Dan Selman wrote: “The key to Knowledge Representation and Reasoning is building your ontology. It doesn’t matter if you are reasoning using a Knowledge Graph, a rules engine or a relational database — you need to do the work of understanding the concepts in your domain, and their properties and relationships.

That is fundamentally different to “find me chunks of text similar to this chunk of text”.

I think folks are getting confused because an LLM can be a useful aid in understand the concepts in a domain, extracting instances of known domain concepts from text, and searching for text based on vector embedding, and they are a Swiss Army knife for data format transformation and text generation.

Let’s use the right tool for the right problem. LLMs are great at many things, but reasoning is not one of them.” Link