How Yoko Li makes towns, tamagoes, and tools for local AI

Hello all, and welcome to another episode of How I Fly, a series where I interview developers about what they do with technology, what they find exciting, and the unexpected things they’ve learned along the way. This time I’m talking with Yoko Li, an

Fly.io logo
Fly.io
•
Xe Iaso
•10 min read•intermediate•
--
•View Original
JSONLLaMAMidjourneyOllamaReinforcement Learning

Overview

The article discusses Yoko Li's innovative work in AI, focusing on her projects like AI Town and AI Tamago, which utilize emergent behavior and large language models. It highlights the significance of the Local AI Starter Kit in democratizing AI access and the challenges faced in generating structured outputs from AI models.

What You'll Learn

1

How to create a virtual town with AI agents that interact with each other

2

How to implement a Tamagochi-style virtual pet using a large language model

3

How to set up a Local AI Starter Kit for document search and retrieval

Prerequisites & Requirements

  • Understanding of AI concepts and large language models
  • Familiarity with JSON and API usage(optional)

Key Questions Answered

What is AI Town and how does it work?
AI Town is a virtual environment populated by AI agents that communicate with each other, creating emergent behaviors. Users can observe how relationships between these agents evolve over time, showcasing the randomness of AI responses.
How does AI Tamago differ from traditional Tamagochi?
AI Tamago utilizes a large language model, specifically LLaMA 2 7B, to manage the game state and user interactions, allowing for more dynamic and unexpected pet behaviors compared to the fixed state machines of traditional Tamagochi.
What is the Local AI Starter Kit and its significance?
The Local AI Starter Kit simplifies the process of setting up AI applications by enabling users to run models on their own hardware with their own data, eliminating the need for multiple API keys and billing setups. This democratizes access to AI technology.
What challenges are associated with generating structured outputs from AI models?
Generating structured outputs like JSON from AI models can be challenging due to the stochastic nature of these models, which can lead to inconsistent outputs. Workarounds include using models specifically tuned for code generation or leveraging grammar files for more deterministic results.

Technologies & Tools

Some links below are affiliate links. We may earn a commission if you make a purchase.

AI/ML
Llama 2
Used in AI Tamago to manage game state and user interactions.
AI/ML
Ollama
Provides JSON output support for AI models, facilitating structured data generation.
AI/ML
Local AI Starter Kit
A collection of tools for setting up AI applications on local hardware.

Key Actionable Insights

1
Experiment with emergent behavior in AI applications to enhance user engagement.
By allowing AI agents to interact dynamically, developers can create more engaging and unpredictable experiences that mimic real-life interactions.
2
Utilize the Local AI Starter Kit to streamline AI project setups.
This kit allows developers to quickly implement AI features without the overhead of managing multiple APIs, making it easier to prototype and test ideas.
3
Incorporate fuzzy searching techniques for improved document retrieval.
Using vector embeddings for search queries can yield more relevant results, enhancing user experience in applications that require document searches.

Common Pitfalls

1
Relying on large language models for deterministic outputs can lead to inconsistencies.
Due to their stochastic nature, these models can produce varied outputs for similar inputs, making it difficult to achieve reliable structured data formats like JSON.

Related Concepts

Emergent Behavior In AI
Document Retrieval Techniques
Large Language Model Applications