Playing Dumb: How a Chess Transformer Simulates Users Across Skill Levels
cs.AI, cs.LG
Submitted: 2026-11-02
Updated: 2026-11-02
Code: https://github.com/CSSLab/maia2-skill-adaptation
Project page: https://hxim.github.io
Terminology
Sources
- Understanding intermediate layers using linear classifier probes
- A Theory for Emergence of Complex Skills in Language Models
- Eliciting Latent Predictions from Transformers with the Tuned Lens
- Language Models are Few-Shot Learners
- Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision
- Discovering Latent Knowledge in Language Models Without Supervision
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- Transcoders Find Interpretable LLM Feature Circuits
- Emulate or Estimate? The Divergent Strengths of Base and Post-Trained Language Models for Opinion Simulation
- Language Models Represent Space and Time
- Finding Neurons in a Haystack: Case Studies with Sparse Probing
- Evidence of Learned Look-Ahead in a Chess-Playing Neural Network
- Agent Safety From Within: Detecting Harmful Trajectories from LLM Internal States
- Understanding the Dynamics of Demonstration Conflict in In-Context Learning
- Measuring Progress in Dictionary Learning for Language Model Interpretability with Board Game Models
- Physics of Skill Learning
- Frontier Models are Capable of In-context Scheming
- Locating and Editing Factual Associations in GPT
- Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
- Learning to Imitate with Less: Efficient Individual Behavior Modeling in Chess
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents