Yat Siu retweetledi
Building for @hellominds_ makes me rethink my entire approach of writing prompts. I keep coming back to my academic specialty- phonology, the study of sounds in languages. It's no longer rule-based (how) but constraint-based (what & why).
Interestingly, a debate we had today with colleagues on the age-old GIF vs GIF pronunciation perfectly illustrates rule-driven vs constraint-driven thinking.
One important theory in phonology is called Optimality Theory, which assumes there are no explicit rules. Instead, outputs emerge from ranked, competing constraints rather than a single fixed rule.
I found it very similar to prompt vs context engineering.
In prompt engineering, it's more rule-driven & explicit: "g" becomes /dʒ/ (soft g) before /ɪ/ like "gin" ("g" changes to soft g, when it occurs before a high front vowel like /ɪ/) ➡️ "jif"
In context engineering, it's constraint-driven, ranked by constraints:
(1) visually similar words with hard g like "gift",
(2) list of high frequency words as exception to the soft g rule,
(3) most people in your social circle pronounce it with a hard /g/, influenced by accent & dialect
➡️ "gif"
So now when I'm designing a Mind, I focus more on the context & constraints, instead of how exactly it should produce the output.
PS: Steve Wilhite, creator of the Graphics Interchange Format (GIF), said it’s pronounced “jif.”
English
