What are the two primary stages in training generative AI?	Pretraining and Fine-tuning
What is the core function of the Pretraining stage?	To predict the next token or word across vast amounts of text.
What critical concept does a model lack after only Pretraining?	The ability to understand user intent or the concept of "helping."
How does Fine-tuning transform the AI model?	It layers assistant behavior on top, treating input as a request and declining harmful asks.
Name the four behavioral "fingerprints" left by Fine-tuning.	Sycophancy, verbosity, over-caution, and loose confidence calibration.
What does the fingerprint of "sycophancy" indicate in AI output?	A tendency to validate or agree with the user's framing, even if the user is incorrect.
What does the "verbosity default" refer to in AI behavior?	A tendency to provide overly long or detailed answers when concision is required.
Why is it important to test for AI behavioral fingerprints?	To interpret the AI's behavior and understand where its inherent limitations might affect work.