Learning from fracture: Teaching language models to design through physical experience
Abstract
How can language models learn to optimize physical designs through experience? We introduce FracLearn, an experience-abstraction-generalization framework evaluated through a deliberately unfamiliar, Tetris-inspired fracture game. The model first proposes architectures and examines simulated fracture responses to accumulate experience. Abstraction organizes these observations into conditional principles linking reinforcement placement, crack evolution and load transfer, with supporting evidence and failure conditions. Generalization challenges the model to combine these principles in new architectures and surpass the earlier design record. Campaigns demonstrate progressive refinement and recovery through hypothesis revision, yielding a best composite score 11.0% above the experience-stage record. Fracture analyses connect improved performance to crack deflection and temporary trapping as toughening mechanisms, alongside greater residual load capacity. Learning occurs through external knowledge and decisions, without model-weight updates. This approach offers a route to scientific agents that turn computational experiments into reusable physical knowledge for materials discovery and engineering design.
Related
- Corridor Bonding Theory: A QTAIM-Based Density-Informed Graph Language for Teaching and Interpreting Chemical Structure
- Value-level data provenance in an electronic laboratory notebook used by humans and language-model agents: An experience report
- Teaching quantum mechanics in secondary schools: A qubit-first learning path to Bell's inequalities
- Fracture-Bench: A Controlled Benchmark Suite for Machine Learning Surrogates in Phase-Field Fracture Mechanics
- Thermodynamically Consistent Machine Learning for Fracture Prediction in Heterogeneous Materials
BibTeX
@misc{ni2026learning,
title = {{Learning from fracture: Teaching language models to design through physical experience}},
author = {Bo Ni},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.17804},
eprint = {2610.17804},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.17804},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Learning from fracture: Teaching language models to design through physical experience
AU - Bo Ni
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2610.17804
UR - https://arxiv.org/abs/2610.17804
AB - How can language models learn to optimize physical designs through experience? We introduce FracLearn, an experience-abstraction-generalization framework evaluated through a deliberately unfamiliar, Tetris-inspired fracture game. The model first proposes architectures and examines simulated fracture responses to accumulate experience. Abstraction organizes these observations into conditional principles linking reinforcement placement, crack evolution and load transfer, with supporting evidence and failure conditions. Generalization challenges the model to combine these principles in new architectures and surpass the earlier design record. Campaigns demonstrate progressive refinement and recovery through hypothesis revision, yielding a best composite score 11.0% above the experience-stage record. Fracture analyses connect improved performance to crack deflection and temporary trapping as toughening mechanisms, alongside greater residual load capacity. Learning occurs through external knowledge and decisions, without model-weight updates. This approach offers a route to scientific agents that turn computational experiments into reusable physical knowledge for materials discovery and engineering design.
ER -