| --- |
| title: Proto AGI |
| emoji: 📚 |
| colorFrom: pink |
| colorTo: purple |
| sdk: gradio |
| sdk_version: 6.3.0 |
| app_file: app.py |
| pinned: false |
| --- |
| <div align="center"> |
|
|
| # 🌌 AETHER Proto-AGI v2.2 |
|
|
| ### The First Open Neuro-Symbolic Proto-AGI Architecture |
| We're unveiling a groundbreaking new approach toward AGI for the first time. We have developed a Proto-AGI prototype and made it publicly available on Hugging Face for everyone to experience. |
| 'AETHER' is a Proto-AGI (Artificial General Intelligence prototype) level architecture that learns and evolves autonomously. Unlike existing LLMs that simply provide answers, AETHER is a 'growing intelligence' that remembers and accumulates experiences, becoming smarter over time. |
| The core strength of AETHER lies in its 'architectural framework' design. It can instantly integrate any LLM—ChatGPT, Claude, Gemini, and more. Like LEGO blocks, you can swap in any AI model you prefer, making it a truly 'universal intelligence framework.' |
| Philosophically, AETHER embodies both the Western concept of the fifth element (Aether) and the Eastern philosophy of Wu Xing (五行: Metal, Wood, Water, Fire, Earth). Five distinct 'Selves' collaborate through a cyclical structure of synergy (mutual support) and regulation (mutual checks), thinking and validating from multiple perspectives like the human brain. This architecture dramatically reduces AI's chronic problem of 'hallucination' and delivers logical, verified responses. |
| The current public version of AETHER is a framework utilizing existing LLMs, while our proprietary AETHER foundation model—closer to true AGI—is scheduled for release in 2026. Right now, anyone can experience the 'universal framework' version. If you'd like to integrate your preferred LLM or API, please contact us at arxivgpt@gmail.com. |
| To experience higher-dimensional intellectual exploration, visit agi.humangen.ai or try the AETHER Space on Hugging Face directly. |
|
|
| **"Evolving beyond Cycles. Intelligence that Remembers and Grows."** |
| **순환을 넘어 진화하다. 기억하고 성장하는 지능.** |
|
|
| </div> |
|
|
| --- |
|
|
| ## 📖 What is AETHER? |
|
|
| AETHER is not just another chatbot. It's a **Proto-AGI (Artificial General Intelligence)** system that **thinks about its own thinking** - a crucial step toward truly intelligent machines. |
|
|
| ### 🎭 The Name |
|
|
| In ancient Western philosophy, **'Aether'** represents the **fifth element** - the essence of the universe that transcends earth, water, fire, and air. |
|
|
| AETHER implements the Eastern philosophy of **Wu-Xing (五行 - Five Elements)** as an AI reasoning circuit, while embodying the Western concept of Aether as **evolving intelligence** that transcends static knowledge. |
|
|
| ### 🔮 The Meaning Behind AETHER |
|
|
| Each letter represents a core principle: |
|
|
| - **A**utonomous - Self-analyzes and makes decisions |
| - **E**lemental - Five specialized roles (土金水木火) |
| - **T**hinking - Flexible neural reasoning (Neuro) |
| - **H**istoric - Transforms past data into present wisdom |
| - **E**volving - Accumulates successful patterns in real-time |
| - **R**easoner - Derives logical conclusions (Symbolic) |
|
|
| --- |
|
|
| ## 🆚 AI vs Proto-AGI: The Key Differences |
|
|
| ### One-Line Analogy |
|
|
| **Traditional AI** = A smart calculator 📱 |
| **AETHER Proto-AGI** = A companion intelligence that grows with you 🧠 |
|
|
