AGI researcher, CS PhD, inventor, philosopher, musician. AI, HPC, CogSci, DeepTech, Crypto. Founder, @cic_agi, @cypher_coin ⏩️⏩️⏩️
Dead City, Antarctica
Joined December 2009
- Tweets47.2K
- Following5.3K
- Followers3K
- Likes96.7K
It's not an industry, it's a clown car. 🤡🚗
Yann LeCun was right & nobody in this industry wants to say it out loud.
A 109 million parameter model beat models up to 72 billion on a specific test: how closely its internal representations matched those of image models.
About 660× smaller.
Randall Balestriero and LeCun’s paper examines why models trained on text and images can develop similar representations of the world.
Their finding: scale alone doesn’t explain it. The examples paired during training matter.
Two observations of the same scene give the model a shared concept to learn. Predicting the next word from earlier text provides a less direct connection.
In their experiments, the better-paired training method needed 42× fewer examples to reach comparable alignment with vision models. Their small text JEPA also aligned better than every language model they tested, including 72B models.
That measures alignment, not overall intelligence. But it gives researchers a concrete way to improve this capability beyond adding parameters: change what the model learns to connect.
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
This is probably my favorite study.
Orange juice blocks endotoxin release
AND
Decreased TRL4 expression.
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
This teacher-turned-cognitive scientist shared a disturbing reality that left the room stunned.
“Our kids are LESS cognitively capable than we were at their age.”
Every previous generation outperformed its parents since we began recording in the late 1800s.
So, what happened?
Screens.
Dr. Jared Horvath explained:
“Gen Z is the first generation in modern history to underperform us on basically every cognitive measure we have, from basic attention to memory, to literacy, to numeracy, to executive functioning, to EVEN GENERAL IQ, even though they go to more school than we did.”
“So why? … The answer appears to be the tools we are using within schools to drive that learning (screens).”
“If you look at the data, once countries adopt digital technology widely in schools, performance goes down significantly, to the point where kids who use computers about five hours per day in school for learning purposes will score over two-thirds of a standard deviation LESS than kids who rarely or never touch tech at school. And that’s across 80 countries.”
But screens aren’t just decimating learning and making new generations less intelligent than the ones before them.
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
Marc Andreessen revealed how Harvard Business School was built on a broken 1941 theory, and how it's now collapsing...
Andreessen co-founded Netscape in 1994 and a16z in 2009.
He has sat on Meta's board since 2008.
He has spent 30 years backing founders and watching managerial CEOs lose to them.
The pattern traces back to one book: James Burnham's The Machiavellians (1941).
Burnham argued every great company had been founder-run.
Henry Ford ran Ford. Bob Noyce ran Intel.
Today, Elon Musk runs Tesla, SpaceX, and Starlink.
Then, he said, something broke.
Between the 1880s and the 1920s, a new philosophy replaced the founder. It was called managerialism.
The professional manager would now hold a portable skill, usable across any business.
The consequences were:
- Harvard Business School
- Stanford Business School
- Management as a universal skill
- The 1970s conglomerate
"That assumes the managers are going to do a good job," Andreessen says.
For 30 years, they haven't.
Managers can run something static, he says. Soup is soup. A bank is a bank. A car is a car.
But when the industry changes, the manager freezes.
Look at SpaceX.
"Imagine being a professionally trained manager, trained at a top management school, working for a rocket launch company, competing with SpaceX."
Then Elon's rockets started landing on their butt.
"Your management skills ... what good are they at that point?"
Andreessen's conclusion:
"You're much more likely to build something important in the 21st century if you start with the founder and train them on management."
The value of the brain of an average X user is about -$100000.
How long before SI decides to completely scan all of our brains? Perhaps the greatest repository of information in the reachable cosmos is the set of all human brains with their deeply buried, partially forgotten, and half-confused memories. By correlating the data across all human brains, the SI could probably infer far more real truth than from any other source. Beware: our brains are valuable ore! Probably our pets and some wildlife, too.
Because of brainless maggots trying to steal every idea, we can no longer have scientific/technical debates on the internet.
Apparently this was the Convair Kingfish design, via Canceled Aircraft page on FB. It's remarkably close to SOTA hypersonic airframes AFAICT so I think it was a very good effort, after all! One only truly learns through studying the history of a subject.
facebook.com/permalink.php?s…
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
Our own military pilots keep seeing things in the sky they can't explain. Not airplanes. Notdrones. Glowing balls of light that stop on a dime and take off faster than anything we've built.
You and I grew up calling them UFOs. Here's the part that will get you: the government finally admitsthey're real. They just won't tell us what they are.
I've been trying to crack this story for a good while now. What a former intelligence officer toldCongress, under oath...
Blockchains are criminal-friendly by design. I am probably the only Layer 1 developer trying to change that. But of course an industry that is dependent on crime to function did not take notice yet. :)
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
100% Cancer Remission Achieved in ALL Patients in Groundbreaking Study
Dr. Andrea Cercek, one of the lead researchers, said:
“It’s absolutely incredible ... We’ve certainly never seen this before.”
All 12 patients experienced complete rectal cancer remission after receiving dostarlimab, a monoclonal antibody.
This particular monoclonal antibody achieved 100% cancer remission without chemotherapy, radiation, surgery, or the toxic side effects that come with such therapies.
The authors wrote:
“All 12 patients (100%; 95% confidence interval) had a clinical complete response, with no evidence of tumor on magnetic resonance imaging ... and no cases of progression or recurrence had been reported during follow-up (range, 6 to 25 months after treatment). No adverse events of grade 3 or higher have been reported.”
How does this monoclonal antibody actually work?
Dr. Cercek explained that it essentially unlocks the body’s ability to heal itself.
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
A major study links six common artificial sweeteners to accelerated brain aging, with the highest consumers experiencing a 62% faster decline in memory and thinking skills.
sciencedaily.com/releases/20…
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
Since many of you have asked me privately for introductory material on analytic philosophy after the Kripke post, I wanted to propose to you Michael Dummett's ''Origins of Analytical Philosophy'', a nice short read which requires very little technical knowledge, outside of the basics of logic, as the book focus on the history of the ideas found in analytical philosophy.
Yann backs me up again 😅
TURING AWARD WINNER SAYS THERE'S "ABSOLUTELY NO WAY IN HELL" LLMS REACH HUMAN-LEVEL AI
yann lecun says scaling LLMs won't get us there, and definitely not in the next two years
"the idea that we're gonna have a country of genius in the data center, that's complete BS"
that's a direct shot at dario amodei, it's his phrase for where anthropic is heading
instead he thinks we'll get something that feels like "a PhD sitting next to you. but it's not a PhD"
a PhD invents solutions to new problems. what we're building finds the ones that already exist, just really fast
Do not ever use ChatGPT
futurism.com/artificial-inte…
Eray Özkural, PhD - OG AI ⏭️🌟 retweeted
Saul Kripke, it's fair to say, is one of the most important analytic philosopher of the past century, and although many books were written as ''discussions and criticism'' of his ideas, it's a bit harder to find a good one which is closer to an overview of his ideas on metaphysics and philosophy of language.
In ''Kripke: Names, Necessity and Identity'' by Christopher Hughes (Oxford) you will find just that, and more!
The book is fairly short (259 pages), and requires basic knowledge in epistemology, metaphysics, linguistics and logic. Very well written. If you have struggled to understand Kripke, this is a great way in.
Check it out!
