1) Mission, Product, Structure
Why: Making AI truly intelligent, error-free, and ethical. Our partner, not nemesis
What: The first knowledge map and concept database in humankind
How: Develop a parsing module to accrete two databases relying on algorithms and humans-in-the-loop hybrid method. Database #1 containing concepts with comprehensive lists of attributes, properties, and inter-conceptual operator, and Database #2 comprising facts and knowledge accompanied by some descriptive elements.
2) Problems Solved
- LLMs and WOMs understanding is shallow. They do not truly embed semantic knowledge and make mistakes
- LLMS and WOMs are incapable or zero-shot learning
- Those models cannot be interpreted, cannot unlearn and cannot be ethically constrained with certainty
- Perplexity results are not systematically reliable, and they are subject to copyright infringements
- Existing models require heavy, expensive compute
3) The End Game
Once the databases are comprehensive-enough and the parsing module fine-tuned, capabilities can be expanded in two steps.
- Add perception and inference modules to permit scene understanding and zero-shot identification and knowledge generation => reasoning without planning.
- addition of planning and evaluation modules, thus yielding full reasoning capabilities coupled with factual knowledge that can be constrained with understood ethical values, i.e. safe AGI.
This end result is described as Project Zero Shot
4) Investor pitch
Diamant is building the maiden knowledge map and concept database with the ultimate goal of making AI truly intelligent, error-free, and ethical. The databases can be used to a) provide education-related services, b) answer knowledge-related questions from app users and c) by LLMs to control the quality of their output (e.g. fact checking) via APIs.