Self-prompting — leveraging your mental cockney builder
When I am doing DIY jobs at home, I have a cockney1 builder who gives me useful tips in his Bob Hoskins accent2. He generally tells me that the way I’m approaching a job is too time consuming; to stop wasting time and to work more quickly. Not a real cockney builder, just in my head. My mental cockney builder is a useful reminder to me that professionals tend to approach building tasks more efficiently than amateurs.
1 This is of course stereotyping, but I think it’s positive stereotyping because cockney accents are good for delivering blunt truths forcefully. You’re only supposed to blow the bloody doors off!
2 I’ve just realized that I can only do internal voices in the accents I can impersonate with my physical voice. That’s curious. I need to investigate that.
A widely used technique when prompting AI is to give the AI a persona to use when processing the task. For example:
- Write a product description for this new product as if you are a professional copywriter, focusing on its innovative features and benefits.
- Explain the importance of regular exercise for heart health as if you are a renowned cardiologist.
- Describe the impact of climate change on our planet as if you were Richard Feynman, emphasising scientific principles and future implications.
I recently used this technique to good effect, asking the AI to respond as if it was Alan Watts when considering a bit of instagram wisdom[^3]. It’s a great method and can result in a much better response from a generative AI. [^3]: Comparing ChatGPT, Perplexity, Claude and Gemini
I’ve come to realize that I have been using a similar mental technique for years on my own mind. When thinking about a question, it’s often good to consider it from the point of view of someone who is really good at that subject. What would Steve Jobs say about your product idea? What would William Blake say about your poem? Sometimes a blunt truth becomes obvious when you consider it from the point of view of an expert.
This makes me wonder, if we can get good improvements in results from generative AI if we spend time learning how to make great prompts, can we do the same with our own minds? I guess applying mental models is one method to do that. The “ask Steve Jobs” method is about approaches and problem-solving styles. Can we formalize that? #TODO
My cockney builder thinks my writing is “a bleedin’ waste of time”, but I made a mistake asking him. Let’s see what Hemingway says…
Hemingway also thinks my writing is a bleedin’ waste of time, but he did say to keep on writing, so that’s good.