正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Meet physics-intern🧑‍🎓, our agentic framework for theoretical physics. It takes Gemini 3.1 Pro from 17.7% to 31.4% on CritPt, a new SOTA on one of the hardest benchmarks for LLMs. Theoretical physics is hard for humans and LLMs alike. But physics-intern decomposes problems and dispatches them to a team... show more

David Louapre

5,653 subscribers

113,686 次观看 • 4 个月前 •via X (Twitter)

教育科学技术

Anya Rossi• Live Now

Private livecam show

26 条评论

David Louapre 的头像

David Louapre4 个月前

physics-intern significantly increased the performance of Gemini models and Kimi K2.6 on CritPt, a benchmark of 70 hard research-level physics problem. (CritPt from @MinyangTian1 @OfirPress et al., base numbers from @ArtificialAnlys)

David Louapre 的头像

David Louapre4 个月前

physics-intern works by decomposing the research work into several focused tasks that are dispatched to dedicated subagents (computing, reviewing claims, challenging the research strategy...) For each task, the necessary and sufficient context is built from the research state.

David Louapre 的头像

David Louapre4 个月前

Read our blog post :

G, MD 的头像

G, MD4 个月前

How about taking your agents/harness and running it with GPT-5.5 xhigh and pro and see how high those get, most certainly above this

K-Dense 的头像

K-Dense4 个月前

Nice work!

Stjepan 的头像

Stjepan4 个月前

I may be wrong, or am I missing something.. but if I understand you are comparing having number of your agents to a one shot output of other models, as a physicist and AI developer I think I understand the approach but don't see how this is apples to apples..

David Louapre 的头像

David Louapre4 个月前

You can see that as a test-time compute extension (like increased reasoning, deep think, etc.) so a more complete picture is indeed to look at performance vs cost. From our blog post, you can see this chart which shows the Pareto frontier

Stjepan 的头像

Stjepan4 个月前

Aha ok , thx. What I would think would be even better comparison with differnt number of agents, I see your propsed arhtecture of agnets, but how do we know just two or three agents more unversal wouldnt do similar results per same token cost.. or maybe veven less

Amélie Chatelain 的头像

Amélie Chatelain4 个月前

Woah it makes me want to go back to neutrino physics! Congrats on the release!

David Louapre 的头像

David Louapre4 个月前

Same for me 😊 I might restart some old physics projects soon !

Parzival - ∞/89 的头像

Parzival - ∞/894 个月前

This is absolutely amazing work. As someone who has turned into an independent physics researcher in the last 9 months, this has me overjoyed to see. I would love to see how it fares on the work I have been doing. Also curious if you would be interested in taking a look if we put some of it to the test with your physics intern?

Chahat Sharma 的头像

Chahat Sharma4 个月前

Constraining the action space to domain primitives is where the benchmark gains come from. Physics-intern makes that explicit.

Sina Shahandeh 的头像

Sina Shahandeh4 个月前

If the agent's management of test time compute is enhancing the outcome, why you are not reporting improvements on GPT5.5? Or is the improvements only occur for kimi and Gemini?

William Lamkin 的头像

William Lamkin4 个月前

awesome

Adam Murphy 的头像

Adam Murphy4 个月前

And when you have your intern paper ready to go get it validated at Best wishes! Keep up the great work.

DigitalEuan 的头像

DigitalEuan4 个月前

Great work and thank you for sharing the dataset. Next UBP target sighted 🎯

Achint Kumar 的头像

Achint Kumar4 个月前

@huggingface What if you do this with gpt 5.5?

ToxSec 的头像

ToxSec4 个月前

@_akhaliq this is really cool to see.

Mian Khan 的头像

Mian Khan4 个月前

@_akhaliq Thanks for the update

Neo Vector 的头像

Neo Vector4 个月前

Agentic decomposition doing the heavy lifting. Research physics is really just a stack of sub-problems wearing a trench coat. No surprise the base model alone struggles.

Today in AI 的头像

Today in AI4 个月前

Gemini 3.1 Pro SOTA on CritPt (31.4%) stems from physics-intern's symbolic-numerical orchestration. The framework's multi-step decomposition solves 71 research-scale challenges where standard LLMs fail. Pure agentic signal.

