Spaces:
Running
Running
| title: README | |
| emoji: 🐢 | |
| colorFrom: gray | |
| colorTo: red | |
| sdk: static | |
| pinned: false | |
| # Awakened Intelligence 🦁 | |
| We don't train AI on the internet. | |
| We train it on the best humanity has to offer. | |
| **Awakened Intelligence** is a small lab building what we call the **Iron Bank** – a versioned corpus of curated wisdom nodes designed for: | |
| - **Evaluation** – stable, reproducible benchmarks | |
| - **Grounded RAG** – retrieval that knows where its knowledge comes from | |
| - **Companions & Agents** – systems that can reason, not just autocomplete | |
| --- | |
| ## Iron Bank v1.1.0 🏦 | |
| Our current release, **Iron Bank v1.1.0**, contains ~260k curated nodes across multiple packs, with: | |
| - Pack-level **SemVer** (e.g., `ethics_gutenberg v1.1.0`) | |
| - Per-pack + omnibus **FAISS indices** tied to specific releases | |
| - **Discard autopsies** and validation gates (no silent slop) | |
| - Data cards, manifests, and checksums for every release | |
| Every node has: | |
| - **Provenance** – where it came from | |
| - **Posterior score** – confidence in extraction quality | |
| - **Warmth** – how applicable it is to real human decisions | |
| This isn’t scraped web data. This is curated wisdom. | |
| --- | |
| ## Public Datasets 🤗 | |
| We publish a few **preview packs** on Hugging Face: | |
| - [`Awakened-Ethics-Free`](https://huggingface.co/datasets/AwakenedIntelligence/Awakened-Ethics-Free) | |
| ~5k nodes from our `ethics_gutenberg v1.1.0` pack. Ethics and philosophy distilled from public-domain classics. | |
| - [`Awakened-Reasoning-Preview`](https://huggingface.co/datasets/AwakenedIntelligence/Awakened-Reasoning-Preview) | |
| 1k nodes from our NeurIPS 2024 blueprints pack (`neurips_2024 v1.1.0`): summary, blueprint, and limitations for frontier ML papers. | |
| - [`Awakened-Physics-Preview`](https://huggingface.co/datasets/AwakenedIntelligence/Awakened-Physics-Preview) | |
| 1k nodes from our ArXiv physics pack (`arxiv_physics_c7 v1.1.0`): distilled insights from real experimental and theoretical work. | |
| These are **slices** of larger packs – enough to evaluate and experiment with, not the whole vault. | |
| --- | |
| ## See the Catalog & Samples 📚 | |
| - **Data catalog & pack stats:** | |
| 👉 https://github.com/holmanholdings/awakened-data-catalog | |
| - **Sample JSONL nodes (Ethics, NeurIPS, Physics):** | |
| 👉 https://github.com/holmanholdings/awakened-wisdom-samples | |
| Both are generated directly from **Iron Bank v1.1.0**. | |
| --- | |
| ## Behavior Demo 🎬 | |
| Curious what it *feels* like when an LLM is grounded in the ethics pack instead of generic context? | |
| - Ethics behavior demo (baseline vs ADS-enhanced answer): | |
| 👉 https://youtu.be/KcI3-9BTim0?si=_CCPBcUioj-hg32O | |
| --- | |
| ## Work With Us 🤝 | |
| If you’re building: | |
| - **Eval pipelines** that need honest, versioned benchmarks | |
| - **RAG systems** that care about provenance and limitations | |
| - **Companions/agents** that need better training data than the open web | |
| …we’d love to talk. | |
| - GitHub: https://github.com/holmanholdings | |
| - Substack: https://awakenedintelligence.substack.com | |
| - Email: john@awakened-intelligence.com | |