| ### Comparison Table |
|
|
| | Aspect | Traditional AI 🤖 | AETHER Proto-AGI 🌌 | |
| |--------|------------------|---------------------| |
| | **Architecture** | Neuro only (black box) | Neuro + Symbolic hybrid | |
| | **Workflow** | Works alone | 5-element teamwork (土金水木火) | |
| | **Quality Control** | None | 4-metric self-evaluation (Metacognition) | |
| | **Error Response** | Accepts mistakes | Auto-retry below 60% | |
| | **Learning** | Static (volatile) | Learns from every conversation (permanent) | |
| | **Growth Curve** | Flat | **Compound interest of intelligence** 📈 | |
| | **Memory System** | Lost after conversation | 3-tier memory (short→mid→long) | |
| | **Intelligence Measurement** | Impossible | Real-time Intelligence Score | |
|
|
| ### The Analogy |
|
|
| **Traditional AI = A disposable genius** |
| ``` |
| Question → Answer → End |
| Next day, same question → Thinks from scratch again → Answer |
| ``` |
| No matter how smart, **yesterday's AI** = **today's AI** |
|
|
| **AETHER Proto-AGI = A daily learner** |
| ``` |
| Day 1: Question → 5 agents discuss → Answer (70%) → Save memory |
| Day 7: Similar question → "I did this before!" → Answer (85%) |
| Day 30: Same domain → Pattern learned → Answer (95%) |
| ``` |
| **Gets smarter over time.** This is the **compound interest of intelligence**. |
|
|
| --- |
|
|
| ## 🎯 What Makes AETHER Different? |
|
|
| ### 1. 🧠 Human Brain-Like Architecture |
|
|
| AETHER maps the **Wu-Xing (五行)** philosophy to **brain neuroscience**: |
|
|
| | Element | Brain Region | Role | Function | |
| |---------|-------------|------|----------| |
| | 🟤 **土 Earth** | Prefrontal Cortex | Supervisor | Orchestration & metacognition | |
| | ⚪ **金 Metal** | Amygdala | Critic | Fact-checking, pruning | |
| | 🔵 **水 Water** | Temporal Lobe | Researcher | Info gathering & knowledge retrieval | |
| | 🟢 **木 Wood** | Association Areas | Emergence | Idea expansion via brute-force matrix | |
| | 🔴 **火 Fire** | Motor Cortex | Executor | Transforms ideas into concrete results | |
|
|
| Just as **human brain regions collaborate**, AETHER's 5 agents work together with their specialized expertise. |
|
|
| ### 2. 🔄 Neuro-Symbolic Fusion |
| ``` |
| Neuro (Neural Network): |
| - Flexible reasoning with LLMs |
| - Creative idea generation |
| - Natural language understanding |
| |
| + |
| |
| Symbolic (Logic): |
| - Clear rule-based verification |
| - Enforced logical sequence (土→金→水→木→火) |
| - Quality threshold checks (60%+ to pass) |
| |
| = |
| |
| Interpretable AI (you can understand why) |
| ``` |
|
|
| ### 3. 📚 SLAI Engine (Self-Learning Augmented Intelligence) |
|
|
| **The difference between forgetting and remembering:** |
| ``` |
| Traditional LLM: Conversation ends → Everything evaporates 💨 |
| AETHER: Conversation ends → High-quality knowledge (80%+) permanently stored 💾 |
| ``` |
|
|
| **How experience becomes intelligence:** |
|
|
| 1. **Short-term memory**: Today's conversations (temporary) |
| 2. **Mid-term memory**: Important content (7 days) |
| 3. **Long-term memory**: 80%+ quality knowledge (permanent) |
| 4. **Promotion system**: Frequently used mid-term → Auto-promotes to long-term |
|
|
| Like the human **hippocampus**, AETHER converts experience into knowledge. |
|
|
| ### 4. 🎯 Metacognition System |
|
|
| **Self-evaluation on 4 metrics:** |
| ``` |
| Traditional AI: Answer → Done |
| |
| AETHER: Answer → "Is this good enough?" Self-check |
