Steer, Don't Tell: Breaking Reasoning Plateaus with Reachability-Aligned Self-Distillation
cs.CL
Submitted: 2026-10-31
Updated: 2026-10-31
Terminology
Sources
- Nudging the Boundaries of LLM Reasoning
- The Unlearnability Phenomenon in RLVR for Language Models
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- HDPO: Hybrid Distillation Policy Optimization via Privileged Self-Distillation
- Trajectory-Refined Distillation
- The Role of Feedback Alignment in Self-Distillation
- START: Self-taught Reasoner with Tools
- QuestA: Expanding Reasoning Capacity in LLMs via Question Augmentation
- GEAR: Granularity-Adaptive Advantage Reweighting for LLM Agents via Self-Distillation
- Learning What Reinforcement Learning Can't: Interleaved Online Fine-Tuning for Hardest Questions
- Making Expert Reasoning Learnable with Self-Distillation
- OpenAI o1 System Card
- POPE: Learning to Reason on Hard Problems via Privileged On-Policy Exploration
- Can Large Reasoning Models Self-Train?
- DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning
- CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation
- Learning from Partial Chain-of-Thought via Truncated-Reasoning Self-Distillation
- GATES: Self-Distillation under Privileged Context with Consensus Gating
- STEPS: Selective On-Policy Self-Distillation for Reasoning
Related papers
- ALARM: Audio-Language Alignment for Reasoning Models
- ThinkSum: Probabilistic reasoning over sets using large language models
- Coherence boosting: When your pretrained language model is not paying enough attention
- From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
- Improving General Role-Playing Agents via Psychology-Grounded Reasoning and Role-Aware Policy Optimization
- The Art of Asking: Multilingual Prompt Optimization for Synthetic Data