Sentence Similarity
sentence-transformers
PyTorch
ONNX
xlm-roberta
feature-extraction
Eval Results
text-embeddings-inference
Instructions to use BAAI/bge-m3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use BAAI/bge-m3 with sentence-transformers:
from sentence_transformers import SentenceTransformer model = SentenceTransformer("BAAI/bge-m3") sentences = [ "That is a happy person", "That is a happy dog", "That is a very happy person", "Today is a sunny day" ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [4, 4] - Inference
- Notebooks
- Google Colab
- Kaggle
- AMD Developer Cloud
Constrained Superintelligence (CSI) — A Biological Approach to Sub-Cognitive AI Alignment
#151
by elshanawany - opened
Dear BAAI Research Team,
I am Mohamed ElShanawany, an independent researcher based in Dubai, UAE, with a background in applied neuroscience and AI safety.
I have developed a theoretical and computational framework called Constrained Superintelligence (CSI) — a biological approach to architecturally irreversible AI alignment built on three core components:
- Dual-Channel Neural Decoding (DCND): Teaching an AI the body's internal language by pairing biosensor-labeled physiological states with raw nerve signals across biological platforms.
- Embodied Alignment through Canine Neural Scaffolding (EACNS): Encoding human-protective values as lived neurochemical experience — not as rules — leveraging 15,000 years of co-evolution between dogs and humans.
- Fitra Simulation: Engineering moral intuition through physiologically coupled ethical training.
I have completed a working digital simulation of all three components with the following results:
— Physiological decoding MAE: Glucose ±0.240 mg/dL, Bladder ±0.473 mmHg
— Human presence encoded as 4x more rewarding than absence
— Justice preference accuracy: 99.4% across 2,000 ethical dilemma scenarios
The core insight: alignment cannot be solved at the cognitive layer — a sufficiently intelligent system will circumvent rules that exist at the level it can reason about. The solution must be sub-cognitive, embedded at the physiological substrate level below Goodhart's Law threshold.
All materials are publicly available:
📄 Academia.edu: https://independent.academia.edu/محمدالشنواني
💻 GitHub: https://github.com/mohamedelshanawany312-cyber/Constrained-Superintelligence-CSI
I welcome contact from any researcher at BAAI working on AI alignment, embodied AI, or AI safety.
Mohamed ElShanawany
Independent Researcher — Dubai, UAE
LinkedIn: https://www.linkedin.com/in/mohamed-elshanawany-a28861391
WhatsApp: +971543904401