Yann LeCun Leaving Meta — Large Language Models Are a Dead End
November 16th, 2025Warning: Not investment advice.
I provide ideas and analysis to a Cryptogon reader who is very active in equity investing. What follows is based on information that I’ve provided to this investor.
Making money on long side trades in tech related stocks, and stocks in general, has been easy in recent months. This is due to frenzied investments into companies that are directly and indirectly related to the Large Language Model (LLM) artificial intelligence paradigm.
While LLMs are capable of accomplishing some incredible parlor tricks—I have scripts and programs that I could not have coded on my own, for example—Big Tech CEOs are trying to convince everyone that LLMs are going to lead to artificial general intelligence.
AGI is not going to emerge from LLMs.
In addition to the diminishing returns around scaling LLMs, the technology routinely produces results that contain significant amounts of hallucinated material. While output might appear impressive at times, it will often contain hallucinated statements and sources.
The hallucination problem is getting worse, not better.
It’s very possible that hundreds of billions of dollars have been spent to travel down a dead end path.
I have no ability to quantify how bad this could be for stocks as increasing numbers of people figure this out. With LLMs potentially representing a massive bubble and the consumer increasingly living paycheck to paycheck, I’m urging caution with long side bets in equities.
If President Warp Speed Trump’s $2000 checks go out, that could provide a sort of short term sugar high, but don’t be fooled.
I don’t know much about the AI field except that it’s prone to extreme boom-bust cycles, and this boom cycle has dwarfed all of the rest combined. Additionally, it has pulled the wider economy into its reality distortion field.
Is the “World Models” paradigm the way to AGI, or even something less polluted with hallucinations than LLMs?
My guess is that World Models are going to be infinitely harder to pull off than LLMs and that it’s a case of trying to keep doubling down. But, as I said, I don’t know much about this field.
Here’s what I do know:
When I ask for the title of a band’s upcoming album and the LLM makes up an album name and fake sources to back up the hallucination…
When I ask for book recommendations on a topic and I get hallucinated titles and authors…
When I ask how to do something using the free version of DaVinci Resolve and it gives me instructions on how to do the task using the paid version, and repeatedly insists that the task is doable in the free version…
There’s a serious problem.
Therefore, if it becomes widely acknowledged that LLMs are unreliable and, perhaps, irredeemably flawed, there could be a lot of downside potential in stocks.
Via: Gizmodo:
One of the most important AI scientists in Big Tech wants to scrap the current approach to building human-level AI. What we need, Yann LeCun has indicated, are not large language models, but “world models.”
…
Why is he leaving a company that’s been spending lavishly, poaching the most highly-skilled AI experts from other firms, and, according to a July blog post by CEO Mark Zuckerburg, making such astonishing leaps in-house that supposedly the development of “superintelligence is now in sight”?
He’s actually been hinting at the answer for a long time. When it comes to human-level intelligence, LeCun has become notorious lately for saying LLMs as we currently understand them are duds—no longer worth pursuing, no matter how much Big Tech scales them up. He said in April of last year that “an LLM is basically an off-ramp, a distraction, a dead end.”
More: Meta’s star AI scientist Yann LeCun plans to leave for own startup
Amazon’s Starlink Rival Is Called Leo
November 14th, 2025A few days ago, my (very rural) area in New Zealand became eligible for free hardware on Starlink. That’s extremely tempting…
I’d get the Residential Lite service tier, which deprioritizes traffic in favor of users who pay for the full priced regular service. If you have Starlink’s Residential Lite service, what speeds do you get, peak and off-peak? Ping? Jitter?
Via: PC Magazine:
Project Kuiper started off as a code name for Amazon’s satellite internet ambitions. “We’ve hit some important milestones under that name, but there’s lots more on the horizon, and it seemed like a good time to share our permanent brand for the program,” says VP Rajeev Badyal.
The name is a nod to low-Earth orbit (LEO), where the company’s satellites will reside as they circle the planet. The lower orbit has already enabled SpaceX to deliver satellite internet speeds of 200Mbps and higher to users on the ground, and Amazon plans to do the same, as higher-orbiting geostationary satellites struggle to compete with Starlink.
Dr. Suzanne Humphries: Deadly Secrets & Monkey Business
November 13th, 2025“What we saw in 2020 was not a new thing.”
