OraclePhys: A Systematic Framework for LLM Fine-Tuning on Structural Mechanics
The study yields two findings: first, the label's answer form -- not its bit count -- causally determines what fine-tuning teaches: a ranking objective installs an out-of-distribution forward model where the untrained base sits at the guessing prior.
Mingyu Li, Guorui Song, Jing Lin et al.
· 0 citations