Thinking in Different Spaces: Domain-Specific Latent Geometry Survives Cross-Architecture Translation
cs.LG, cs.AI
Submitted: 2026-03-20
Updated: 2026-11-02
Terminology
Sources
- Revisiting Model Stitching to Compare Neural Representations
- Probing Classifiers: Promises, Shortcomings, and Advances
- Training Verifiers to Solve Math Word Problems
- Jump to Conclusions: Short-Cutting Transformers With Linear Transformations
- TruthfulQA: Measuring How Models Mimic Human Falsehoods
- Distributed Representations of Words and Phrases and their Compositionality
- Emergent Linear Representations in World Models of Self-Supervised Sequence Models
- Steering Llama 2 via Contrastive Activation Addition
- Improving Instruction-Following in Language Models through Activation Steering
- Extracting Latent Steering Vectors from Pretrained Language Models
- BERT Rediscovers the Classical NLP Pipeline
- Steering Language Models With Activation Engineering
- ExpertSteer: Intervening in LLMs through Expert Knowledge
- Universal and Transferable Adversarial Attacks on Aligned Language Models
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