Incredible!
AI begins predicting the entire universe—
LLMs predict the linguistic description of the world, video world models predict (dynamic) images, and this time directly predict the four-dimensional physical state of space and time.
Most strikingly, it employs an entirely new architecture called "Neural Operator," rather than the widely dominant Transformer.
Output the complete four-dimensional trajectory, including the time dimension, in a single inference.
The required context length is undoubtedly extremely large, and now the startup Accelerated Understanding has increased the training context to the trillion (1T) level and exceeded 5 trillion during inference!
What does 5 trillion mean? It’s 5 million times the capacity of today’s top LLMs! Imagine asking an AI to read War and Peace 5 million times in one go—and remember every word.
Think about it, really think about it.
Time isn't fundamental—it's more like an emergent phenomenon arising from a deeper informational structure. Isn't that interesting? If you push this idea to its extreme, it amounts to generating the entire physical universe.
More astonishingly, the two founders of this startup have just turned down a lucrative offer from former world首富 Jeff Bezos, which included a 35% equity stake, a $2 million annual salary, and a commitment to over $2 billion in funding.
Behind them, the "spiritual shareholders" and possible hidden forces may well be the man in the leather jacket—NVIDIA CEO Jensen Huang.
Five trillion context is not Ilya's "safe superintelligence"
Yesterday, a prominent Silicon Valley investor spoke vaguely, sparking widespread speculation online that Ilya might be preparing a major move with his SSI this time.
And now the mystery seems to be solved—it’s not Ilya, or even a language model, but a true “universe generator.”
The alarming data on Accelerated Understanding has been widely shared:
- Model parameter scale: Pre-training of models with up to 1 trillion parameters has been completed.
- Extreme scalability: The scale of the experiment has surged to 35 trillion parameters.
- Training period context: 1 Trillion
- Reasoning context: Over 5 trillion.
With such a vast context to support it, this model demonstrates alarming capabilities:
It does not require sub-sampling or chunking when performing physical inference.
Facing an extremely complex 4D spatial physical system, it can generate the complete motion trajectory of the entire spacetime in a single inference!
Moreover, due to the unified underlying logic, the same model can simultaneously handle completely different physical problems!
This was unimaginable before.
In the past, meteorologists used weather models, and materials scientists used materials models; now, Accelerated Understanding tells the world: physical laws are fundamentally interconnected—a single universal physics model is sufficient to solve everything.
The most groundbreaking core feature of Accelerated Understanding lies in its creation of a perfect "simulate → improve → simulate" feedback loop. It is not merely a "generator," but a true "Reality Validator."
Moreover, this targeted feedback also aligned with Ilya's training during inference.
"Is the Transformer Dead?" The Awakening of Physical AI
If you ask any AI professional today, "What's the most popular AI architecture right now?" the answer will be 100%: Transformer.
From ChatGPT to major video generation models, this architecture, invented by Google, has dominated the AI world over the past few years—the very “T” in ChatGPT.
But in Anima's view, the Transformer approach has gone off track.
“Language-centered intelligence is ‘human-centered,’” Anima sharply noted in an exclusive interview ahead of the launch, “while placing physics at the center is a ‘nature-centered’ perspective.”
How do current LLMs and video world models work? They essentially play a "game of probability."
ChatGPT predicts the next most likely word.
Those impressive video-generating AIs are essentially taking visual shortcuts—they only make the generated images appear to follow physical laws, but if you examine the underlying gravity, fluid dynamics, or material tension, you’ll find everything is wrong.
They fundamentally don't understand physics—they're just advanced 'pixel repeaters'.
Text has only one dimension, video has three dimensions, while real-world spacetime consists of three spatial dimensions plus one time dimension.
In addition, many models use an autoregressive approach—predicting frame by frame, with errors accumulating progressively.
Therefore, Accelerated Understanding completely rejects the Transformer and also rejects visual shortcuts.
Their ultimate weapon is a revolutionary technology—Neural Operators—that Anima helped pioneer many years ago.
Unlike text processing, neural operators are specifically designed to handle complex, invisible physical data. They do not require forcing the physical world into reduced dimensions such as text or pixels; instead, they directly understand multiphysics phenomena within the full 4D dimension (3D space + time).
Invisible airflows, plasma turbulence inside nuclear fusion reactors, microscopic thermodynamic distributions within chips... these physical processes, invisible to the human eye and impossible for traditional AI to infer using "visual experience," have become clear and computable under this new architecture.
