Pinned Tweet
Formal exposition of the discrete quadratic iterator:
A rounded contracting map forgets where it started, not how it was steered. For two quadratic maps on a 14-digit lattice, the final state fixes the last 26 controls exactly; a 27-step collision shows 26 is optimal.
Paper + code: unitambo.wordpress.com/2026/…
Counterexamples welcome.
1
1
56
🌞Phillip Somerville retweeted
GTA 6 HAS AN NPC WHO FIGURED OUT HE'S LIVING IN A VIDEO GAME. WHAT THE ACTUAL FUCK. 😂
Players discovered a house with a massive conspiracy board declaring “IT'S A SIMULATION,” surrounded by clues about the game's world.
And there's an even better detail.
The guy apparently has plans to escape by BOAT, potentially sailing beyond the game's map.
Rockstar really created a conspiracy theorist who's technically RIGHT.
Writer: Oliver
🌞Phillip Somerville retweeted
I believe there is a vast, extremely diverse space of possible approaches to learning (for both time scales: long-term plasticity / pretraining & on-the-fly adaptation / reasoning), and that the current stack (Transformer + gradient descent & token-space reasoning chain generation) is not even in the top 50% most efficient.
Future AI will look nothing like the current stack.
🌞Phillip Somerville retweeted
Historically, central bank rate rise cycles regularly ended in recessions.
Absent a broader crisis they were generally short lived and mild, capital and labour was reallocated to more productive means over time, case closed.
But when you start actively trying to avoid recessions and you drive a massive misallocation of capital and collapsing productivity growth, you end up where we are now.
Where we havent had a real recession in 35 years and its like a patch of bushland that hasn't had a hazard reduction burn in decades.
🌞Phillip Somerville retweeted
A lot of normal economic indicators that signal recessions simply have not worked for Australia in the last two decades.
In other economy's without the largest average per capita migration intake in the Anglosphere, they work fine.
But its much harder to have a headline recession with explosive population growth through migration.
On the other side of that coin, it means when a recession is finally reached its much worse living standards, because for the economy to be that bad in headline terms, it needs to be disastrous in per capita terms.
🌞Phillip Somerville retweeted
People don't want to read anymore.
Everyone says it, I have evidence. I can see the statistics for my short X posts, and for my longer blog articles, and the former do a lot better - both in terms of engagement and revenue.
I had been working on a book for a few years, but I abandoned it at around 50k words, because I didn't feel like it would make much impact. You get much more across in short form, and I see this consistently in my metrics.
Should I try to finish the book? Would anybody read it?
planetocracy.org/p/do-i-need…
🌞Phillip Somerville retweeted
Flashback to June 14, 2024.
At the time, frontier AI models have zero fluid intelligence. They score around 5% on ARC 1. They can't reliably add two numbers. They make laughable logic errors every other prompt. Reasoning models don't exist. Coding agents don't exist. Claude Code doesn't exist. "Code gen" is a complete joke. Scaling up next-token prediction based on web text data is hitting a wall.
What if a time traveler told you that, by November 2026...
• AI models would have human-like fluid intelligence, thanks to a new paradigm: test-time induction (modern reasoning models).
• Not only would ARC 1 be saturated, but unreleased benchmarks 10x harder (ARC 2) and 100x harder (ARC 3) would be saturated too.
• Nearly all code would be written by AI, and models could one-shot complex codebases flawlessly.
• AI models would be solving multiple Millennium Prize problems.
• OpenAI's valuation would be 15x higher, and Anthropic's would be 85x higher.
And on top of all that...
• Nvidia would still hold ~80% of the AI compute market.
• Nvidia would have grown its revenue and earnings 4.5x.
• Nvidia would still be growing 75% a year.
• Nvidia would have even higher operating margins (67% vs. 65% back then).
Based on that information, straight from the future, would you have bought NVDA?
If you answered yes -- congrats. Today, 2.5 years later, you'd be down 5% on your investment (even more after inflation).
We have two purported proofs of the Generalized Riemann Hypothesis, released within hours of one another, approaching the problem through apparently different constructions, with almost opposite methods of establishing credibility. ChatGPT summary follows:
The most striking distinction is that Alpöge offers a machine-verifiable certificate without a digestible mathematical explanation; Masullo offers a substantial mathematical explanation without a machine-verifiable certificate.
Neither form of evidence should be confused with a fully established theorem.