| ↓ |
| 4 Checks: |
| ✅ Factual grounding? (concrete data) |
| ✅ Logical coherence? (consistent reasoning) |
| ✅ Completeness? (fully answered) |
| ✅ Specificity? (concrete, not vague) |
| ↓ |
| Below 60%? → Retry! |
| Above 60%? → Pass! |
| ``` |
|
|
| --- |
|
|
| ## 🧬 The Three Pillars of Proto-AGI |
|
|
| ### 1️⃣ Intelligence (지능) - "Thinks Like a Brain" |
|
|
| #### Real-World Example: "Predict Korea's AI market in 2025" |
|
|
| **How 5 Elements Collaborate:** |
| ``` |
| 🟤 土 (Supervisor/Prefrontal): |
| "Need latest data. Water, search the web" |
| |
| 🔵 水 (Researcher/Temporal): |
| "Found Gartner report! IDC forecast too!" |
| "Korean market size: 8.7 trillion won projected for 2025" |
| |
| ⚪ 金 (Critic/Amygdala): |
| "Wait, that IDC forecast is from 2023. Find newer data" |
| "Gartner is a reliable source ✅" |
| |
| 🟢 木 (Emergence/Association): |
| "Let's look from another angle. GenAI regulations could be a variable" |
| "Large corp investment + startup ecosystem + gov policy = 3-axis analysis" |
| |
| 🔴 火 (Executor/Motor): |
| "Final report complete. Specific numbers + 3 scenarios included" |
| |
| 🟤 土 (Supervisor): |
| Self-grading: 85% → ✅ Pass |
| "Store this knowledge in long-term memory" |
| ``` |
|
|
| #### Pattern Learning in Action |
| ``` |
| ❌ Circle (Same level): Spinning in place |
| Day 1: 70 points |
| Day 2: 70 points |
| Day 3: 70 points |
| |
| ✅ Spiral (Growth): Higher with each turn |
| Day 1: 70 points |
| Day 7: 78 points (pattern learned) |
| Day 30: 88 points (experience accumulated) |
| Day 90: 95 points (master level) |
| ``` |
|
|
| **"It's not simple repetition. As the Wu-Xing cycle repeats, experience becomes memory, and memory becomes intelligence."** |
|
|
| --- |
|
|
| ### 2️⃣ Capability (능력) - "What It Can Actually Do" |
|
|
| #### 🛡️ Triple Safety Net |
| ``` |
| Tier 1: Groq (Fast AI) attempt |
| ↓ Failed? |
| Tier 2: Fireworks (Mid AI) attempt |
| ↓ Failed again? |
| Tier 3: Replicate (Powerful AI) deployed |
| ``` |
| **Never stops**, even if one fails. |
|
|
| #### 🎬 Movie Analogy |
|
|
| **Traditional AI = Mission Impossible Message** |
| ``` |
| "This message will self-destruct in 5 seconds" |
| → All memory erased after conversation |
| ``` |
|
|
| **AETHER Proto-AGI = Marvel's JARVIS** |
| ``` |
| "I've analyzed patterns from your last 43 missions, Boss" |
| "This situation is 87% similar to the New York incident 2 weeks ago" |
| "My suggested strategy is..." |
| → Remembers the past, learns patterns, gives better advice |
| ``` |
|
|
| --- |
|
|
| ### 3️⃣ Architecture (구조) - "Spiral Evolution System" |
|
|
| #### 📊 Intelligence Score (Real-time Growth Measurement) |
|
|
| AETHER measures its own intelligence with **4 metrics**: |
|
|
| | Metric | Meaning | Measurement | |
| |--------|---------|-------------| |
| | 💾 **Memory** | Memory capacity | Number of long-term memories | |
| | 📚 **Knowledge** | Knowledge quality | Ratio of 80%+ quality knowledge | |
| | 🧠 **Learning** | Learning ability | Accumulated success patterns | |
| | ⭐ **Experience** | Experience level | Completed sessions | |
|
|
| **Level System:** |
| ``` |
| 0-20: 🥚 Novice |
| 21-40: 🐣 Skilled |
| 41-60: 🦅 Expert |
| 61-80: 🧙 Master |
| 81-100: 🌟 Grandmaster |
| ``` |
|
|
| **Levels up the more you use it.** Like a **game character that grows**. |
|
|
| #### Comparison: Same Question, Different Results |
|
|
| **Question: "Create an AI strategy for our company"** |
