Substantially, the idea is: how can you assess reliability of the framework on which you would be building specialized functions, if the framework is as impredi

Substantially, the idea is: how can you assess reliability of the framework on which you would be building specialized functions, if the framework is as impredictable as a LLM? (Although the article offers more sides to the problem, and stresses on the grave engineering risk of letting data and commands share the same channel.)