General Physics Transformer shows foundation-model paradigm works for physics simulation
burny_tech · x · 2026-10-27
The paper Towards a Physics Foundation Model introduces GPhyT, trained on 1.8 TB of diverse simulation data, showing the train-once-deploy-anywhere paradigm can work for physics. Key insight: transformers can infer governing dynamics from context, simulating fluid-solid interaction, shock waves, thermal convection, and multiphase dynamics without being given the equations. It outperforms specialized architectures by over 7x across domains, generalizes zero-shot to unseen systems via in-context learning, and stays stable in long-horizon rollouts. Shared as context for the ongoing experiment adding Kimi's Attention Residuals to this architecture.
More from Research
- FSPO Beats GRPO and SDPO on Reasoning With Fewer Tokens via Temporal Self-Distillation — IITK · 2026-10-27
- DiscoDemo Uses RL to Generate Diverse Robot Demos, Boosting Real-World Success from 39% to 78% — DAVIAN-Robotics · 2026-10-27
- AI Evaluator Forum Releases AEF-1: Minimum Operating Conditions for Independent AI Evaluations — RishiBommasani · 2026-10-27
- Model breeding begins: neuron combo makes Qwen believe it's Cookie Monster — yacineMTB · 2026-10-27
- EMNLP paper uses sparse autoencoders for text-based causal confounding adjustment — anjalie_f · 2026-10-27
- Volta releases Blackwell kernel verifier letting LLMs generate PTX directly, skipping compiler layers — Azaliamirh · 2026-10-27