The Art of the AI Prompt

AKA how to talk to Large Language Models like a human

Don't miss the mark illustration

1.

Don't miss the mark

A lot of energy has gone into one-shot prompts. These miss the mark.

Large Language Models are built to simulate human conversation and predict the next conversational unit. Humans don’t converse by showing up and giving all of the information all at once.

That’s not a conversation, it’s a boss showing up, giving orders and then being surprised when they don’t get a useful answer1.

Do

Hi, I need to prepare a workshop series for a client. Let’s have a conversation to work through how we can structure a 5 day series. What questions do you think I should be asking to best prepare for it?

Don’t

I’m an innovation designer, I work at a UK-based agency, our clients are nonprofits and charities. I am organising a workshop for a UK-based charity with revenue of £100m / year. I need to prepare

  • x,
  • y,
  • z,
  • a,
  • b,
  • c,
  • d,
  • e; and
  • f

The workshop is 5 days long. Channel Don Norman, Jakob Nielsen and Jake Knapp. Synthesise each day and give me a rationale about each day’s activities that can be shared with their leadership team.

Footnotes

  1. Since LLMs are simulating human conversation they may also be displaying the tendency for subordinates to not give full answers. There’s lots of academia around social hierarchy but Deborah Tannen and Pierre Boudieu both have interesting viewpoints on it. ↩