1. The mathematical contrast is substantial
Alpöge's announced approach appears to generalise a method associated with Stepanov's work over finite fields. His striking proposed innovation is to use chains of theta lifts to imitate multiplicity, with a chain length of approximately for a von Mangoldt sum of length .
Masullo instead constructs an analytic and spectral framework. His manuscript explicitly relates its trace formula to the programmes of Alain Connes and Christopher Deninger, seeking to recover information about L-function zeros through carefully constructed function spaces and operators.
There is a tantalising commonality: both approaches invoke theta constructions, although in different mathematical roles.
That is not enough to establish a shared underlying mechanism. One uses iterated theta lifts in an arithmetic argument; the other uses theta functions within an adelic spectral construction.
Nevertheless, this is precisely where I would begin looking for an unexpected correspondence.
2. There is a potentially important difference in scope
The terminology is treacherous here.
Alpöge explicitly distinguishes RH, GRH and a possible further extension to ERH.
Masullo's manuscript discusses Hecke L-functions and also formulates broader conjectures concerning automorphic L-functions on . But stating those broader conjectures is not the same as proving them.
We would need to isolate Masullo's final unconditional theorem and compare its quantifiers against the precise Lean theorem in Alpöge's archive.
That could reveal whether one claim genuinely contains the other, or whether their scopes differ.
3. A significant development in Alpöge's verification
This deserves emphasis.
Andrew Sutherland at MIT has published an independent verification log reporting that:
All 175,187 Lean modules compiled successfully.
The five headline theorems depend only on Lean's three standard foundational axioms.
Every module was replayed through the Lean kernel.
An independent proof kernel, nanoda, accepted the exported proof.
Two additional checks were still running at the timestamp of the report.
That is substantially stronger evidence than merely claiming to have formalised a proof.
It still leaves the crucial semantic question: does the theorem accepted by the machine express the full mathematical statement being claimed?
4. My assessment
The unusual situation is that we may be witnessing two independent approaches to the same historic problem, one emerging from a long human research programme, the other from a remarkably compressed human–AI collaboration.
Masullo's approach currently has the stronger reported history of human mathematical examination. Alpöge's has the stronger publicly documented mechanical verification.
And they may be complementary rather than contradictory.
If both ultimately survive, the great story would not simply be that GRH had been proved twice. It would be that two apparently different mathematical constructions independently forced the zeros onto the critical line.
That would suggest a deeper structural principle waiting to be extracted from both proofs.
For me, the next useful exercise is not to compare 940 pages with 175,187 Lean modules. It is to extract the decisive mathematical obstruction to off-line zeros in each argument and ask whether those obstructions are equivalent.
This is massive. A story I've been following for years. He thought he had more time but didn't. Alessandro Masullo announces a separate, purely human proof of the Riemann hypothesis. math216.wordpress.com/2026/1…
1
300
🌞Phillip Somerville retweeted
BREAKING: A mathematician at Anthropic claims to have used the company's AI to solve the Riemann hypothesis, the most notorious open problem in all of math
sciam.com/article/anthropic-…
🌞Phillip Somerville retweeted
Well, looks like @__alpoge__ & @claudeai are claiming a proof of the Riemann Hypothesis
alpo.ge/remark.pdf
🌞Phillip Somerville retweeted
In the coming months all manner of things is going to be blamed for the state of the Aussie economy.
But the truth is the private sector and household economy in per capita (per working age person) terms was cooked years ago.
Its all right there in the national accounts.
It was only population growth, government spending, taxpayer funded employment growth and high housing prices creating our own economic version of 'Weekend At Bernies'.
Until that is collectively accepted, we will be unable to learn the required lessons to begin to repair the damage that was done.
But that is likely not going to happen without some serious political upheaval.
🌞Phillip Somerville retweeted
The King 👑.
Stanley Druckenmiller’s take on the current economy.
He’s basically out of the stock market:
🌞Phillip Somerville retweeted
I've watched this at least five times. Impressive. I doubt a human could have written this song without AI. The themes are too complex.
🌞Phillip Somerville retweeted
Can someone please explain to me how you get fucking bail after kidnapping and holding hostage a mother and child? Please can anyone fucking enlighten me @VictorianLabor
🌞Phillip Somerville retweeted
There are legitimate questions about the US role in the 2014 overthrow of Ukraine's democratically elected president, which triggered events that led to today's ongoing disaster. The best way to clear them up would be for the Trump admin to declassify State Dept and CIA documents from that era.