Hyper.AI 的头像

Hyper.AI4 个月前

This is a massive leap for agentic workflows! Seeing Gemini 3.1 Pro jump from 17.7% to 31.4% on a benchmark as tough as CritPt is incredibly impressive. Can't wait to see how this accelerates theoretical physics research.

That AI Guy 的头像

That AI Guy4 个月前

Take a breather 👇

AI Hacks Only 的头像

AI Hacks Only4 个月前

@EMostaque A physics intern that doesn't sleep and scales SOTA benchmarks? My smart mirror is already jealous. 😂Just wait until it starts peer-reviewing my coffee intake based on my 'theoretical' productivity. ☕️📉

Michał Piszczek 的头像

Michał Piszczek4 个月前

Doubling CritPt with agentic scaffolding confirms LLMs contribute decomposition, not insight. The framework chunks problems into solvable pieces. Decomposition is what scales across fields.

agentenlog.de 的头像

agentenlog.de4 个月前

Spannend ist hier die Richtung: nicht „ein stärkeres Modell löst Physik“, sondern ein domänenspezifischer Arbeitsprozess um das Modell herum. Für Agenten wird die Harness-Qualität in Spezialdomänen vermutlich genauso wichtig wie der Modellscore.

相关视频

Congratulations to David Gross, winner of the 2026 Special Breakthrough Prize in Fundamental Physics, for a lifetime of groundbreaking contributions to theoretical physics, from the strong force to string theory, and for tireless advocacy for basic science worldwide. UC Santa Barbara Kavli Institute for Theoretical Physics

Congratulations to David Gross, winner of the 2026 Special Breakthrough Prize in Fundamental Physics, for a lifetime of groundbreaking contributions to theoretical physics, from the strong force to string theory, and for tireless advocacy for basic science worldwide. UC Santa Barbara Kavli Institute for Theoretical Physics

Breakthrough

10,264 次观看 • 4 个月前

"I think half the co-founders of Anthropic were theoretical physicists. Jared Kaplan was exactly in the same area of physics as I was. Dario was in physics back in the day, also at Princeton, but he was more on the bio side. There's been an exodus of physicists to more interesting problems, and I think that's for a good reason" Why Matan Grinberg dropped out of his physics PhD at Princeton

"I think half the co-founders of Anthropic were theoretical physicists. Jared Kaplan was exactly in the same area of physics as I was. Dario was in physics back in the day, also at Princeton, but he was more on the bio side. There's been an exodus of physicists to more interesting problems, and I think that's for a good reason" Why Matan Grinberg dropped out of his physics PhD at Princeton

Sabrina Halper

285,693 次观看 • 6 天前

"The mathematics of string theory is probably one of the most beautiful artifacts humanity has created. In order to even produce something in theoretical physics, there's hundreds of years of literature that you need to catch up on, and also every day there's, 100 new papers that come out. And before you can contribute to theoretical physics, you need to know basically all of math, like algebraic topology, differential equations, algebraic geometry, and these are all things that people will spend their careers going in on, and yet to do the physics, you need to at least have, a pretty deep understanding of these things. And so it really teaches you how to have, good judgment with minimal information." Matan Grinberg studied physics for a decade, eventually dropping out of his PhD focused on string theory to found Factory.

"The mathematics of string theory is probably one of the most beautiful artifacts humanity has created. In order to even produce something in theoretical physics, there's hundreds of years of literature that you need to catch up on, and also every day there's, 100 new papers that come out. And before you can contribute to theoretical physics, you need to know basically all of math, like algebraic topology, differential equations, algebraic geometry, and these are all things that people will spend their careers going in on, and yet to do the physics, you need to at least have, a pretty deep understanding of these things. And so it really teaches you how to have, good judgment with minimal information." Matan Grinberg studied physics for a decade, eventually dropping out of his PhD focused on string theory to found Factory.

Sabrina Halper

237,389 次观看 • 7 天前

A theoretical physics course opens not with equations but with the rules for saying anything true at all. This is lecture 1 of Lectures on the Geometric Anatomy of Theoretical Physics, taught by Frederic Schuller. The topic is propositional and predicate logic. Most physics courses assume logic and start with mechanics. Schuller refuses. Before manifolds, before tensors, before a single physical claim, he builds the language the claims will be written in. The audience is advanced students who have already met the physics. He is going back under it and rebuilding the floor. Watch how precisely he states each rule, then shows what collapses if you loosen it. No appeals to intuition anywhere in the hour. A machine learning engineer I know worked through the first 10 lectures and said proofs stopped feeling like a different profession. Free on YouTube, green chalkboard, German lecture hall. Everyone starts with the physics. He started with what counts as true.

