Bookmarks

Manipulating Chess-GPT's World Model

The author explores how Chess-GPT, a language model for chess, can improve its performance by manipulating its internal understanding of player skill and board state. By using linear probes and skill interventions, the model's chess-playing ability was significantly enhanced, especially in games with random initializations. The findings suggest that Chess-GPT learns a deeper understanding of chess rather than just memorizing patterns.

Generative Agents: Interactive Simulacra of Human Behavior

The content discusses generative agents that simulate believable human behavior for interactive applications. These agents populate a sandbox environment, interact with each other, plan their days, form relationships, and exhibit emergent social behaviors. The paper introduces a novel architecture that allows agents to remember, retrieve, reflect, and interact dynamically.

Subcategories