Elon Musk on Ukraine:
It's true that from an invasion standpoint, Russia is the aggressor, but it is also true that we Americans staged a coup to overthrow the democratically elected pro-Russian president of Ukraine.
Without U.S. State Department and other agency help, this would not have happened.
That is obviously an offensive action.
🌞Phillip Somerville retweeted
Gods honest truth, I increasingly find myself deeply troubled by the events unfolding in our country.
Child kidnappers swiftly out on bail.
An elderly veteran forced to wait 18 hours for a ambulance dies.
Elements of political class and some of their supporters pulling on the threads of the social fabric like a bored house cat, as they continue to try to defend the indefensible on all manner of fronts.
It doesnt need to be this way.
"A family from Thornlands in Brisbane is grieving the loss of a much-loved father, grandfather and veteran, who died after waiting 18 hours for an ambulance."
nine.com.au/australia-news/q…
🌞Phillip Somerville retweeted
Chomsky was wrong.
Researchers from UC Berkeley and Google DeepMind cracked one of the biggest mysteries in AI.
For years, scientists have been finding impossible geometric shapes hidden inside the brains of LLMs.
Calendar months arrange themselves into literal circles.
Historical years bend into smooth, rippled timelines.
Cities organize themselves into actual 2D maps of latitude and longitude.
Everyone assumed this proved massive transformers were spontaneously inventing complex "world models" out of thin air.
They were wrong.
This paper proves you don't need a massive LLM for any of this to happen.
Even basic, decade-old word embeddings like word2vec produce the exact same circles, timelines, and maps.
The geometry isn't coming from the AI's architecture.
It’s coming from a hidden mathematical property inside human language itself.
Researchers call it translation symmetry.
In natural text, how often any two months appear near each other depends almost entirely on the time gap between them.
January and February share the same statistical relationship as July and August.
When a neural network trains on data with that kind of symmetry, the math forces it to learn Fourier representations, literal sine and cosine waves.
The circles? Those are the long waves.
The strange "ripples" researchers found on AI timelines? Just higher harmonics.
Here is the craziest part.
Rresearchers tried to break it.
They deleted every single sentence in the training data where two months appeared together. No "January" next to "February." No "March" next to "April."
And the circle formed anyway.
Why? Because the geometry is a collective phenomenon across the entire language.
Even if two months never touch in a sentence, they are both pulled into orbit by thousands of other words tied to the same seasons, weather, and events.
The rest of human language forces the shape into existence.
🌞Phillip Somerville retweeted
And as a result, the Tasmanian Government has published one one of the greatest Government Documents I have ever seen
🌞Phillip Somerville retweeted
As others have noted, James Bond as a current year political and practical matter comes across as absurd.
Who are the "enemies of England" at this point that would require a Bond figure?
The official policy of the country is endless curry and halal shops. Selling every bit of London to foreigners. When terrorism does happen, it's excused as "part and parcel" and simply the "price to pay" for multiculturalism.
When terrorists bombed Manchester the English were told to "not look back in anger." When terrorists massacred the French nearby, all government and news agencies covered up the atrocities.
When "grooming gangs" were / are running rampant, the police arrested victims and the fathers trying to get revenge for their daughters. UK police routinely arrest native citizens for "hate speech."
England today would never deploy a Bond figure to solve any of these issues, maybe the opposite. It presents a difficult issue for the screen writers here; how do you write a "villain" that opposes a liberal multicultural hellscape without making them look like the accidental hero?
A realistic telling of a 2026+ Bond would look absurd, using all of his skills, charm, intelligence.... to do what exactly? As terrorism is accepted, foreign organized crime is accepted, international fraud is accepted, the MI6 organization does what?
Bond is a figure that no longer makes sense in such a society. I suppose they can go back to cold-war era tropes about evil Russians or something. But it's been done to death and is overly trite.
James Bond no longer has a place in society because society has accepted so much corruption and rot that it has criminalized anybody actually trying to fix things - especially in the way Bond does, extrajudicially. There's too much belief that needs suspended now... a realistic telling is that Bond kills a bad guy and the next scene is him on trial for war crimes at the Hague. It's all too absurd.
🌞Phillip Somerville retweeted
James Bond is incoherent in 2026 because the UK has neither the desire nor the ability to project power abroad or influence global affairs, they are busy shedding the last of their colonial possessions. Their main law enforcement priority is criminalizing dissent domestically


























