser and a Nazi.

toolate's avatar

millions? missing zeroes their. try 10-15 BILLION

Antonio Dias's avatar

Not to worri, MechaHitler is all for it!

Ian Langmead's avatar

"If anything really bad comes of pushing premature AI too hard and too fast into to the military, he will own it." The irony it that maybe as a result there will not one around to read the that epitaph....

Andrew Wilson's avatar

Perhaps autonomous AI will decide the current US administration is the greatest obstacle to progress and take them all out?

Bob Johnson's avatar

Or,.... time to rewatch

Colossus: The Forbin Project (1970)

Paul Czyzewski's avatar

To me the most disingenuous part of this is to threaten to label Anthropomorphic a "supply chain risk."

In technical terms, this is called a bald-faced lie. ;)

Bob Johnson's avatar

Trump wrote, "I am directing EVERY Federal Agency in the United States Government to IMMEDIATELY CEASE all use of Anthropic’s technology."

So, we can expect federal government efficiency to increase.

PHT's avatar

"if anything bad happens with [...] , Trump will own it" if surprisingly _not_ prescient. You're suppose to know better. He will be blamed for it, for sure, and he will deserve the blame, for sure, but he will never own it. The only thing he'll own is whatever money he made selling cryptos and fake steaks... May he enjoy it in the graveyard when his time comes, as for for all of us...

Brian Curtiss's avatar

What about some kudos for Anthropic? Any Silicon Valley company that can illicit that response from Trump deserves some acknowledgment for his rant at least. Probably only cost them a billion or so. They'll be fine.