This paper maps concepts from AI alignment onto a basic, three step interaction cycle, yielding a corresponding set of alignment objectives: 1) specification alignment: ensuring the user can efficiently and reliably communicate objectives to the AI, 2) process alignment: providing the ability to verify and optionally control the AI's execution process, and 3) evaluation support: ensuring the user can verify and understand the AI's output.

the notion of a Process Gulf, which highlights how differences between human and AI processes can lead to challenges in AI control.

The key design skill is less about beautiful all-encompassing Figma documentation with all the kinds of journey maps and personas, or mastering a “process”. It’s about being so keenly situationally aware of what unknowns are in front of you so you can pick out what tools or design activities target them precisely.

It's not helpful to think of design as process of discovering, defining, ideating and delivering (or whatever version of those double diamond steps you prefer) because following those steps often does not get you closer to a solution.

Every software framework you've ever used is in the abstraction game: it takes a general-purpose tool, picks a specific set of common use-cases, and puts up scaffolding and guard rails that make it easier to build those specific use cases by giving you less to do and fewer choices to think about. The lines between these three are blurry. Popular abstractions become standardizations.