Ex- Bell Labs (Holmdel image processing), ex- simultaneous interpreter (Italian, French, English), entrepreneur. Working on 2 stealth projects
La Jolla Village, San Diego
Joined May 2022
- Tweets645
- Following730
- Followers106
- Likes867
Starting a blinded, matched-harness comparison of GPT-6 Astra vs GPT-5.6 Sol. Not benchmark scores: I’m testing how each constructs objects from ambiguous evidence—when it preserves alternatives, commits, and repairs after new evidence. Exact prompts + raw outputs will be public.
Paolo Pignatelli retweeted
“There is no such thing as probability at a fundamental level. The world is completely deterministic…
Not all colleagues agree with this - to put it mildly. But the ones who understand it do.” - @DavidDeutschOxf
Paolo Pignatelli retweeted
The CEO of a $3 billion AI company mass disqualified an entire generation of workers.
And told the brainrot kids they're the ones who'll win.
Amjad Masad, CEO of Replit used by 20 million developers just dropped a bomb.
He said the single most important skill in the AI age is not coding, engineering, credentials.
It's idea generation.
Why? Because AI is driving the cost of building things to zero.
He said the single most important skill in the AI age is not coding, engineering, or credentials.
It's no longer "can you build it?" It's "can you think of it fast enough?"
And here's where it gets controversial.
Masad says the people best positioned to win aren't the Ivy League kids or the seasoned professionals.
It's the terminally online, the "brainrotted."
The chronically plugged in doom scrollers.
His logic is that if you're addicted to social media and you know every trend, every meme, every emerging market before it hits mainstream, you have a massive advantage.
Then he went further.
ADHD? That's not a liability anymore, it's a superpower.
People who crave novelty, who jump between ideas, who can't sit still, AI rewards that behavior.
You can now prototype 10 ideas in the time it used to take to build one.
The old rules said focus, specialize, follow the path.
The new rules say move fast, spot trends, generate relentlessly, then let AI handle the execution.
Replit's own platform proves it.
Their AI Agent lets anyone describe an app in plain English and have it built autonomously.
Revenue grew 5x, six months, they cut their team to 65 people and became more productive.
Revenue grew 5x in six months, they cut their team to 65 people, and became more productive.
Attention spans, internet addiction, and restless curiosity might be the exact traits the AI economy demands.
The gatekeepers just lost the gate.
The CEO of the company behind Claude went on the record.
Saying things that should be front page news in every country on Earth.
His name is Dario Amodei and he runs Anthropic.
He just said we are “near the end of the exponential.”
The end of the climb as in like we’re about to arrive.
He says AI will handle end to end software engineering in 1 to 2 years.
Build, test, compile ship with no humans in the loop.
He calls it “a country of geniuses in a data center.”
Millions of AI instances, each one at the level of a Nobel Prize winner, running 24/7, thinking at superhuman speed.
His best guess is 1 to 3.
He puts 90% odds on it happening within a decade and says it would be “crazy” to bet against it by 2035.
The only reason he won’t say 99%?
Someone might invade Taiwan and blow up the chip factories.
Now here’s where it gets dark.
He says AI could wipe out half of all entry level white collar jobs.
Lawyers, consultants, analysts within 1 to 5 years.
Anthropic’s own research says programmers are the most exposed profession on Earth.
74.5% of their tasks can already be done by AI.
And this isn’t the part that scared him most.
He described this AI as the single most serious national security threat humanity has faced in a century.
A digital nation with the IQ of 50 million geniuses.
He’s not some doomer on Reddit, he built this and he’s building the next version right now.
The most powerful man in AI just told the world exactly what’s coming.
Almost nobody is listening.
1. Google DeepMind just reported a fundamental bug in RAG: embedding spaces break down at scale.
In our recent paper zenodo.org/records/17038405, The Energy–Computation Law: Operational Limits, Geometry, and Testable Predictions, we showed why this is not a bug but a physical law:
2. embeddings are bounded by energy–entropy constraints. Once the Bekenstein–Bremermann limit is crossed, collapse is inevitable.
What DeepMind observed empirically, we derived from first principles. Retrieval failure at scale isn’t an accident — it’s physics.
The article about the bug in RAG: marktechpost.com/2025/09/04/…
Exploring a unified theory of information, observation, and language—with practical algorithms. Highlights: Semantic Chemistry (Object Constructor & reactions), Nibbler (hierarchies), LLC bridges, and testable predictions. #AIResearch
Energy-integrated view (optional). If you care about physics: map bits to kBTln2k_BT\ln2kBTln2 to get a free-energy proxy and a computational light-cone; otherwise, the independent edition stands on compression + constraints alone.
Call for collaborators. Benchmarks (parsing, vision, motifs), catalyst libraries, fast port matching, and causal metrics. If this resonates, DM or reply. #SemanticChemistry #KnowledgeGraphs #OpenAI #GPT5 #ChatGPT #AIResearch
next: Acknowledgments
I want to thank @OpenAI for its great help, as well as #Grok4 and #Anthropic. #GPT5 is immensely powerful.
Grateful to @Twelvetwentyone and the Theoretical Physics / Applied / Mathematical / Computational community for accepting “The Brittleness of Plasticity: Mask Evolution Operators for Neural Network Stability.” Bridging physics + AI drift stability.
🧠💡 Neural networks must adapt — but too much plasticity causes brittleness, drift, and catastrophic forgetting.
Our new paper introduces Mask Evolution Operators to stabilize learning without losing flexibility.
🔗 doi.org/10.5281/zenodo.16882…
#AI #MachineLearning #PhysicsInspiredAI
3. Think of it like a damped harmonic oscillator for neural nets — oscillations stay controlled, energy is conserved, and stability is maintained even during continuous adaptation.
Full open-access paper & materials: doi.org/10.5281/zenodo.16882…