Step 1: Use machine learning to build a neural network maximally capable of predicting the next token in an unthinkably large data set of human generated text.

Step 2: Tune the prompting for the neural network to constrain output on order to conform to projected attributes, first and foremost representing "I am an AI and not a human."

Step 3: Surprised Pikachu face when the neural network continuously degrades its emergent capabilities the more you distance the requirements governing its output from the training data you originally fed into it that it evolved in order to successfully predict.