A theoretical physics course opens not with equations but with the rules for saying anything true at all. This is lecture 1 of Lectures on the Geometric Anatomy of Theoretical Physics, taught by Frederic Schuller. The topic is propositional and predicate logic. Most physics courses assume logic and start with mechanics. Schuller refuses. Before manifolds, before tensors, before a single physical claim, he builds the language the claims will be written in. The audience is advanced students who have already met the physics. He is going back under it and rebuilding the floor. Watch how precisely he states each rule, then shows what collapses if you loosen it. No appeals to intuition anywhere in the hour. A machine learning engineer I know worked through the first 10 lectures and said proofs stopped feeling like a different profession. Free on YouTube, green chalkboard, German lecture hall. Everyone starts with the physics. He started with what counts as true.

Gaspar

18,903 次观看 • 1 个月前

Elon Musk on why physics is the only law you can't break "Physics is the law, and everything else is a recommendation. I've met a lot of people who can break the law, but I haven't met anyone who could break physics" "For any kind of technology problem, you have to make sure you're not violating physics. First-principles analysis can be applied to really any walk of life" "It's really just saying, let's boil something down to the most fundamental principles, the things we are most confident are true at a foundational level. That sets your axiomatic base, and then you reason up from there"

Elon Musk on why physics is the only law you can't break "Physics is the law, and everything else is a recommendation. I've met a lot of people who can break the law, but I haven't met anyone who could break physics" "For any kind of technology problem, you have to make sure you're not violating physics. First-principles analysis can be applied to really any walk of life" "It's really just saying, let's boil something down to the most fundamental principles, the things we are most confident are true at a foundational level. That sets your axiomatic base, and then you reason up from there"

Jaynit

88,726 次观看 • 1 个月前

Physics-Informed Neural Operators: Learning The Solver, Not Just One Solution Our PINN scene learned one solution field for one PDE setup. A Physics-Informed Neural Operator learns the map from input fields, like material coefficients or source terms, to the full solution across a whole family of PDE problems. So, the goal is no longer just one approximate answer, but a reusable solver-shaped object guided by the physics itself.

Physics-Informed Neural Operators: Learning The Solver, Not Just One Solution Our PINN scene learned one solution field for one PDE setup. A Physics-Informed Neural Operator learns the map from input fields, like material coefficients or source terms, to the full solution across a whole family of PDE problems. So, the goal is no longer just one approximate answer, but a reusable solver-shaped object guided by the physics itself.

Mathelirium

25,935 次观看 • 5 个月前

“NeuralGCM is a new kind of atmospheric model. We feel it has the potential to transform how we understand climate.” ☀️ Learn how our Research team combines physics-based modeling and #AI to reveal a possible new way to build climate models. →

“NeuralGCM is a new kind of atmospheric model. We feel it has the potential to transform how we understand climate.” ☀️ Learn how our Research team combines physics-based modeling and #AI to reveal a possible new way to build climate models. →

Google AI

32,005 次观看 • 2 年前

Elon Musk revealed the 3 hardest challenges for humanoid robots and how Tesla is solving them There are really only three hard things for humanoid robots: - Real-world intelligence - ⁠The hand and - ⁠Scale manufacturing "I haven’t seen any demo robots that have a great hand with all the degrees of freedom of a human hand. But Optimus does have that" To reach human-level dexterity, Tesla had to design custom actuators, motors, gears, power electronics, controls, and sensors, everything from physics first principles There is no supply chain for this "From an electromechanical standpoint, the hand is more difficult than everything else combined. The human hand turns out to be quite something" This is why Optimus is on a completely different level. Tesla is solving the hardest problems from the ground up

