3 comments

[ 3.6 ms ] story [ 19.8 ms ] thread
So prompt-engineering getting generally less and less necessary is a sign of what? nothing? impending plateau?

If the specialized expert information can leave the heads of the users once it can also be captured and absorbed into a better model - even if the lead time for that is weeks or months. Information wants to be absorbed.

I think the most important point is #3 about agent structuring. Building a good application means understanding the users’ needs and designing a workflow to help them accomplish their tasks. I don’t think most real-world use cases will reduce to “shove text in andn get text out”. Rather, I think pretty significant scaffolding needs to be built for each application, with models integrated with well-defined, atomic roles in the pipeline. If so, better models will make this work better and open new opportunities, but they won’t supplant the need for app development.
I also agree the most important point is #3. It’s the part that requires true intelligence, the ability to understand human needs. AI, until AGI is real, is best positioned as a tool that accelerates human productivity.