Physics Steering: Causal Control of Cross-Domain Concepts in a Physics Foundation Model
cs.LG, cs.AI, physics.comp-ph
Submitted: 2025-11-25
Updated: 2026-11-05
Terminology
Sources
- Refusal in Language Models Is Mediated by a Single Direction
- Late-time acceleration by a residual cosmological constant from sequestering vacuum energy in ultimate collapsed structures
- ChemCrow: Augmenting large-language models with chemistry tools
- Global dissipative solutions of the defocusing isothermal Euler-Langevin-Korteweg equation
- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
- Transformer Feed-Forward Layers Are Key-Value Memories
- Masked Autoencoders Are Scalable Vision Learners
- Axial Attention in Multidimensional Transformers
- Symplectic cuts and open/closed strings I
- Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
- Multiple Physics Pretraining for Physical Surrogate Models
- Progress measures for grokking via mechanistic interpretability
- Centre Stage: Centricity-based Audio-Visual Temporal Action Detection
- ClimaX: A foundation model for weather and climate
- The Well: a Large-Scale Collection of Diverse Physics Simulations for Machine Learning
- The Linear Representation Hypothesis and the Geometry of Large Language Models
- Layer by Layer: Uncovering Hidden Representations in Language Models
- Steering Language Models With Activation Engineering
- Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Bias
- Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis