Constructing a Potato- modeled GLaDOS system for AI initiation
In an exciting blend of technology and gaming nostalgia, Dave Niewinski has created an AI-powered version of GLaDOS's potato-based incarnation from the popular game *Portal 2*. This real-life AI assistant, housed within a custom 3D-printed potato enclosure, serves as a fun and accessible introduction to AI, demonstrating how modern language models can be used in creative, real-world projects with limited resources.
**The Making of PotatOS**
The heart of the project lies in a single-board computer (SBC), concealed within the whimsical, game-accurate potato enclosure. The exact model of the SBC remains unnamed, but it is small, affordable, and powerful enough to run a workable software system, albeit with some limitations in speed and memory.
To power GLaDOS's witty and sarcastic dialogue, the project employs a large language model (LLM), though the specific type, such as GPT-4 or Llama, is not specified in the available sources. The project demonstrates the use of LLMs in a hobbyist environment, focusing on balanced performance and budget rather than high-end processing.
Speech generation is a key feature of the potato-based GLaDOS, allowing it to respond to user input with synthesized voice lines mimicking the original character. While details about the specific voice recognition and synthesis libraries used are not provided, Vosk was employed for speech recognition, which Dave described as "somewhat documented".
**A Practical Demonstration of Building AI Systems**
The project aims to provide a practical demonstration of building AI systems, using a recognizable character from a popular video game as an example. Piper was used for speech generation, trained on audio files from the Portal wiki, while LlamaIndex was employed for preprocessing information from the wiki for retrieval-augmented generation.
**Safety Measures and Offline Functionality**
To ensure safety, Dave made sure not to give the device control over his home. The system was designed to work offline, choosing an Nvidia Jetson Orin Nano for its computational power. The purpose of this latest project is to serve as a fun introduction to building AI systems, not to replace traditional home assistants.
**The Final Code and Availability**
All necessary components, including the Nvidia Jetson Orin Nano, a microphone, a speaker, and supplemental electronics, are housed in the potato-shaped enclosure. The final code for the project is available on GitHub under the name PotatOS, allowing enthusiasts to replicate the project and explore AI systems further.
This project highlights the combination of affordable hardware, custom fabrication, and open-ended AI tools to bring a beloved game character into the real world as a conversational, interactive potato. The AI-powered potato GLaDOS is not just a fun novelty; it's a testament to the potential of AI in creative, accessible, and affordable projects.
In the pursuit of demonstrating AI capabilities in a hobbyist environment, the project utilizes electronics, such as a single-board computer and a microphone, to power the potato-based GLaDOS. Furthermore, the use of a large language model for generating witty and sarcastic dialogue aligns with the field of artificial-intelligence, showcasing its application in programming creative, real-world projects.
