News roundup
(1) Suddenly we see a host of doomsday warnings about AI. Is it real or is it orchestrated propaganda? Only AI knows.
The dangers posed by self-improving AI may be entirely real. So, too, however, are the enormous financial and political interests competing to shape the rules governing it. Before Congress hands a new federal agency the power to halt technological development, destroy “non-compliant” systems, and imprison their creators, lawmakers should determine whether this sudden chorus of doomsday warnings was orchestrated by those hoping to write the rules and secure seats at the regulatory table.
(2) Guess what? You’ll never guess; more ballot fraud.
In California:
James Brass, 47, of Victorville, was arrested today on a federal indictment charging him with paying people on Skid Row in downtown Los Angeles to sign petitions using stolen identities of registered voters to qualify initiatives on ballots in California elections.
Brass is expected to make his initial appearance this afternoon in United States District Court in Riverside.
Also charged in the two-count indictment are two petition circulators who worked for Brass: Courtney Price, 49, of Jacksonville, Florida, and Jateisha Herron, 33, of Boron, California.
(3) John Fetterman does a modified Zell Miller. Here’s how it happened:
Sen. John Fetterman, D-Pa., delivered a surprise video message at the GOP midterm convention Wednesday night, praising President Donald Trump and Republican Sen. Dave McCormick while rejecting what he described as the Democratic Party’s “extremes and socialism.”
“I’m a commonsense Democrat. I’m always going to stand with America,” Fetterman said. “I’m always going to reject the extremes and socialism and that anti-American way of life.”
Fetterman introduced McCormick, who is seeking re-election to the Senate, as “the kind of senator that gets the job done” and someone who will “fight for Pennsylvania.”
Standing in front of an iconic U.S. steel mill, Fetterman said he was proud to work with McCormick and Trump to defend Pennsylvania’s steel industry, which remains a central part of the state’s economic and political identity.
This is not only a breath of fresh air, but I don’t think it will hurt Fetterman (and may in fact help him) if and when he runs for re-election.
(4) There are many essays in recognition of the 25th anniversary of 9/11, which is tomorrow. I plan to write a few things tomorrow, but here are some things you might want to look at today:
(5) Iran is on the ropes.
But I’ll believe it when it happens.
(1) I read through the link and saw this:
Okay, we get it. This is frightening — especially for those of us whose scientific knowledge is limited.
There is a book about government debt with the title, “This Time It’s Different.” This phrase from the finance world is meant to be an amusing jab at people who are either claiming the sky is falling or some nirvana is about to happen. The phrase is really suggesting that, “No. This time is not different.” That is, people should be skeptical. Another interpretation is, if it sounds too good or bad to be true, it probably is.
If, in the year 1940, some not well known scientist said that there would soon be a single bomb that could blow up an entire large city, would knowledge of physics make that claim more scary, or would scientific ignorance make the claim more scary?
It could be a ridiculous claim, however at this point I believe some knowledge of the technology makes the claim more scary. These AI folks are putting together some rather stupendous elements, IMO. Reflexive skepticism of such things usually carries the day. But, “This time may be different.”
@neo: Suddenly we see a host of doomsday warnings about AI. Is it real or is it orchestrated propaganda?
Two, two mints in one!
It’s real and it’s propaganda. It’s not just local Bernie Sanders stuff. The Chinese are funding at least some of the AI hysteria in America, obviously to help China compete in the AI arms race.
AI doomsday is expressed as P(doom), probability of doomsday. Just about everyone agrees that P(doom) is a positive percentage. The low estimates are around 5-10%.
This is not new.
https://en.wikipedia.org/wiki/P(doom)
I’m something of an AI booster, partly because it’s an arms race which IMO we must win against China. But the risks are real and existential.
Fingers crossed. 🙂
Tommy Jay,
Elon Musk in his recent Economist Magazine interview was more sanguine – he said there is only a 10% mankind is doomed.
I am personally baffled by this talk because as I noted earlier, these LLM’s can’t even give an answer to a straight question where there is certainty but a lot of verbiage out there of denial, nuance, excuses etc.
Yann LeCun, Meta’s chief of AI before he departed to start his own company has said many times that the LLM’s are garbage and not worth the effort.
MIT Technolgy Review – “Yann LeCun’s new venture is a contrarian bet against large language models” – Google it: The URL is too messy.
Re: This time it’s different.
I heard that in the dotcom boom, while sleeping on the floor as a dotcom programmer.
The dotcom boom became a bubble. And my company (BlueLight.com) went under as well as our main client, Kmart. But the internet still took over and became one of the most significant technologies in history.
The current AI bubble may well bust. But AI won’t stop.
9/11
Since 9/11 we’ve been invaded by 4 million Hijrah seeking Muslims and the people who run our government say nothing about Hijrah because they are clueless as what it is and that goes for the GOP, who should know better but do not.