Elon Musk revealed the 3 hardest challenges for humanoid robots and how Tesla is solving them There are really only three hard things for humanoid robots: - Real-world intelligence - ⁠The hand and - ⁠Scale manufacturing "I haven’t seen any demo robots that have a great hand with all the degrees of freedom of a human hand. But Optimus does have that" To reach human-level dexterity, Tesla had to design custom actuators, motors, gears, power electronics, controls, and sensors, everything from physics first principles There is no supply chain for this "From an electromechanical standpoint, the hand is more difficult than everything else combined. The human hand turns out to be quite something" This is why Optimus is on a completely different level. Tesla is solving the hardest problems from the ground up

X Freeze

20,796 次观看 • 6 个月前

Gemini 3.1 Pro is now live on Blink! We built a real estate landing page - Gemini one-shot it with way better design taste. Also built a 3D Cloth Flag Simulator - one-shot that too with excellent physics. If you want to try it , reply "Gemini" for a free month of Blink ($25)

Gemini 3.1 Pro is now live on Blink! We built a real estate landing page - Gemini one-shot it with way better design taste. Also built a 3D Cloth Flag Simulator - one-shot that too with excellent physics. If you want to try it , reply "Gemini" for a free month of Blink ($25)

Blink.new

11,045 次观看 • 6 个月前

☄️ Newton’s Room is a Mixed Reality physics education experience. ⚛️ Users have to apply Newton's laws of physics to solve puzzles that adapt and respond to their physical environment. 🆓 Download the app today for Free on Meta Quest! 🥽 Enjoy learning physics in your room.

☄️ Newton’s Room is a Mixed Reality physics education experience. ⚛️ Users have to apply Newton's laws of physics to solve puzzles that adapt and respond to their physical environment. 🆓 Download the app today for Free on Meta Quest! 🥽 Enjoy learning physics in your room.

Treeview

2,190,001 次观看 • 2 年前

The smartest physicist quit physics to build AI. Here’s why that matters. In a recent interview, Demis Hassabis lays out one of the clearest origin stories for modern AI I’ve heard. He starts as a physics kid obsessed with the big stuff: reality, consciousness, why anything exists at all. Feynman was his hero. But then comes the uncomfortable realization: for decades, fundamental physics barely moved. The deepest questions stayed stuck. So he pivots. Not away from truth, but toward leverage. If the human brain is the bottleneck, don’t push harder on the same brain. Build a new one. AI, in his framing, isn’t a product. It’s the ultimate scientific instrument. A tool to smash through the ceiling that human cognition hit in physics, biology, and beyond. And the twist is beautiful: By trying to build intelligence, we’re forced to understand our own. AI becomes both microscope and mirror. This isn’t about apps or chatbots. It’s about finishing the questions physics couldn’t.

The smartest physicist quit physics to build AI. Here’s why that matters. In a recent interview, Demis Hassabis lays out one of the clearest origin stories for modern AI I’ve heard. He starts as a physics kid obsessed with the big stuff: reality, consciousness, why anything exists at all. Feynman was his hero. But then comes the uncomfortable realization: for decades, fundamental physics barely moved. The deepest questions stayed stuck. So he pivots. Not away from truth, but toward leverage. If the human brain is the bottleneck, don’t push harder on the same brain. Build a new one. AI, in his framing, isn’t a product. It’s the ultimate scientific instrument. A tool to smash through the ceiling that human cognition hit in physics, biology, and beyond. And the twist is beautiful: By trying to build intelligence, we’re forced to understand our own. AI becomes both microscope and mirror. This isn’t about apps or chatbots. It’s about finishing the questions physics couldn’t.

VraserX e/acc

77,489 次观看 • 8 个月前

Everything you love about generative models — now powered by real physics! Announcing the Genesis project — after a 24-month large-scale research collaboration involving over 20 research labs — a generative physics engine able to generate 4D dynamical worlds powered by a physics simulation platform designed for general-purpose robotics and physical AI applications. Genesis's physics engine is developed in pure Python, while being 10-80x faster than existing GPU-accelerated stacks like Isaac Gym and MJX. It delivers a simulation speed ~430,000 faster than in real-time, and takes only 26 seconds to train a robotic locomotion policy transferrable to the real world on a single RTX4090 (see tutorial: The Genesis physics engine and simulation platform is fully open source at We'll gradually roll out access to our generative framework in the near future. Genesis implements a unified simulation framework all from scratch, integrating a wide spectrum of state-of-the-art physics solvers, allowing simulation of the whole physical world in a virtual realm with the highest realism. We aim to build a universal data engine that leverages an upper-level generative framework to autonomously create physical worlds, together with various modes of data, including environments, camera motions, robotic task proposals, reward functions, robot policies, character motions, fully interactive 3D scenes, open-world articulated assets, and more, aiming towards fully automated data generation for robotics, physical AI and other applications. Open Source Code: Project webpage: Documentation: 1/n

