![IntuitMachine's tweet photo. MIT researchers have been digging into the "brains" of 60 different scientific AI models, and have stumbled upon something wild.
It turns out, whether an AI is reading text or looking at 3D atoms, they are all starting to agree on the same hidden truth about our universe.
Here is the pattern you can't unsee. 🧵
1/
First, the premise.
We have AI models for everything now.
• Some read protein sequences (like text).
• Some look at 3D crystal structures (like vision).
• Some predict forces in materials.
They are built differently. They are trained differently. They should think differently.
2/
But a new paper from MIT just asked a massive question:
"Are these models actually learning the same physics?"
The answer is yes. And it’s kind of spooky.
3/
The researchers took nearly 60 models—from LLMs reading SMILES strings to complex 3D potentials—and peered inside their latent spaces (their internal "thoughts").
They found that as models get smarter, their internal representations of matter start to look identical.
4/
Think of it like this:
If you ask a poet and a physicist to describe a sunset, they use different languages. But if they are both experts, they are describing the exact same reality.
The AI models are converging on a "Universal Representation of Matter."
5/
This chart in the paper is the smoking gun.
It shows that an LLM (trained on text) and a 3D Atomistic Model (trained on geometry) align almost perfectly when looking at molecules.
The text model "hallucinated" the 3D structure implicitly. It learned the physics just by reading the chemistry.
6/
But that's not even the most interesting part.
This convergence gives us a new way to spot "fake" intelligence.
The researchers found that high-performing models all cluster together in this "truth" space.
But the weak models? They scatter.
7/
It’s the Anna Karenina principle of AI:
"All happy (smart) models resemble one another; every unhappy (dumb) model is unhappy in its own way."
If a model diverges from the pack on standard data, it hasn't learned a new trick. It’s just lost in a local sub-optimum.
8/
However, there is a catch.
When the researchers threw "out-of-distribution" data at the weak models (stuff they hadn't seen before), the behavior flipped.
Instead of scattering, the weak models collapsed. They all started making the same low-information mistakes.
9/
This reveals a massive problem in Materials Science AI specifically.
The study shows these models are currently "data-governed." They are memorizing their specific training sets rather than learning universal laws.
They aren't "foundational" yet. They are just really good parrots.
10/
So, what does this mean for the future of Science?
Efficiency: We don't need massive, expensive, symmetry-enforcing architectures. We can "distill" the knowledge from big models into simple, fast ones.
Truth: We can use "alignment" to fact-check AI. If a model disagrees with the consensus of other top models, it's likely wrong.
11/
The most profound takeaway?
pattern-matching
And the fact that different AIs are independently deriving the same laws suggests that these models aren't just pattern matching.
They are uncovering reality.
12/
If this research holds up, in 5 years we won't distinguish between "protein models" and "materials models."
We will just have "Matter Models."
One foundation to simulate it all.
13/
This paper is a dense but rewarding read. It fundamentally changes how I think about "generality" in AI.
If you want to dive deeper, grab the PDF here: [Link to 2512.03750v1.pdf]
And SUBSCRIBE to me for more breakdowns of the science that is quietly changing the world.](https://pbs.twimg.com/media/G9fsOhYWwAAJ12I.jpg)
Olivia
Online
Rotman Criollo M.
@RotmanC
Only if we work together will we succeed. PhD hydrogeologist & geological engineer. Working at @IMEDEA_UIB_CSIC. Tweets are my own
Earth, planet
Joined January 2011
685 Following
258 Followers
1.6K Posts
Pinned Tweet
Back to work? Consider starting 2023 by submitting an abstract to #EGU23 session "ERE 2.6 - Lessons learned on interactions of groundwater and geothermal energy systems" https://t.co/SnD5wxg7tS
One week to deadline (January 10th)
#hydrogeology #GeothermalNews @iahgroundwater
Recibe El Jueves en casa cada mes con -60% de dto. Años de humor que merecen celebrarse.
"Heredamos este legado del terror de lo que se nos ha quedado en el cuerpo de lo que fue esa España y es necesario recuperar también el legado de resistencia".
La Ruta al Exilio es un recorrido de jóvenes por los rincones de nuestra memoria. Han estado en Andalucía en lugares cargados de historia y de dolor, pero también de dignidad y hemos hecho un diario de viaje con ellos.
Un programa especial para celebrar el 95 aniversario de nuestro #14ABRIL: https://t.co/gdo7C0wtSD
¿Qué efectos tuvo la regularización de 600.000 inmigrantes en España en 2004?
