Funnily enough, I was reading Minsky's book recently. I second the recommendation. I think he's missing many technical details*, but the basic approach seems to be correct.
*(For example, the idea of a "hierarchy of feedback loops" from perceptual control theory would explain a lot of the interactions between agents in his theory.)
I also put the abstract of the paper into GPT-4, and gave it the following prompt:
> Simplify the above. Use paragraph headings and bold key words.
I quite liked its output, as it made it easier to see the core ideas in the paper:
ABSTRACT
Generative Agents: This paper introduces generative agents, computational software agents that simulate believable human behavior. They can be used in various interactive applications like immersive environments, communication rehearsal spaces, and prototyping tools.
Architecture: The generative agent architecture extends a large language model to store a complete record of the agent's experiences in natural language. It enables the agents to synthesize memories, reflect on them, and retrieve them dynamically to plan behavior.
Interactive Sandbox Environment: The generative agents are instantiated in a sandbox environment inspired by The Sims, where users can interact with a small town of twenty-five agents using natural language.
Believable Behavior: The generative agents produce believable individual and emergent social behaviors, such as autonomously spreading party invitations and coordinating events.
Components: The agent architecture consists of three main components: observation, planning, and reflection. Each contributes critically to the believability of agent behavior.
KEYWORDS: Human-AI Interaction, agents, generative AI, large language models
*(For example, the idea of a "hierarchy of feedback loops" from perceptual control theory would explain a lot of the interactions between agents in his theory.)
I also put the abstract of the paper into GPT-4, and gave it the following prompt:
> Simplify the above. Use paragraph headings and bold key words.
I quite liked its output, as it made it easier to see the core ideas in the paper:
ABSTRACT
Generative Agents: This paper introduces generative agents, computational software agents that simulate believable human behavior. They can be used in various interactive applications like immersive environments, communication rehearsal spaces, and prototyping tools.
Architecture: The generative agent architecture extends a large language model to store a complete record of the agent's experiences in natural language. It enables the agents to synthesize memories, reflect on them, and retrieve them dynamically to plan behavior.
Interactive Sandbox Environment: The generative agents are instantiated in a sandbox environment inspired by The Sims, where users can interact with a small town of twenty-five agents using natural language.
Believable Behavior: The generative agents produce believable individual and emergent social behaviors, such as autonomously spreading party invitations and coordinating events.
Components: The agent architecture consists of three main components: observation, planning, and reflection. Each contributes critically to the believability of agent behavior.
KEYWORDS: Human-AI Interaction, agents, generative AI, large language models