UFO-Z☀️ @ufobtc27

@Boca_Tina Congrats Tina

@xiangyuli 哈哈哈哈

@tig88411109 @bearliu 哈哈哈哈

@chenna1985 Good 👍 also including DFT, MD

Artificial intelligence is not replacing human intuition in maths and physics, but reimagining how questions are asked, explored and understood go.nature.com/4vG4Vsl

x.com/i/article/2064…

@BetaTomorrow 好文章

@wilmingzhou 6

Boltzmann generators go big: scalable equilibrium sampling for materials Predicting whether a crystal phase is stable, or computing its free energy, usually means running molecular dynamics for a long time. The problem is that MD explores configuration space one step at a time, so consecutive snapshots are correlated and the system mixes slowly. Boltzmann Generators offered a clever alternative: train a normalizing flow to map random noise straight onto the Boltzmann distribution, giving uncorrelated equilibrium samples in one shot, plus free energies through reweighting. The catch was scale. Earlier BGs maxed out around 500 atoms, with poor sampling efficiency and training runs approaching a full GPU-year, slower than the MD they were meant to replace. Schebek, Noé and Rogal fix this with a clean idea: stop modeling the whole configuration at once and learn from local atomic environments instead. They pair augmented coupling flows with a graph neural network that builds per-particle embeddings from each atom's neighborhood. By modeling displacements from the ideal lattice rather than absolute positions, the input stays size-independent, so a model trained on a few hundred atoms transfers directly to systems above 1000 atoms. Scaling becomes linear rather than quadratic, and training needs only the energy function, no MD samples. The payoff is large. For mW ice, a global BG needed over 330 GPU days and reached effective sample sizes around 0.2%, while the local model converges in about 4 GPU days and pushes efficiency up by an order of magnitude. It reproduces radial distribution functions and free energies that match MD to within 10^-3 kBT per particle, resolving tiny phase-stability differences in Lennard-Jones crystals, water ice, and the silicon phase diagram. Conditioning on temperature, pressure, and interaction parameters lets one model cover a whole family of materials. The real lever is cost. A trained flow needs only tens of thousands of energy evaluations for a large cell, against tens of millions for standard MD free energy methods, and that gap widens sharply with expensive machine-learned interatomic potentials. In materials development, energy storage, and catalysis, this means screening phase stability and thermodynamics at realistic system sizes without paying the usual simulation tax. Paper: Schebek et al., Nature Communications (2026) — CC BY 4.0 | doi.org/10.1038/s41467…

Mark Rocket Science @SpaceX 基本是热流体力学，燃烧科学 这块多尺度 CFD ，空天凝聚态物理还有很多大空间可为！

snowboat @snowboat84

a week ago

@ufobtc27 @reyjudge From Jim Cantrell The books that he borrowed from me but never returned are: - Rocket Propulsion Elements by George P. Sutton, Oscar Biblarz - Fundamentals of Astrodynamics by Roger R. Bate, Donald D. Mueller, Jerry E. White - International Reference Guide to Space Launch

1 1 3 357 4

@snowboat84 @reyjudge 碉堡了，我感觉可能还有钱学森的星际航行概论

@snowboat84 @reyjudge 我日，张博，分享下，一起学啊哈哈哈哈

Yann LeCun décrit un monde qui l'a enfanté, et ce monde est en train de mourir. Le modèle qu'il défend est celui du XXe siècle. La recherche fondamentale d'un côté (universités, PhD, papiers), l'application industrielle de l'autre, des décennies plus tard. Une chaîne longue, lente, découplée. La découverte en amont, la valeur en aval, et vingt ans entre les deux. Elon a prouvé l'inverse. Quand l'ingénierie et la recherche sont totalement intriquées, quand tu pars d'un problème réel à résoudre et pas d'un papier à publier, tu vas infiniment plus vite. SpaceX, Tesla, Neuralink ne sont pas des applications de la recherche académique, ce sont des labos de recherche qui se trouvent être aussi les boîtes les plus innovantes du monde. Et la vérité que personne n'ose dire, c'est que l'écrasante majorité des papiers ne créent aucune valeur. Des gens qui publient pour publier, optimisés pour la citation et pas pour le réel. L'alignement académique récompense le statut. L'alignement capitalistique récompense une seule chose : que ça marche, vraiment, dans le monde. C'est exactement le point de Thiel. Historiquement, les génies créaient une valeur immense et n'en captaient presque rien, parce qu'ils étaient déconnectés de tout véhicule capable de la capturer. Créer de la valeur et capturer de la valeur sont deux choses distinctes, et l'académie a passé un siècle à exceller dans la première en abandonnant la seconde. Dimon le dit à sa façon : Elon est notre Einstein. Sauf que cet Einstein-là n'a pas eu besoin de l'université pour produire ses percées. Il a eu besoin d'un problème, d'une équipe d'ingénieurs et d'un alignement commercial brutal. Ses breakthroughs dans le spatial, l'automobile et le cerveau ont créé plus de valeur réelle que tout le système académique réuni sur vingt ans. Et avec l'IA, le basculement s'accélère. La valeur du diplôme s'effondre, celle de l'école aussi, parce que l'intuition d'ingénierie branchée sur le réel devient le seul moteur qui compte. LeCun n'a juste pas remarqué que le monde qui l'a fait roi est déjà derrière nous.

Yann LeCun @ylecun

2 weeks ago

@JohnProperBTC @Xnarkycritic @DavidSacks This is targeted R&D with a clear (commercial) purpose. We're talking about basic research here. The kind of federally-funded basic research done 5, 10, or 20 years ago in universities that Elon's projects rely on. The kind of federally-funded research projects that every

10 2 70 27K 11

@brivael The world changed a lot already Tra Sci Paradigm is shifting

@HaoYin20 Congrats, amazing bro 😎

@snowboat84 👍

World Labs CEO Dr. Fei-Fei Li: "The world is not made of words." "Language models have given machines an extraordinary command of concepts, vocabulary, and reasoning, but the physical world, virtual or real, runs on a different substrate." "Where language models learn the statistical structure of text, world models learn the statistical structure of space and time: how light falls on a surface, how a garden looks from an angle no camera has captured, how objects respond to force and follow the laws of physics." "Language gave machines a way to talk about that world. World models are how machines will finally come to understand, imagine, reason and interact with it." Full piece: drfeifei.substack.com/p/a-functional…

Fei-Fei Li @drfeifei

2 weeks ago

x.com/i/article/2062…

168 962 5K 990K 6K

A new study suggests that motion of #BiomolecularCondensates is different from that of colloidal particles. Externally-controlled dynamic states, such as condensate motion, lead to polarization and dipole force fields. 🔗 go.aps.org/4x8sds4

@PhysRevE Nice work

@0xBeyondLee 牛逼