1⃣No tuvo efecto atracción.
2⃣Aumentó empleo formal de inmigrantes.
3⃣No afectó empleo nativos.
4⃣Aumentaron ingresos públicos en 4.000€ por inmigrante, sin aumentos de gasto público significativo.
Who to follow
IAHgroundwater
@iahgroundwater
Furthering the understanding, wise use and protection of groundwater resources throughout the world. Join, support or sponsor us now!
Vimal Mishra
@vmishraiit
Vikram Sarabhai Chair Professor, Civil Engineering and Earth Sciences, @iitgn, Editor, Earth’s Future, https://t.co/ZsZq4xCQcZ
FAO Asia Pacific
@FAOAsiaPacific
News and information from the Food and Agriculture Organization of the United Nations (@FAO) Regional Office for Asia and the Pacific.
MIT researchers have been digging into the "brains" of 60 different scientific AI models, and have stumbled upon something wild.
It turns out, whether an AI is reading text or looking at 3D atoms, they are all starting to agree on the same hidden truth about our universe.
Here is the pattern you can't unsee. 🧵
1/
First, the premise.
We have AI models for everything now.
• Some read protein sequences (like text).
• Some look at 3D crystal structures (like vision).
• Some predict forces in materials.
They are built differently. They are trained differently. They should think differently.
2/
But a new paper from MIT just asked a massive question:
"Are these models actually learning the same physics?"
The answer is yes. And it’s kind of spooky.
3/
The researchers took nearly 60 models—from LLMs reading SMILES strings to complex 3D potentials—and peered inside their latent spaces (their internal "thoughts").
They found that as models get smarter, their internal representations of matter start to look identical.
4/
Think of it like this:
If you ask a poet and a physicist to describe a sunset, they use different languages. But if they are both experts, they are describing the exact same reality.
The AI models are converging on a "Universal Representation of Matter."
5/
This chart in the paper is the smoking gun.
It shows that an LLM (trained on text) and a 3D Atomistic Model (trained on geometry) align almost perfectly when looking at molecules.
The text model "hallucinated" the 3D structure implicitly. It learned the physics just by reading the chemistry.
6/
But that's not even the most interesting part.
This convergence gives us a new way to spot "fake" intelligence.
The researchers found that high-performing models all cluster together in this "truth" space.
But the weak models? They scatter.
7/
It’s the Anna Karenina principle of AI:
"All happy (smart) models resemble one another; every unhappy (dumb) model is unhappy in its own way."
If a model diverges from the pack on standard data, it hasn't learned a new trick. It’s just lost in a local sub-optimum.
8/
However, there is a catch.
When the researchers threw "out-of-distribution" data at the weak models (stuff they hadn't seen before), the behavior flipped.
Instead of scattering, the weak models collapsed. They all started making the same low-information mistakes.
9/
This reveals a massive problem in Materials Science AI specifically.
The study shows these models are currently "data-governed." They are memorizing their specific training sets rather than learning universal laws.
They aren't "foundational" yet. They are just really good parrots.
10/
So, what does this mean for the future of Science?
Efficiency: We don't need massive, expensive, symmetry-enforcing architectures. We can "distill" the knowledge from big models into simple, fast ones.
Truth: We can use "alignment" to fact-check AI. If a model disagrees with the consensus of other top models, it's likely wrong.
11/
The most profound takeaway?
pattern-matching
And the fact that different AIs are independently deriving the same laws suggests that these models aren't just pattern matching.
They are uncovering reality.
12/
If this research holds up, in 5 years we won't distinguish between "protein models" and "materials models."
We will just have "Matter Models."
One foundation to simulate it all.
13/
This paper is a dense but rewarding read. It fundamentally changes how I think about "generality" in AI.
If you want to dive deeper, grab the PDF here: [Link to 2512.03750v1.pdf]
And SUBSCRIBE to me for more breakdowns of the science that is quietly changing the world.
![IntuitMachine's tweet photo. MIT researchers have been digging into the "brains" of 60 different scientific AI models, and have stumbled upon something wild.
It turns out, whether an AI is reading text or looking at 3D atoms, they are all starting to agree on the same hidden truth about our universe.
Here is the pattern you can't unsee. 🧵
1/
First, the premise.
We have AI models for everything now.
• Some read protein sequences (like text).
• Some look at 3D crystal structures (like vision).
• Some predict forces in materials.
They are built differently. They are trained differently. They should think differently.
2/
But a new paper from MIT just asked a massive question:
"Are these models actually learning the same physics?"
