How is "Chain-of-Thought" explainability in the strictest sense of the term in regards to interpreting the neural networks decision making process?