Our team's project for the hackathon is kind of like building a super-smart robot that thinks a lot like a human. Imagine your brain as having two sides: one side is really good at math and solving problems, while the other side loves to paint and write stories. We want our robot to think like that too!
To make this happen, we're using three parts: two big ones to think like the two sides of your brain and a smaller one to make sure the big parts work together nicely. Imagine having two big toy boxes: one filled with math games and another with art supplies. Then, there's a little toy that helps you decide when to play with each box.
We'll put all of this together on a website, so people can see how the robot thinks and plays with its toys.
Now, diving into the details:
Two Large Language Models (LLMs): These are complex artificial intelligence algorithms. One is designed to focus on analytical and logical reasoning (like the logical side of the brain), and the other is geared towards symbolic and creative understanding (like the artistic side).
Mediating Model: This smaller model acts as a bridge between the two LLMs. It ensures that both LLMs function in harmony, deciding when to use logical reasoning and when to engage in creative thought, integrating the two to form a cohesive chain of thought.
Web User Interface (WebUI): This is the platform where users will interact with the system. It allows loading and managing the two LLMs and the mediator, presenting it in a way that is accessible and user-friendly.
Integration and Architecture: As a full-stack project, our approach will involve the design of both front-end (what users see) and back-end (how the system processes data) components. We'll use standard web technologies and machine learning frameworks to ensure a robust and scalable system.
The goal is to emulate a dichotomous thinking pattern, reflecting human-like cognition in a machine, thus crafting a complex yet elegant solution that showcases both the scientific rigor and creative artistry of human thought.