Teach the Way It Thinks: Guiding LLM Post-training Data Engineering with Sparse Autoencoders
cs.LG, cs.AI, cs.CL
Submitted: 2026-05-26
Updated: 2026-11-05
Terminology
Sources
- Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
- Neuron-Aware Data Selection In Instruction Tuning For Large Language Models
- From Data-Centric to Sample-Centric: Enhancing LLM Reasoning via Progressive Optimization
- Training Verifiers to Solve Math Word Problems
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- Scaling and evaluating sparse autoencoders
- Prompt Curriculum Learning for Efficient LLM Post-Training
- DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoning
- Large-Scale Data Selection for Instruction Tuning
- Tulu 3: Pushing Frontiers in Open Language Model Post-Training
- Solving Quantitative Reasoning Problems with Language Models
- LearnAlign: Data Selection for LLM Reinforcement Learning with Improved Gradient Alignment
- LIMR: Less is More for RL Scaling
- Rho-1: Not All Tokens Are What You Need
- SparseRM: A Lightweight Preference Modeling with Sparse Autoencoder
- Task-Specific Data Selection for Instruction Tuning via Monosemantic Neuronal Activations
- Locating and Editing Factual Associations in GPT
- Interpretability Can Be Actionable
- Training language models to follow instructions with human feedback
- Learning to Reason with Curriculum I: Provable Benefits of Autocurriculum
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