If Communication is Oxygen, and Prompt hacking is essentially communication, then…
1: Communication is Oxygen
At Automattic, we like to say that communication is the oxygen of a distributed company. Why is that exactly?
For remote work to work, you have to provide sufficient context so your coworkers are on the same page. They cannot grab you across the desk to clarify what you meant, and any follow-up question can take up to a full day to be answered.
We try to share context early, and preemptively. We prepare information for each other to be easily surfaced and reached. We stuff the context window.
2: Prompt hacking is Async Communication.
With the release of ChatGPT, prompt hacking became a thing that captured the imagination of journalists around the world: “If only I describe my task in this weird way, AI will do it.“
Now, with the initial hype cycle stabilizing a bit, they are deeming it dead and a bygone fad.
For me, it was never it’s own thing. When I try to write a prompt, it is no different than describing a Github issue:
- Describe the task to be done in sufficient detail
- Try to be mindful of the receiver’s context (window)
- Share as much information as possible to avoid follow-up questions
- Describe the shape of the result
- Sidenote: I use please a lot because I don’t see the prompt as something different than any other “delegation”.
I am trying to set up the other party for success and make it easier for them to deliver the result I would like to see.
3: Prompt hacking is oxygen
This post started as an interesting title, but it got me thinking about the implications:
- Async communication is essentially zero-shot prompting. How can we use our async work expertise to elevate our prompts?
- What other learnings from Remote work setup can make us better at working with AI models?
- Can learnings from prompt hacking make us better at managing?
We’re still early.