What Is Lost in Post-Training? Default Collapse and the Loss of In-Context Steerability Across Diverse Perspectives
cs.LG
Submitted: 2026-10-02
Updated: 2026-10-02
Sources
- Diagnosing and Repairing Persona Collapse in LLM Advice
- Where does output diversity collapse in post-training?
- Constitutional AI: Harmlessness from AI Feedback
- Deliberative Alignment: Reasoning Enables Safer Language Models
- A Roadmap to Impactful Pluralistic Alignment Research
- When Models Refuse: Political Steerability and Feature Richness as Measures of Ideological Depth
- Fine-Tuning Language Models from Human Preferences
- Proximal Policy Optimization Algorithms
- The Llama 3 Herd of Models
- Gemma 4 Technical Report
- Diverse Preference Optimization
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks