Han Liu

Computational Physicist at University of Minnesota, BS and PhD from Peking University | Building CFD Solvers with HPC, Quantum Computing, and AI

[Han Liu @hanliufluid]

@MikePFrank Yes

[Han Liu @hanliufluid]

@VukRosic99 说得不错

[Han Liu @hanliufluid]

@Yonah_x 说的很好，学到了很多，谢谢

[Han Liu @hanliufluid]

All benchmarks validate against published results.
Mass conservation within 2%. Correct vortex shedding. Physically accurate two-phase topological transitions.
What an exciting time to be a builder!

[Han Liu @hanliufluid]

I set up 10 AI coding agents and got a 26,000-line physics solver in two weeks.
I didn't read a single line of it.
Here's what happened 🧵

[Han Liu @hanliufluid]

The part that shocked me most: the agents found actual bugs in AMReX itself, and built missing functionality from scratch to keep going. Nobody told them to. They just needed it to move forward, so they did it.

[Han Liu @hanliufluid]

@Yuchenj_UW 100% agree. Lock-in only makes sense when you're miles ahead. Right now the gap is small, why give people a reason to try the other options?
Imo they is should be everywhere. The more places people use your model, the harder it is to leave.

[ア @yuruyurau]

a=(x,y,d=mag(k=(4+sin(y*2-t)*3)*cos(x/29),e=y/8-13))=>point((q=3*sin(k*2)+.3/k+sin(y/25)*k*(9+4*sin(e*9-d*3+t*2)))+30*cos(c=d-t)+200,q*sin(c)+d*39-220)
t=0,draw=$=>{t||createCanvas(w=400,w);background(9).stroke(w,96);for(t+=PI/240,i=1e4;i--;)a(i,i/235)}//#つぶやきProcessing

[Han Liu @hanliufluid]

@HiroNishikawa Very cool design!

[Han Liu @hanliufluid]

@Jmendezcarvajal And the key is always the solver’s efficiency, which does not improve significantly by moving to the cloud.

[Han Liu @hanliufluid]

@FrdGrn @IterIntellectus You have Hinton!

[George Haller @GeorgeHallerETH]

We can now extract simple and predictive reduced-order models for nonlinear dynamical systems directly from videos! https://t.co/ta9AbKZg34

[Han Liu @hanliufluid]

@DylanoA4 Thank you for sharing this.

[Han Liu @hanliufluid]

@LabbediBadis Nice setup!

[Chennakesava Kadapa @chenna1985]

You are not solving the PDEs using Neural Networks in the actual sense of finding the solution of PDEs for the given boundary and loading conditions.

You do something with the data obtained with classical solution methods, but you are not actually solving the PDEs.

#AI #ML #NNs

[Han Liu @hanliufluid]

@sama The dynamics of information flow through your trained nodes when processing this prompt.

[Han Liu @hanliufluid]

Watch “Chopin - Nocturne op.9 No.2 Colorful Liquids in motion” on #Vimeo https://t.co/ekyC18AlcS