On this video, Charles provides high-level intuitions and a default playbook for prompting language fashions. We think about two very completely different sources of instinct: “language fashions are statistical fashions of textual content” and “prompts are magic spells”. Then, we overview prompting methods, like decomposition, reasoning, and reflection.
Obtain slides from the bootcamp web site right here: https://fullstackdeeplearning.com/llm-bootcamp/spring-2023/prompt-engineering/
Intro and outro music made with Riffusion: https://github.com/riffusion/riffusion
Watch the remainder of the LLM Bootcamp movies right here: https://www.youtube.com/playlist?list=PL1T8fO7ArWleyIqOy37OVXsP4hFXymdOZ
00:00 Intro
02:15 Language fashions are statistical fashions of textual content
04:50 However “statistical mannequin” provides dangerous instinct
08:35 Prompts are magic spells
09:57 Prompts are portals to alternate universes
16:42 A immediate could make a want come true
23:03 A immediate can create a golem
27:10 Limitations of LLMs as simulators
32:17 Prompting methods are largely methods
33:40 Few-shot studying is not the correct mannequin for prompting
37:40 Character-level operations are onerous
40:23 The prompting playbook: reasoning, reflection, & ensembling