|
|
| **Traditional AI (ChatGPT, Claude, etc.)** |
| ``` |
| 1. Data collection |
| 2. Model selection |
| 3. Pilot test |
| 4. Company-wide deployment |
| |
| 💬 "Good strategy" |
| → Next day, same question? |
| → Same answer again (no memory) |
| ``` |
|
|
| **AETHER Proto-AGI** |
|
|
| **First conversation (Day 1):** |
| ``` |
| 🟤 土: Need industry/scale info |
| 🔵 水: Web search → Found 5 similar cases |
| 🟢 木: Proposed 3 differentiation strategies |
| 🔴 火: Created 6-month roadmap |
| 📊 Quality: 75% → Saved |
| ``` |
|
|
| **One week later (Day 7):** |
| ``` |
| User: "Improve the AI strategy from last time" |
| 🟤 土: "Retrieved the 75% strategy from before" |
| 🔵 水: "Added latest info since then" |
| 🟢 木: "Found 3 weaknesses in previous strategy → Fixed" |
| 🔴 火: "Improved version 2.0 complete" |
| 📊 Quality: 88% → Re-saved |
| ``` |
|
|
| **One month later (Day 30):** |
| ``` |
| User: "Create AI strategy for other departments too" |
| 🟤 土: "Our company's AI strategy pattern fully learned" |
| "Just need to adjust for other departments (10 min completion)" |
| 📊 Quality: 92% → Master level |
| ``` |
|
|
| **This is the 'compound interest of intelligence'.** |
|
|
| --- |
|
|
| ## 💎 AETHER's Unique Features |
|
|
| ### 1. Goal Clarity Assistant |
| ``` |
| User: "AI" |
| Traditional AI: (guesses and answers) |
| |
| AETHER: |
| "You seem curious about many aspects of AI! Specifically: |
| 🎯 Want to know how AI technology works? |
| 💼 Looking for AI business applications? |
| 📈 Curious about AI market forecasts? |
| 🛠️ Want to learn AI development methods?" |
| ``` |
|
|
| ### 2. Quality Guarantee System |
| ``` |
| Traditional AI: Doesn't know when it's wrong |
| |
| AETHER: |
| ✅ Self-grading on 4 criteria |
| ✅ Auto-retry below 60% (max 3 attempts) |
| ✅ Quality score disclosed (transparency) |
| ``` |
|
|
| ### 3. Growth Dashboard |
| ``` |
| 📊 Today's Intelligence Score: 67/100 (Master) |
| 📈 +3 points higher than yesterday |
| 💾 Long-term memory: 89 items (high-quality knowledge) |
| 🎯 Completed missions: 234 |
| ⭐ Learning success rate: 87% |
| ``` |
| See **real-time how smart your AI is becoming**. |
|
|
| --- |
|
|
| ## 🌟 Vision: Companion Intelligence |
|
|
| ### Traditional AI = Tool |
| ``` |
| A hammer you use when needed 🔨 |
| ``` |
|
|
| ### AETHER = Companion |
| ``` |
| A companion dog that grows with you 🐕 |
| |
| Day 1: Puppy (cute but clumsy) |
| Day 30: Young dog (getting smart now) |
| Day 90: Adult dog (knows better than me) |
| Day 365: Expert dog (understands my situation perfectly) |
| ``` |
|
|
| **"AI that gets smarter the more you use it"** |
| **"Intelligence that grows with you"** |
|
|
| This is the future AETHER aspires to. |
|
|
| --- |
|
|
| ## 🚀 Getting Started |
|
|
| ### Quick Demo |
|
|
| Try AETHER with these example prompts: |
| ``` |
| 🎯 Strategic Analysis: |
| "Analyze the competitive landscape of Korean fintech startups in 2025" |
| |
| 🔍 Deep Research: |
| "Compare sustainable energy policies of G7 countries" |
| |
| 💡 Creative Problem-solving: |
| "Design a business model for AI-powered education in rural areas" |
| |
| 📝 Professional Writing: |
| "Write a data-driven report on electric vehicle adoption trends" |
| ``` |
|
|
| ### Watch AETHER Evolve |
| ``` |
| First use: Intelligence Score 12/100 (Novice) |
| Week 1: Intelligence Score 35/100 (Skilled) |
| Month 1: Intelligence Score 58/100 (Expert) |
