Your book Rebooting AI offers a well considered solution. Build a knowledge graph (ontology) that covers human knowledge in a taxonomy of concepts. A semantic web. The scale needed to is on the order of ten million concepts with a branch depth of 5-10 edges. (2-3x Wikipedia) The ontology connecting concept nodes is constructed by NLP usi…
Your book Rebooting AI offers a well considered solution. Build a knowledge graph (ontology) that covers human knowledge in a taxonomy of concepts. A semantic web. The scale needed to is on the order of ten million concepts with a branch depth of 5-10 edges. (2-3x Wikipedia) The ontology connecting concept nodes is constructed by NLP using Common Crawl to extract 50-100 billion RDF triples and classify subject/object predicates to connect nodes by relationships. This semantic AI model (SAM) is the solution you posit.
LLMs might perform much better with long tail knowledge trees grouping tokens by topics rather than starting with random weights. SAM could be used to detect factual errors and hallucinations. Even red team the LLM or construct steering prompts to align the LLM with legal or other constraints. Investment in a Web 3.0 SAM (reading and curating) can save LLM (write only).
Your book Rebooting AI offers a well considered solution. Build a knowledge graph (ontology) that covers human knowledge in a taxonomy of concepts. A semantic web. The scale needed to is on the order of ten million concepts with a branch depth of 5-10 edges. (2-3x Wikipedia) The ontology connecting concept nodes is constructed by NLP using Common Crawl to extract 50-100 billion RDF triples and classify subject/object predicates to connect nodes by relationships. This semantic AI model (SAM) is the solution you posit.
LLMs might perform much better with long tail knowledge trees grouping tokens by topics rather than starting with random weights. SAM could be used to detect factual errors and hallucinations. Even red team the LLM or construct steering prompts to align the LLM with legal or other constraints. Investment in a Web 3.0 SAM (reading and curating) can save LLM (write only).