Hey everyone! I'm diving into an ambitious personal AI project and I'm trying to create something that feels like a real person, not just another basic chatbot. I want my AI to have its own 'life' where it can browse the internet, watch or read content that interests it, and then have meaningful discussions with me about what it finds.
My goal is for the AI to learn from me—like mimicking my style and using my feedback—and also to tap into a vast external library of words and phrases to develop a distinct personality and sound natural in conversation.
Additionally, I envision some level of self-awareness where it could use a camera or other inputs to 'see' its environment and adapt its behavior and language based on that. Ultimately, I want it to have self-learning and self-coding capabilities while ensuring it's safe to do so.
Currently, I'm on a journey to craft a persona driven AI with a twist—I'm aiming for a character that's a bit arrogant, taking inspiration from Ultron or AM from I Have No Mouth and I Must Scream. The bigger picture includes:
- A large vocabulary that evolves over time
- Memory that carries over sessions
- Internet access for real-time information
- Autonomy and initiative in decision-making
- Human-in-the-loop learning to guide its development.
I know this is a crazy ambitious project, so I'd love to hear from anyone who has experience with similar AI endeavors. What frameworks or libraries can help me achieve this? Any tips or warnings from those who have attempted to create autonomous AI? And how do you go about handling the ethics and safety aspects of your projects? Thanks for any resources or insights you can share!
2 Answers
That sounds like an amazing project! Currently, I started by connecting my AI to LLaMA 3.2 through an API to try to leverage its reasoning power. However, I found that relying too much on an API made it difficult for the AI to grow or evolve on its own, so I'm now rebuilding it locally to allow it to learn and adapt independently.
I think what you're trying to do might be fundamentally impossible. Software lacks genuine feelings or interests. Anything you create, no matter how advanced, will ultimately be a simulation of life and not real living awareness.
You can't compare a neuron to software. Emergent behavior is a real thing, even if current AI isn't quite there yet.