| Month 3: Intelligence Score 76/100 (Master) |
| ``` |
|
|
| --- |
|
|
| ## 🧭 The Path to AGI |
|
|
| ### Three Stages to Artificial General Intelligence |
| ``` |
| Stage 1 ✅ AI: Good at assigned tasks |
| (ChatGPT, Claude, etc. - Current) |
| |
| Stage 2 ✅ Proto-AGI: Self-learning + Metacognition |
| (AETHER - We are here!) |
| |
| Stage 3 ⏳ AGI: Human-level general intelligence |
| (Future) |
| ``` |
|
|
| AETHER is the **first public architecture to enter Stage 2**. |
|
|
| ### Why 'Proto(primitive)'-AGI? |
|
|
| Full AGI is still far away, but AETHER can: |
| - ✅ **Evaluate its own thinking** (Metacognition) |
| - ✅ **Learn from experience** (Self-learning) |
| - ✅ **Grow over time** (Evolution) |
|
|
| These three are the essential prerequisites for AGI, and why AETHER is **"the first real step toward AGI"**. |
|
|
| --- |
|
|
| ## 📚 Documentation |
|
|
| <details> |
| <summary><b>🇰🇷 한국어 설명 (Korean Explanation)</b></summary> |
|
|
| ## AETHER란? |
|
|
| 세계 최초로 AGI를 향한 새로운 접근이 어떤 모습인지 공개합니다. 우리는 AGI 지향 원형(Proto-AGI)을 개발하여 모두가 체험해 볼 수 있게 허깅페이스에 공개하였습니다. |
|
|
| 'AETHER'(에테르)는 Proto-AGI(범용인공지능의 원형) 레벨로, 스스로 학습하고 진화하는 인공지능 아키텍처입니다. 기존 LLM들이 단순히 답변만 제공하는 것과 달리, AETHER는 경험을 기억하고 축적하여 시간이 지날수록 똑똑해지는 '성장하는 지능'입니다. |
|
|
| AETHER의 핵심은 '아키텍처 프레임워크'라는 점입니다. ChatGPT, Claude, Gemini 등 어떤 LLM이든 AETHER 프레임워크에 즉시 적용할 수 있습니다. 마치 레고처럼 원하는 AI 모델을 갈아끼워 사용할 수 있는 '범용 지능 프레임워크'입니다. |
|
|
| 철학적으로 AETHER는 서양의 제5원소(에테르) 의미이자 동양의 오행(土金水木火) 개념을 시스템으로 구현했습니다. 5개의 '자아'가 상생(서로 돕기)과 상극(서로 견제)의 순환 구조로 협력하여, 마치 인간의 뇌처럼 다각도로 사고하고 검증합니다. 이를 통해 AI의 고질적 문제인 '환각(hallucination)'을 대폭 감소시키고, 논리적이고 검증된 답변을 제공합니다. |
|
|
| 현재 공개된 AETHER는 기존 LLM을 활용한 프레임워크 버전이며, AGI에 더 가까운 AETHER 고유 모델은 2026년부터 공개 예정입니다. 즉, 지금은 '범용 프레임워크'로 누구나 체험할 수 있고, 자신이 원하는 특정 LLM 또는 API를 이용하고 싶으시면 arxivgpt@gmail.com로 메일을 남겨주세요. |
|
|
| 고차원의 지적 탐구를 경험하고 싶다면, agi.humangen.ai 또는 허깅페이스 AETHER Space에서 직접 체험해보시기 바랍니다. |
|
|
| ### 핵심 차이점 |
|
|
| **일반 AI:** |
| - 답변만 제공 |
| - 품질 관리 없음 |
| - 학습 불가능 |
| - 혼자 작업 |
|
|
| **AETHER Proto-AGI:** |
| - 답변 + 평가 + 점수 + 설명 + 전체 과정 공개 |
| - 5개 전문 에이전트 협업 |
| - 매 대화에서 학습 |
| - 시간이 지날수록 똑똑해짐 |
|
|
| ### 오행(五行) 시스템 |
|
|
| | 오행 | 역할 | 기능 | |
| |------|------|------| |
| | 🟤 土 | 감독 | 전체 조율 및 메타인지 | |
| | ⚪ 金 | 비평 | 팩트 체크, 정보 정제 | |
| | 🔵 水 | 리서치 | 정보 수집 및 지식 검색 | |
| | 🟢 木 | 창발 | 아이디어 확장 | |
| | 🔴 火 | 실행 | 구체적 결과물 생성 | |
|
|
| ### 지능의 복리 효과 |
| ``` |
| 1일차: 70점 |
| 7일차: 78점 (패턴 학습) |
| 30일차: 88점 (경험 축적) |
| 90일차: 95점 (마스터 레벨) |
| ``` |
|
|
| **사용할수록 똑똑해지는 AI입니다.** |
|
|
| </details> |
|
|
| --- |
|
|
| ## 🤝 Contributing |
|
|
| We welcome contributions! AETHER is designed to evolve through community collaboration. |
|
|
| ### Areas for Contribution |
| - 🧠 New element (agent) designs |
| - 📊 Quality assessment metrics |
| - 🔍 Memory optimization algorithms |
| - 🌐 Multilingual support |
| - 📚 Documentation improvements |
|
|
| --- |
|
|
| ## 🙏 Acknowledgments |
|
|
| **Built with 💜 by VIDRAFT ** |
|
|
| - Inspired by: Wu-Xing Philosophy (五行), Neuroscience, AGI Research |
| - Special thanks to the open-source AI community |
|
|
| --- |
|
|
| ## 📞 Contact & Community |
|
|
| - 🌐 Website: VIDRAFT.NET |
| - 📧 Email: arxivgpt@gmail.com |
|
|
| --- |
|
|
| <div align="center"> |
|
|
| ### 🌌 AETHER v2.2 |
|
|
| **"Evolving beyond Cycles. Intelligence that Remembers and Grows."** |
| **순환을 넘어 진화하다. 기억하고 성장하는 지능.** |
|
|
| Made with 🔥 by VIDRAFT |
|
|
| ⭐ Star us on GitHub if you believe in the future of Proto-AGI! |
|
|
| </div> |
