The Little Prince vs the Utilitarian Philistine
Why does this matter?
This short story showcases what LLMs are great for (coherence) and what’s hard to achieve with them (consistency).
Consistency (hard)
Unlike traditional computing and creation, LLMs today destroy all previous creation whenever you create a new prompt. That makes it very hard to achieve consistency across outputs and makes it really difficult to attain consistency across models. Definitely a great opportunity for a new set of tools to start capturing state and allow for a more permanent and sticky way to experiment with preferences.
If you are interested in learning how a prompt works, I recommend following Nick St. Pierre @nickfloats. For instance, check this simple yet powerful way to think about prompt structure for Midj-ourney:
Coherence (easy)
What makes LLMs so different and so powerful is that they help find and create coherence across disperate ideas or point of views. They are a great tool to inspect, associate, and brainstorm with. For the Little Prince tale above, I wanted to created a short story that included my favorite book (The Little Prince), my favorite song (Flaming Youth by KISS) in a contemporary situation.
Go ahead. Try to find coherence. It’s never been this easy.