Tag
The paper introduces LongChang, an automated framework that simulates worlds with altered physical laws to test whether LLMs can autonomously discover those laws through free exploration. Among 12 evaluated models, only GPT-6-Astra achieved a 16.7% discovery rate, while the rest scored 0%.
提出 RLAWM，通过分层建模与求解奖励惩罚模型内部世界模型与真实物理公理之间的偏差，显著提升 LLM 在物理推理基准上的表现。
PhysMent introduces an interactive benchmark that evaluates LLM physical reasoning via iterative tool-mediated experimentation with a MuJoCo physics simulator, revealing model weaknesses in multi-step procedural tasks.
This paper introduces a four-stage diagnostic to test whether LLMs can reason in unfamiliar physics frameworks, finding that frontier models achieve low pass rates and exhibit a qualitative-versus-quantitative asymmetry.
This paper audits multimodal physics evaluation pipelines, revealing issues like train-eval contamination, translation drift, and MCQ saturation. It releases new datasets (PhysCorp-A, PhysR1Corp, PhysOlym-A) and a training recipe (Physics-R1) that significantly improves performance on held-out olympiad problems.
The paper introduces SeePhys Pro, a benchmark to diagnose modality transfer issues in multimodal RL for physics reasoning, revealing that models struggle with representation-invariant reasoning and often rely on residual textual cues rather than visual evidence.