Every Western country’s Leftists have united 100% with Islam and their own Muslim invasion.
I will tell you very simply that if this continues there will be a Northern Ireland type Black Civil War. UK Professor, David Betz has written and spoken about this very carefully but forcefully.
Muslims are not here to integrate but to dominate. their imams say this in mosques all over the US every Friday. Won’t see that on CNN
@John Galt III: I am personally baffled by this talk because as I noted earlier, these LLM’s can’t even give an answer to a straight question where there is certainty but a lot of verbiage out there of denial, nuance, excuses etc.
You are basically asking a current AI to opine on a controversial topic to your satisfaction, while much of its training data goes against your position. Fine. However, AI does a lot better on a lot more queries that people actually care about.
LLMs are basically huge (I mean, yuge!) linear algebra transformations operating on huge datasets of human texts and data. Their responses are based on vast numbers of weighted averages including feedback.
LLMs are much more like humans than Hollywood computers (True/False/Explode) of the 50s/60s.
Plus AI is always improving. Already I see that current ChatGPT doesn’t hallucinate as much as Chat of a year or two ago.
When I was a tournament chess player I could handle the chess program Sargon II. Ten years later I didn’t stand a chance against the 80s PC chess programs.
Today the World Chess Champion doesn’t stand a chance against your cell phone.
“Muslims are not here to integrate but to dominate.” John Galt III
Yes and that will be their undoing. In the last year I’ve seen an exponential growth on the internet in awareness of Islam’s true nature.
I saw a video of Mr. Coxon describing his opinion and stance. It’s mostly the same as that discussed in neo’s link. One of his points was specifically, that the current AI systems in common use now are not that good or smart. But, he’s seen how fast the whole industry is moving and has some idea of where it is headed in the very near future. He’s saying the systems will be vastly more capable very soon.
LLMs are basically huge (I mean, yuge!) linear algebra transformations operating on huge datasets of human texts and data. Their responses are based on vast numbers of weighted averages including feedback. — huxley
My limited researching says that these are neural nets that, at least in the beginning, model how animal neurons interact with each other. I think they still do also, though the researchers tend not to use the “neural” nomenclature now. There were various algorithms used in the past, some of which wouldn’t scale with massively parallel processing computer systems. They figured out ways to avoid that, which didn’t damage the model; and so the “hyper scale” processing race is on.
Now these models are vectorized linear algebra computations that can be paralleled across 10’s of millions of processor cores simultaneously. Each neural interaction takes about one clock cycle in time, and they are assembling effectively many trillions of neurons. The human brain has about 85 billion.
As Mr. Coxon describes, he is worried about the situation where the AI system can modify and improve itself on its own. Will it discover some incredible improvement we humans haven’t or couldn’t even imagine?
AI isn’t just LLMs, the same techniques can be applied to other areas, like driving, image recognition, or a drone attack. Ukraine is the leading exporter of training data for drone war. There is also work on reasoning, which is a whole nother kettle of fish. The advantages of compute are speed, and increasing scale. The field is advancing incredibly fast. In the open source project I am part of, I’d say about 6 months since AI became good enough to be usable, to now, when everyone is using it.
Maybe it’ll discover that God doesn’t play dice with the universe.
(Or maybe the opposite…$
huxley,
“You are basically asking a current AI to opine on a controversial topic to your satisfaction, while much of its training data goes against your position. Fine. However, AI does a lot better on a lot more queries that people actually care about.”
Oh, I see. So, if if lies and spits out propaganda that’s ok. So, you are saying it does well on telling the average dolt what Dolly Parton’s bra size is but falls flat when asked more serious quetsions. Got it. As I wrote, Yann LeCunn who is one the best AI developers agrees large LLMs are garbage.
The future by the way maybe Small LLM’s on your personal computer as they are catching up to LLM’s very, very rapidly. Good!!!
huxley,
Another example:
Amodei, the founder of Anthropic and its CEO was married 3 years ago to a woman who tried to sell Jeffrey Epstein a porn site she concocted.
Up until very recently any inquiry of Claude baffled this favorite LLM of yours by saying it wasn’t sure he was married but gave his sister’s name as his wife. Is that the kind of answer that “people care about? Those are called lies or at least purposeful obsfuscation.
I suggest a thorough study of Gellman’s Amnesia.
@ Neo > “(1) Suddenly we see a host of doomsday warnings about AI. Is it real or is it orchestrated propaganda? Only AI knows.”
But Sarah Hoyt Has Thoughts!
https://accordingtohoyt.com/2026/09/10/the-corrupt-stupid-leading-the-delusional/
Here’s the infamous Coxon tweet, which I picked up from Alex Berenson’s substack.
https://x.com/hilbertspaess/status/2097476196791709843?utm_source=substack&utm_medium=email