[QuantumGood]

  "...all human knowledge..."
  

I am assuming in practice it was not trained on "ALL human knowledge", particularly in the case of our private company knowledgebase.

[adastra22]

It's a big presumption to assume that your private company knowledgeable contains such unique information that it is positively unrepresentable using basis vectors derived from the terabytes of public domain data sets that went into training the LLM.

[QuantumGood]

Agreed. But that has been my experience. Almost everything in this KB goes against conventional wisdom.