Via: Children’s Health Defense:
Books:
Dissolving Illusions: Disease, Vaccines, and the Forgotten History by Dr. Suzanne Humphries
Dr. Mary’s Monkey by Edward T. Haslam
Me & Lee: How I Came to Know, Love and Lose Lee Harvey Oswald by Judyth Vary Baker
Make America Great Again: Trump Organization’s Increasing Employment of Foreign Workers
November 13th, 2025Via: The Hill:
The Trump Organization requested 184 foreign workers to work across various company properties, a record number that has increased over the years.
The company sought to hire workers through H-2A and H-2B visas for temporary positions at Mar-a-Lago, two golf clubs and at Trump Vineyard Estates in Charlottesville, Va., according to data from the Department of Labor.
Over the course of Trump’s first term and the first nine months of his second term, the Trump Organization’s visa requests increased from 121 in 2021 to 184 in 2025, according to Forbes. Overall, the company has filed to hire 566 foreign laborers, primarily to work as servers, farm workers, kitchen staff, clerks and housekeepers.
Massie: Trump Pressures Boebert, Mace on Epstein Files
November 13th, 2025Massie just dropped a bomb:
He says the White House directly pressured GOP holdouts, Boebert, Mace, MTG, to abandon the Epstein files petition, telling one of them, “We can’t let Massie have a win.”They weren’t trying to protect “process.”
They were trying to block the vote… pic.twitter.com/ajMDlMaNgC— Brian Allen (@allenanalysis) November 13, 2025
Club of Rome: Giant Leaps
November 13th, 2025? THE “GIANT LEAP” — THE WEF’S MOST RADICAL BLUEPRINT YET
The global elites have finally said the quiet part out loud: their so-called “Giant Leap” is not a proposal — it’s a plan to reshape society from the top down, whether the public agrees or not.
In their own words, this… pic.twitter.com/OjgrqJTA5a
— Jim Ferguson (@JimFergusonUK) November 13, 2025
The speaker in the clip above is Sandrine Dixson-Declève from the Club of Rome. This is from her opening plenary speech at the “Beyond Growth” conference in the European Parliament on May 16, 2023.
U.S. Has Stopped Minting Pennies
November 13th, 2025Via: USA Today:
America’s last penny was struck at the U.S. Mint in Philadelphia, the end of a coin production that started in 1793 and ended Nov. 12. We’re already starting to miss them.
Some stores have begun running short of pennies and have had to round cash transactions up or down to the nearest five cents when customers don’t have exact change, USA TODAY reported.
President Donald Trump order the Treasury Department to stop minting pennies as a budget-cutting measure in February. But a resulting shift to nickels, which are more expensive than pennies to produce, could cause other losses in the federal budget.
Google Will Continue to Allow Sideloading Apps on Android
November 13th, 2025Rare good news!
I’m genuinely surprised that Google backtracked on this.
Via: Google – Android Developers Blog:
While security is crucial, we’ve also heard from developers and power users who have a higher risk tolerance and want the ability to download unverified apps.
Based on this feedback and our ongoing conversations with the community, we are building a new advanced flow that allows experienced users to accept the risks of installing software that isn’t verified.
Study 329: One of the Most Infamous Cases of Scientific Fraud in Modern Psychiatry
November 11th, 2025Via: Brownstone Institute:
It began with a lie.
In 2001, the Journal of the American Academy of Child & Adolescent Psychiatry (JAACAP) published a paper declaring that the antidepressant paroxetine (Paxil) was “generally well tolerated and effective” for adolescent depression.
That conclusion was false.
The manufacturer, GlaxoSmithKline (GSK), knew from its own data that the drug failed to outperform placebo and carried a serious risk of suicidal behaviour.
Instead of telling the truth, GSK hired a public-relations firm to ghostwrite the paper, enlisted academic co-authors who never saw the raw data, and used the publication to promote Paxil to doctors treating children.
It became known as Study 329 — one of the most infamous cases of scientific fraud in modern psychiatry.
For years, the fraud stood unchallenged. Regulators issued warnings but never forced a correction. The journal refused to retract. The paper remained in circulation — cited hundreds of times, shaping prescribing habits, and legitimising a lie that cost young lives.
Make Al Qaeda Great Again
November 11th, 2025The US fought the "war on terror" for more than 20 years — it's ongoing. It not only cost trillions of dollars and more than a million lives, but radically and permanently eroded core civil liberties for Americans.
Now, someone the DOJ called an Al Qaeda terrorist less than a… https://t.co/efxKy1YqJ3
— Glenn Greenwald (@ggreenwald) November 11, 2025