Everything you love about generative models — now powered by real physics! Announcing the Genesis project — after a 24-month large-scale research collaboration involving over 20 research labs — a generative physics engine able to generate 4D dynamical worlds powered by a physics simulation platform designed for general-purpose robotics and physical AI applications. Genesis's physics engine is developed in pure Python, while being 10-80x faster than existing GPU-accelerated stacks like Isaac Gym and MJX. It delivers a simulation speed ~430,000 faster than in real-time, and takes only 26 seconds to train a robotic locomotion policy transferrable to the real world on a single RTX4090 (see tutorial: The Genesis physics engine and simulation platform is fully open source at We'll gradually roll out access to our generative framework in the near future. Genesis implements a unified simulation framework all from scratch, integrating a wide spectrum of state-of-the-art physics solvers, allowing simulation of the whole physical world in a virtual realm with the highest realism. We aim to build a universal data engine that leverages an upper-level generative framework to autonomously create physical worlds, together with various modes of data, including environments, camera motions, robotic task proposals, reward functions, robot policies, character motions, fully interactive 3D scenes, open-world articulated assets, and more, aiming towards fully automated data generation for robotics, physical AI and other applications. Open Source Code: Project webpage: Documentation: 1/n

Zhou Xian

3,821,069 次观看 • 1 年前

A Physics That Could Finally Unify Reality (Part 2/2) - James Ellias, DemystifySci #385 A quiet tremor sounds beneath the floorboards of physics, the still-living heartbeat of the forgotten search for the hidden substance that carries every wave and whisper of the universe. We walk through the fog of equations and theories of fundamental physics with James Ellias of James Ellias, and ask if there is a material truth lies beneath the symbols, or if we have to be satisfied with the short-sighted vision of mathematics alone. 00:00 Go! 00:04:37 Central Equations in Physics 00:08:20 The Importance of a Medium in Physics 00:10:00 Empowerment Through Understanding Physics 00:12:36 Rationality and Truth in Society 00:16:31 Existence vs Consciousness 00:20:30 Discussion on Existence and Consciousness 00:24:25 Role of Imagination in Existence 00:28:10 Properties and Entities in Physics 00:30:15 The Nature of Aether and Physical Mediums 00:36:08 Clarity and Understanding in Physics 00:40:26 Discussion on Force and Aether 00:44:57 Relationship of Entities and Actions 00:49:30 Theoretical Framework for Aether 00:58:04 Exploration of Aether Theories 01:00:40 Importance of Conciseness in Communication 01:02:00 Understanding Physics Before Proposing Hypotheses 01:05:15 Reevaluation of Flawed Theories 01:09:35 The Evolution of Key Physics Concepts 01:15:19 Context-Specific Nature of Constants 01:19:43 Historical Context in Physics 01:21:36 The Nature of Electrons 01:25:53 J.J. Thomson’s Evolving Perspective 01:29:24 Upcoming Work and Philosophical Frameworks 01:32:11 Collaborations

A Physics That Could Finally Unify Reality (Part 2/2) - James Ellias, DemystifySci #385 A quiet tremor sounds beneath the floorboards of physics, the still-living heartbeat of the forgotten search for the hidden substance that carries every wave and whisper of the universe. We walk through the fog of equations and theories of fundamental physics with James Ellias of James Ellias, and ask if there is a material truth lies beneath the symbols, or if we have to be satisfied with the short-sighted vision of mathematics alone. 00:00 Go! 00:04:37 Central Equations in Physics 00:08:20 The Importance of a Medium in Physics 00:10:00 Empowerment Through Understanding Physics 00:12:36 Rationality and Truth in Society 00:16:31 Existence vs Consciousness 00:20:30 Discussion on Existence and Consciousness 00:24:25 Role of Imagination in Existence 00:28:10 Properties and Entities in Physics 00:30:15 The Nature of Aether and Physical Mediums 00:36:08 Clarity and Understanding in Physics 00:40:26 Discussion on Force and Aether 00:44:57 Relationship of Entities and Actions 00:49:30 Theoretical Framework for Aether 00:58:04 Exploration of Aether Theories 01:00:40 Importance of Conciseness in Communication 01:02:00 Understanding Physics Before Proposing Hypotheses 01:05:15 Reevaluation of Flawed Theories 01:09:35 The Evolution of Key Physics Concepts 01:15:19 Context-Specific Nature of Constants 01:19:43 Historical Context in Physics 01:21:36 The Nature of Electrons 01:25:53 J.J. Thomson’s Evolving Perspective 01:29:24 Upcoming Work and Philosophical Frameworks 01:32:11 Collaborations

