What are the two techniques discussed for strongly influencing LLM output besides changing the initial prompt?	Pre-filling assistant messages and stop sequences.
What is the primary purpose of pre-filling assistant messages?	To steer the LLM's response or direct its stance on a topic.
When pre-filling an assistant message, what assumption does the LLM make about the content provided?	It assumes the content was authored and continues its response from the end of that text.
How can pre-filling assistant messages be used to steer the model's opinion?	By including a partial, biased statement (e.g., "coffee is better because") in the assistant message.
What is a stop sequence?	A specific string or sequence provided to force the LLM to stop generating output immediately upon generating it.
What happens when an LLM generates a defined stop sequence?	The entire generation immediately stops, and no further text is produced.
How can stop sequences be used to refine output formatting?	By including unwanted characters (like commas or spaces) in the stop sequence to clip them from the final output.
In the context of pre-filling, what is the practical limitation of the generated response?	The LLM continues from the end of the pre-filled text, often resulting in an incomplete sentence that requires manual stitching.