For over a decade, we’ve accepted that end-to-end backprop is the only way to train deep networks. But holding the entire network in memory all at once is why AI training is hitting a resource wall.
We found a new way to break the network into blocks and train them independently. The trick? Treating the network’s forward pass like a diffusion model denoising a signal.
This reinterpretation slashes the memory needed to train deep models. In our #ICLR2026 paper (https://t.co/PK5h0mqQSo), we matched end-to-end performance across ViTs, DiTs, and LLMs. We did this while training just one isolated block at a time.
I wonder what will happen in the economy if 80 million people who have reliably consumed 3500 calories of food each day for the last 30 years suddenly start eating only 1800 calories.
Turkish state lenders sold about $6 billion to defend the lira on Thursday, about half shortly after a court decision that removed the main opposition party’s leadership, according to traders familiar with the transactions. https://t.co/Iqafn3xBYa
A Turkish court removed the leader of the country’s main opposition party in a landmark ruling that could strengthen President Recep Tayyip Erdogan’s grip on power while risking political unrest and renewed market turmoil https://t.co/3uuljxGVBi
This is an extraordinary document written by the research arm of China's spy agency (the powerful MSS, basically the CIA and the FBI all wrapped in one) that absolutely zero media has picked up on.
As far as I can see, I'm the first person to write about it even though it was published (in Chinese) on May 13th on https://t.co/6uFjJ6ObmL, a website of China's Ministry of Foreign Affairs.
The document contains perhaps the most authoritative description of where China thinks its relationship with the U.S. stands, and where it’s headed.
The title of the report is “The Great Global Transformation and the Path to U.S.–China Coexistence” and I provide a full translation of it in my article, the link of which is at the bottom of this post.
To summarize briefly the most important - and, perhaps, surprising - aspect of the document: China's spy agency - the one institution whose entire job is to worry about the U.S. threat - has largely stopped worrying.
That's really what transpires from the document. They use a strategic framework borrowed from Mao's "protracted war" theory and, according to this framework, America's offensive phase is finished and China weathered the storm intact.
The question is no longer "how do we survive America?" but "how do we manage America?" - and they're proposing a six-step relationship recovery program.
I'll let you read the full document as well as my analysis of it here: https://t.co/vDvWFZJlrQ
People are realizing that AIs are nowhere near human intelligence and learning abilities.
Yet they have become very useful by compensating for their lack of common sense, lack of understanding of reality, and limited reasoning and planning abilities, by the accumulation of enormous amounts of declarative knowledge.
"The model didn’t 'invent' any 'new mathematics', [but] merely being able to know deeply all the results in a scientific field, and being able to use all known arguments expertly and with just the right choice of parameters, that alone can lead to a ton of breakthroughs, and this is not just limited to mathematics, this type of (extremely) solid expert execution is the bread and butter of many many scientific advances."
A sinking ship? Why the EU and China could be heading for a trade war
Fiery clashes at a conference in Beijing reflect wider tensions that threaten to descend into economic conflict - my report on rapidly worsening EU-China ties for our weekend paper
https://t.co/e9rMYfL20F
Bahceli, Erdogan, Simsek have estimated the cost of the war with the Kurdistan Worker's Party to be around 1.8 to 2.5 trillion dollars over the last four decades
In contrast with today, where Mardin Artiklu is preparing a Kurdish curriculum for Kurds in Syria and if the peace process continues, bilingual education for Kurds in Turkey