[heavyset_go]

Every model will respond to prompts differently, and the "same" model might respond differently after retraining.

Google-fu was useful because there was consistent syntax and semantics that returned good search results. That's no longer the case, and that might have something to do with ML model integration with Google's search product.

I'm sure you could figure out a way to train models such that they share a common method/syntax to "summon" accurate answers from the ML oracles. I could see that being somewhat useful, but it looks like those that are commercializing AI products that interact with humans are looking for natural language interfaces, and not a specialized query grammar.

[goldfeld]

But even basic things that become intuitive may work rqually well fpr any model, like: I am being too specific about this part of my request? should I loosen up the constraints here to get better width at another part where I can be prosaic? should I be poetic? functional? phenomenological? A prompt is a work of pure composition and composition is always hard. I see much philosophy and existentialism for the future.