Hard to explain, easy to verify

submited by
Style Pass
2023-01-24 22:30:10

In the past few months, the democratization of image generation models and improved usability of language models ushered in a wave dubbed “Generative AI”.

Amid this speculation, I think tasks that feel practically useful to complete with AI follow one pattern, which I call “hard to explain, easy to verify”. A task fits this pattern if it is:

There aren’t absolute thresholds for “hard” and “easy”; it depends on a task owner’s skills and their next best alternative to AI for completing the task.

Note that “hard to explain” describes the process underlying the task (i.e. how to complete it), not the outcome of the task (i.e. what success looks like).

Text - Whether it’s for marketing copy or fiction, it’s hard to explain how to write something well, but easy to verify it’s written well.

Images - It’s hard to explain all the steps to draw a cat in the style of an oil painting. But if I showed you one, it’d be easy to verify it depicts what I claim.

Leave a Comment