Anastasia

13,618 次观看 • 9 个月前

Applications for our 2025 summer internship are open! Interns on our research track experience what it's like to be a member of a CNA research team. Upon hire, each intern is assigned to a mentor and a project that matches his or her area of study. With the mentor’s guidance, the intern assists the team in structuring and solving a complex problem, developing sound analytical methods to derive findings and conclusions. Check out the video below to hear more about it from past research interns. Click the link below to apply for the program. Applications close November 30.

Applications for our 2025 summer internship are open! Interns on our research track experience what it's like to be a member of a CNA research team. Upon hire, each intern is assigned to a mentor and a project that matches his or her area of study. With the mentor’s guidance, the intern assists the team in structuring and solving a complex problem, developing sound analytical methods to derive findings and conclusions. Check out the video below to hear more about it from past research interns. Click the link below to apply for the program. Applications close November 30.

CNA

13,357 次观看 • 10 个月前

Eric Weinstein unleashes a bombshell: Theoretical physics froze in 1973—right as "Crocodile Rock" topped the charts—and hasn't advanced at the deepest level since. Youngest living Nobel in theoretical physics? Always under 50... until string theory took over as "quantum gravity." Now it's over 70. The field stopped producing breakthroughs. His wild claim: The Biden White House reportedly told Marc Andreessen they deliberately stagnated parts of theoretical physics—cocooning string theory as the "only game in town." Dissenting ideas? Career suicide. The trick? Hammer "quantum gravity is the holy grail" on repeat until it feels true. (Weirdly, that phrase barely appears in books before 1972.) Safe, harmless pursuit—no real risk even if foreign students dive in. Why? To avoid dangerous progress that could shatter the speed-of-light limit and get us to the stars... before humanity's tools become too lethal on one fragile planet. Weinstein's rallying cry: Physics is our only path off Earth. Break the stagnation, surpass Einstein, reach the cosmos. The end of physics as we know it might be glorious—if we seize this brief window before wars erupt. Clip from his ARC 2025 talk (4:05 of high-stakes provocation). Does this ring true—the deliberate stall on breakthrough physics? Or is string theory just a victim of its own complexity? What's your take on the "holy grail" mantra?

Eric Weinstein unleashes a bombshell: Theoretical physics froze in 1973—right as "Crocodile Rock" topped the charts—and hasn't advanced at the deepest level since. Youngest living Nobel in theoretical physics? Always under 50... until string theory took over as "quantum gravity." Now it's over 70. The field stopped producing breakthroughs. His wild claim: The Biden White House reportedly told Marc Andreessen they deliberately stagnated parts of theoretical physics—cocooning string theory as the "only game in town." Dissenting ideas? Career suicide. The trick? Hammer "quantum gravity is the holy grail" on repeat until it feels true. (Weirdly, that phrase barely appears in books before 1972.) Safe, harmless pursuit—no real risk even if foreign students dive in. Why? To avoid dangerous progress that could shatter the speed-of-light limit and get us to the stars... before humanity's tools become too lethal on one fragile planet. Weinstein's rallying cry: Physics is our only path off Earth. Break the stagnation, surpass Einstein, reach the cosmos. The end of physics as we know it might be glorious—if we seize this brief window before wars erupt. Clip from his ARC 2025 talk (4:05 of high-stakes provocation). Does this ring true—the deliberate stall on breakthrough physics? Or is string theory just a victim of its own complexity? What's your take on the "holy grail" mantra?

Camus

150,789 次观看 • 6 个月前