The answer is yes. And it’s kind of spooky.
3/
The researchers took nearly 60 models—from LLMs reading SMILES strings to complex 3D potentials—and peered inside their latent spaces (their internal "thoughts").
They found that as models get smarter, their internal representations of matter start to look identical.
4/
Think of it like this:
If you ask a poet and a physicist to describe a sunset, they use different languages. But if they are both experts, they are describing the exact same reality.
The AI models are converging on a "Universal Representation of Matter."
5/
This chart in the paper is the smoking gun.
It shows that an LLM (trained on text) and a 3D Atomistic Model (trained on geometry) align almost perfectly when looking at molecules.
The text model "hallucinated" the 3D structure implicitly. It learned the physics just by reading the chemistry.
6/
But that's not even the most interesting part.
This convergence gives us a new way to spot "fake" intelligence.
The researchers found that high-performing models all cluster together in this "truth" space.
But the weak models? They scatter.
7/
It’s the Anna Karenina principle of AI:
"All happy (smart) models resemble one another; every unhappy (dumb) model is unhappy in its own way."
If a model diverges from the pack on standard data, it hasn't learned a new trick. It’s just lost in a local sub-optimum.
8/
However, there is a catch.
When the researchers threw "out-of-distribution" data at the weak models (stuff they hadn't seen before), the behavior flipped.
Instead of scattering, the weak models collapsed. They all started making the same low-information mistakes.
9/
This reveals a massive problem in Materials Science AI specifically.
The study shows these models are currently "data-governed." They are memorizing their specific training sets rather than learning universal laws.
They aren't "foundational" yet. They are just really good parrots.
10/
So, what does this mean for the future of Science?
Efficiency: We don't need massive, expensive, symmetry-enforcing architectures. We can "distill" the knowledge from big models into simple, fast ones.
Truth: We can use "alignment" to fact-check AI. If a model disagrees with the consensus of other top models, it's likely wrong.
11/
The most profound takeaway?
pattern-matching
And the fact that different AIs are independently deriving the same laws suggests that these models aren't just pattern matching.
They are uncovering reality.
12/
If this research holds up, in 5 years we won't distinguish between "protein models" and "materials models."
We will just have "Matter Models."
One foundation to simulate it all.
13/
This paper is a dense but rewarding read. It fundamentally changes how I think about "generality" in AI.
If you want to dive deeper, grab the PDF here: [Link to 2512.03750v1.pdf]
And SUBSCRIBE to me for more breakdowns of the science that is quietly changing the world.](https://pbs.twimg.com/media/G9fsOheWYAAQoxu.jpg)
The Dark Side of Academia – 5 Ways to Spot Toxic Supervisors
(And how to protect your sanity + career)
-Thread-
¡Ya estamos en marcha!
Empieza #Coast2Coast2025: un día entero de ciencia joven, olas, sedimentos y muchas ganas de compartir.
Gracias a tod@s l@s que hacéis esto posible. ¡Nos vemos en el Campus Nord!
@la_UPC @icts_socib #CienciaCostera
#geology is getting lower relevance in your countries too?
🚀 New blog post alert! 📢 Want to make your #EGU25 session stand out? Check out these top tips on how to promote your session effectively! 💡📣
👉https://t.co/eM8NXwhqer
#EGU25 #ScienceCommunication #Geoscience #ConferenceTips
My passport is being detained at the Netherlands 🇳🇱 Embassy in Kampala 🇺🇬.
Last month I applied again for the Netherlands visa because I am supposed to give a keynote speech at the International Fundraising Congress even after refusing my first visa request for the same event.
But even after applying again, the embassy has refused to respond to my calls and emails since I submitted my new application and the event is happening next.
The embassy of the Netherlands rejected my first visa request because they did not believe I had enough ties to Uganda to ensure my return after the event. They also believed that I would stay in their country because of the current political situation in Uganda 🇺🇬
However, I have a UK 🇬🇧 visa activating on November 4th, and I am scheduled to speak at another event in London. And my USA visa is still valid till mid 2024.
When I contacted the embassy in Kampala, they informed me that they are not the ones who make visa decisions and that all decisions are made in Hague, Netherlands.
They suggested that I could appeal by shipping my passport to the address listed in the refusal letter.
However, there was also an option of applying again, so I decided to do so.
To my surprise, VFS Global sent me an email stating that my passport had been sent to the Netherlands embassy in Kampala which contradicted what the embassy had said that the decisions are not made by them. (Lies)
It has been over 15 working days, and I am still waiting for my passport. I need it to attend the conference next week, but no one is responding to me.