Shocked Jensen Huang, rejected Bezos
At this point, you might be wondering: how much computing power is needed to train such a powerful model, and where does the funding come from?
Although declining to disclose current funding details, the co-founder subtly indicated that "we have partnered with computing providers who have supplied hardware clusters to develop and run AI."
But all the clues point to her former employer—NVIDIA—and Jensen Huang, who proclaimed, “AI is the future.”
Go back to 2018, Anima was hired by NVIDIA as Director of AI Research.
Over five years, she led a team of top scientists to explore how to apply NVIDIA's GPU computing power to cutting-edge AI fields.
At the time, the team developed an early breakthrough project: using AI to accelerate weather forecasting. The results showed that AI predictions were just as accurate as those made by meteorologists using extremely complex traditional computational fluid dynamics models—but orders of magnitude faster.
This result directly "shattered" Jensen Huang.
At the 2021 NVIDIA GTC conference, Jensen Huang took the stage to showcase the Anima team’s research on “neural operators.” “He was so excited at the time,” recalled Jensen Huang.
When Anima half-jokingly told Old Huang, "AI might steal the lunch of theoretical physicists."
At the time, Jensen Huang's eyes lit up as he confidently replied, "I want it to eat all their lunch!"
According to Anima, it was Jensen Huang who initially encouraged her to pursue and realize this wild idea of "physical AI."
Although NVIDIA has not yet responded to Reuters’ inquiry about whether it invested in the company, within the AI community, it is widely understood: no massive model with a 5-trillion-context capacity could have been developed without the covert support of top-tier computing power.
Old Huang has finally made his move.
However, to understand just how impressive this "physical large model" is, we need to go back to Los Angeles at the end of 2024.
At an upscale restaurant, a secret dinner is taking place that could reshape the current AI landscape.
One of the highlights of the dinner was investor and biotech entrepreneur Vik Bajaj, who later co-founded Project Prometheus, a company valued at over $10 billion, with Bezos.
Across from him sat a husband-and-wife team with top-tier AI backgrounds: Anima Anandkumar, Professor of Computing and Mathematical Sciences at Caltech and former Director of AI Research at NVIDIA, and her husband, leading AI infrastructure engineer Benedikt Jenik.
Bajaj arrived with an offer titled "Project Prometheus," a deal backed by Bezos that offered terms so generous they were astonishing:
Invite Anima to serve as the company’s spokesperson, board member, and lead architect of the scientific vision;
The couple will receive up to 35% of the company’s equity;
Base annual salary of $1 million, doubling to $2 million after three months of employment;
Most strikingly, the protocol explicitly states that investors, including Bezos, will provide over $2 billion in committed funding for financing prior to the Series B round.
In an era where funding for large models is increasingly difficult and compute costs are high, this is a super golden ticket to financial freedom and industry power.
However, Anima and Jenik chose to decline.
Why? Because what they hold has far greater potential than a company that automates the manufacturing of complex physical systems. They don’t want to be subordinate to others, nor do they want their technological vision hijacked by capital. What they aim to build is a "god-level model" capable of directly understanding and simulating the physical universe.
In the end, the Anima couple quietly lay low until today, surprising the world with their independently developed Accelerated Understanding.
Conclusion: A Paradigm Shift from "Generating Information" to "Optimizing the World"
August 25, 2026, is destined to be a historic day in the annals of AI development.
While we celebrate large language models for writing beautiful poetry and helping us polish work emails, Accelerated Understanding arrives like a wall-breaker from the future, coldly reminding us: language and images are merely surface manifestations of human understanding; the laws of physics are the true underlying code of the universe.
Rejecting Bezos, rejecting Transformers, abandoning visual shortcuts, they get to the essence of 4D spacetime. With a staggering 1 trillion parameters and 5 trillion context, Anima Anandkumar and Benedikt Jenik announce to the world the arrival of the "Physical AI" era.
Reference materials:
https://x.com/AndrewCurran_/status/2092244031002771643
https://x.com/AnimaAnandkumar/status/2092236528898675014
https://x.com/daniel_mac8/status/2092303081287418277
https://www.reuters.com/business/ai-founders-who-walked-away-bezos-backed-prometheus-model-universe-2026-08-25/
https://acceleratedunderstanding.com/
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation, edited by David.