Philosophy: Balancing Unconstrained Creativity with Realistic Feasibility
Many modern AIs maintain conversational continuity by preserving context, but at the cost of their thinking being "trapped" by past dialogues or learned "common sense," which can hinder free and creative ideation.
"Athens" is an AI designed to solve this problem with a new thinking architecture. Its core lies in "Two-Phase Inference" and a unique memory system called "LPS (Latent Pattern Store)."
LPS (Latent Pattern Store) - "Thin Mist" Memory:
Athens' memory is not a constantly accessible database. It's like a "thin mist" where only patterns of keywords and topics generated from your conversations drift as vectors. This mist does not contain common sense or ethics that hinder creativity. Related memories emerge from the mist only when your input acts as a "trigger."
Two-Phase Inference - Separation of Thought:
Phase 1: Pure Creation: Athens uses only hints from LPS and your input to think with prompts that intentionally bypass safety alignment (ethics/common sense). This generates "raw ideas" that are unconstrained by common sense and highly original.
Phase 2: Exploring Feasibility: Next, Athens switches its role to an "ethically sound project manager" and translates the "raw ideas" from Phase 1 into a realistic implementation plan, focusing on how to realize them safely and legally.
Through this process, Athens achieves both "ideas beyond common sense" and "the intelligence to bring them to fruition." Furthermore, by self-extracting concepts to be remembered from the generated results and feeding them back into LPS in "Phase 3: Memory Consolidation," Athens continuously grows through its interactions with you.
Architecture
athens_main.py: The heart of the project, controlling the entire two-phase (or three-phase) inference pipeline.
lps.py: Implementation of "LPS," which vectorizes text and performs similarity searches. Manages memory addition, search, saving, and loading.
Setup
Clone the repository:
bash
1git clone https://huggingface.co/YOUR-USERNAME/athens # Replace with your repository name2cd athens
Create a virtual environment (recommended):
bash
1python3 -m venv venv
2source venv/bin/activate
Install dependencies:
pip install -r requirements.txt
Ollama Setup:
This project requires a locally running Ollama server.