Maybe the useful unit is not a model, but a model-context pair.

via

PromptEngineering

I’ve been thinking about two separate observations from recent AI workflows. First: Different models can be useful because they see the same problem differently. For example, one model may be better at structure, another at expansion, another at critique. Second: The same model can be useful in different ways depending on whether it has context or not. A context-aware session helps build faster. A clean session helps reveal what the context was silently filling in. That made me wonder: Maybe the