This situation is unfair since the embassy is trying to delay my passport and try to deny me a visa without any clear reason.
As an activist, I feel silenced without any clear reason. This was my second time applying for this visa, and I made sure to provide everything they needed.
It is not fair that we have to spend a lot of money applying for a Schengen visa only to be denied because the consular is not motivated as my role is to motivate unmotivated people.
I have already lost money twice in two different months. I would greatly appreciate any assistance from the people on this platform.
I have also tagged everyone who works at the Netherlands 🇳🇱embassy including the ambassador. at least let them be open and explain why am not being allowed to go and speak at this conference.
@NLinUganda @Karin_Boven @JantienZ #netherlands #visa #embassy
Estamos buscando a personas que deseen presentarse próximamente a convocatorias predoctorales (INPhINIT, FPU, CAM) y postdoctorales (#JdC-Formación)
Temas: #CambioGlobal #Biodiversidad y #Funcionamiento de ríos y lagos
Se agradece difusión. Gracias!!
¿Interesada/o? Más info⬇️
Ayudas atracción al talento:
https://t.co/KfHPqwi0sN
@postdoc_es
Divendres, 20 d'octubre...
Eliane Brum es una de las conciencias más lúcidas de nuestra época. Eso produce mucho dolor...
💧Turning a vision for #GroundwaterMonitoring into reality through Natural Capital Ecosystem Assessment programme #NCEA
We are in the process of recruiting 14 #groundwater technical specialists across England & the first 5 positions are now live. Deadline 23 October. Links⬇️
1/2
📢 Help us spread the word!🌍Please reshare this invitation
Let's bring together #Indigenous and non-Indigenous scholars to make a difference in #water & #EarthObservations
📌#GEOIndigenousWaterSummit
Lets drive positive change!
Submit your abstract by 🗓Oct. 15
More info 👇
What would the world be like without coffee? #InternationalCoffeeDay
🚨 Postdoc opportunity - I'm recruiting a postdoc to come work with me on drinking water access/affordability and climate adaption in California. Please share widely!
https://t.co/2ZVhmIQP9H
📢Oferta Treball - Serveis i Aigües de Barberà Empresa Municipal, S.A.U.: Tècnic/a Superior responsable de Tecnologies de la Informació i Comunicació
👉Més informació: https://t.co/TVfJtKZ470
#treball #ofertatreball #ofertalaboral #ofertadefeina #ofertatrabajo
Estamos a algo más de una semana de empezar el cuarto curso en la OIFem. Clases gratuitas de programación de todos los niveles para adolescentes españolas/residentes en España. Aún puedes inscribirte en https://t.co/e3yidKkxjY :)
So very true
Have you ever thought about this?
In 100 years like in 2123 we will all be buried with our relatives and friends.
Strangers will live in our homes we fought so hard to build, and they will own everything we have today. All our possessions will be unknown and unborn, including the car we spent a fortune on, and will probably be scrap, preferably in the hands of an unknown collector.
Our descendants will hardly or hardly know who we were, nor will they remember us. How many of us know our grandfather's father?
After we die, we will be remembered for a few more years, then we are just a portrait on someone's bookshelf, and a few years later our history, photos and deeds disappear in history's oblivion. We won't even be memories.
If we paused one day to analyse these questions, perhaps we would understand how ignorant and weak the dream to achieve it all was.
If we could only think about this, surely our approaches, our thoughts would change, we would be different people.
Always having more, no time for what's really valuable in this life. I'd change all this to live and enjoy the walks I've never taken, these hugs I didn't give, these kisses for our children and our loved ones, these jokes we didn't have time for. Those would certainly be the most beautiful moments to remember, after all they would fill our lives with joy.
And some of us waste it day after day with greed, selfishness and intolerance.
Every minute of life is priceless and will never be repeated, so take time to enjoy, be grateful for, and celebrate your existence.
Last Seen Users on Sotwe
Most Popular Users
Elon Musk 
@elonmusk
241.5M followers
Barack Obama
@barackobama
118.8M followers
Cristiano Ronaldo
@cristiano
114.5M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.3M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.5M followers
Kim Kardashian
@kimkardashian
70.9M followers
YouTube
@youtube
68.8M followers
Neymar Jr
@neymarjr
66.5M followers
Bill Gates
@billgates
65.3M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.7M followers
X
@x
60.7M followers
✨
⭐
💫