Speaker 0:  for a hundred years, scientists have been stuck behind a mathematical wall that turned simple material simulations into thousand hour nightmares. but a new ai framework called thor just smashed that wall, crashing calculation times from months to mere seconds. so thor smashes, the material science bottleneck. researchers at los alamos have launched the thor ai framework, which uses tensor networks to calculate how atoms arrange themselves in solids. it effectively solves the curse of dimensionality that has plagued statistical mechanics for over a century. this turns the impossible into the instant accelerating the discovery of next gen batteries, aerospace alloys, and semiconductors by a factor of 400 x. what do you think of this, john?

Speaker 1:  yeah. so i found this article a couple weeks ago, and so i feel like it's already it's outdated, but it's i don't think it is. i think it's just because how much you and i stay on top of things. but i still think this is relevant, and and we'll talk about we'll talk about the reasons why. so, the reason why classical methods fail because, historically, physics have relied on indirect methods such as molecular dynamic simulations or multi monte carlo methods. and so these simulate particle motion over time to approximate the interval. right? so while they're still useful, they still require an enormous computing power and they take weeks or months, and they still introduce approximation errors. in fact, solving the configurational integral exactly using classical computation could take longer than the age of the universe in some cases. so what thor does is thor stands for tensors for high dimensional object representation and it solves this by combining tensor network mathematics and machine learning potentials. so instead of brute force computation, thor represents the high dimensional problem as a network of small tensors. it uses tensor train cross interpolation to compress the data. it identifies symmetries in crystal structures, and it focuses computation only on the most relevant interactions, which so by doing this, it reduces a massive multidimensional problem into a manageable structure. so, essentially, turning in a possible calculation into a traceable one. and like you said, for weeks to seconds. and it's already been validated on copper metal behavior under pressure, argon, high pressure gas states, and tin, solid solid phase transitions. so, i think this is a super cool, advancement. and it's not just faster computing. it's more of a paradigm shift because it moves from approximation to exact calculation. it enables exploration of previously unreachable physical regimes, and it provides a new computational foundation for material science and thermal dynamics. but what i like about this is it's doing this without quantum computing. right? and so this is something that you would think, based on the amount of computations that are needed, that this would be a problem that quantum computing would solve, not ai. right? you're on mute.

Speaker 0:  when you see an article like this, a 400 x, breakthrough, we always assume it's quantum, but it's interesting that it's just an ai framework they put together. i like to think of this update like trying to find one specific grain of gold on a massive beach. traditional message made you sift through every grain of sand one by one, takes forever. thor is like having a powered metal detector that ignores the sand entirely and points you directly to the golden seconds. it doesn't work harder. it just ignores the noise. yeah. so, peer review, results were published late last year. the framework implementation is currently rolling out in the labs. it's being funded by the us ldrd, primarily available to institutions and academic researchers. it works both in legacy, type binding models and cutting edge h i, hip dash and then machine language potential. so we're gonna see faster development cycles for tech inside your pocket. think of phone batteries that last longer and chips that run cooler arriving years sooner than expected once we start utilizing this kind of technology. the specs, 400 times speed up over traditional los alamos simulation baselines. rank one and rank two tensor training schemes we mentioned optimize the boltzmann weighted densities. it's been validated under extreme stress, gigapascal pressures. underlying science author relies on you already talked about the, tensor train cross interpolation, so i won't get into that. and, this is the biggest leap in compute computational thermodynamics since the introduction of the monte carlo simulation in the nineteen forties. it moves us from stochastic guessing to structured reconstruction, which is crazy.

Speaker 1:  yeah. it's essentially gonna compress years of r and d into days or hours, which is absolutely mind blowing.

Speaker 0:  yeah. and the curse of dimensionality isn't just a catchy name. it refers to the fact of adding just one atom to a simulation increases the mathematical complexity exponentially, not linearly. so the research was birthed at the los alamos national laboratory, the same place that developed the atomic bomb, proving they are still a heavyweights of the big physics and computing. while thor is currently tested on boring crystals like copper and tin, the math behind it, tensor networks, was originally popularized in the world of quantum physics to describe subatomic entanglement. so what are the skeptics saying, john?

Speaker 1:  so, obviously, we've got skeptics out there, but, it's the thing is the thing to think about this is it's not discovering new physics laws. and so we're not trying to break things or anything like that. physics, they already have the equations, but we just couldn't solve them. so what this does is it's making existing ones computationally usable. so, like, the skeptics can talk all they want, but it's not it it isn't it they could say they're taking a shortcut, but i disagree. i think what they're doing is they're just approaching it in a more efficient manner where they'll get well, they're they'll get better outcomes.

Speaker 0:  yeah. this store currently, excels only with ordered crystalline solids. so if you're working with messy amorphous materials like glass or liquid polymers, this framework isn't a magic wand yet. so yeah. we'll have to see. keep an eye on this because this looks like, it has great potential in the future of so many different things. like i said, we could see this changing, things in your pocket incredibly. and if this is the biggest breakthrough since the monte carlo simulation, which you use all the time in project management, not sure if you're aware of that. but, yeah. it's one of it's definitely something that we use a lot. i'd like to see how this is applied to these things going forward. so we are entering the era of instant thermodynamics where the bottleneck is no longer the computer, but our own scientific imagination. if you could design a dream material tomorrow, maybe a room temperature semiconductor or a battery that charges in five seconds, what would you build first? let us know in the chat.