Takeshi Kojima

RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs

Geometry of Nash Mirror Dynamics: Adaptive $\beta$-Control for Stable and Bias-Robust Self-Improving LLM Agents

Emergent Misalignment from Superposition

Vertical Attention: Automatic Exploration of Inter-Layer Connections in Transformer-based Language Models

Topology of Reasoning: Understanding Large Reasoning Models through Reasoning Graph Properties

Curse of Instructions: Large Language Models Cannot Follow Multiple Instructions at Once