‘Good enough’ prompting

I really liked Ethan Mollick’s thoughts this week about ‘good enough’ prompting for AI tools. He references a study of Doctors using GenAI tools which revealed some interesting observations around algorithmic aversion (our hesitancy to take advice from machines when they conflict with our judgement, even when the machine may have a higher degree of accuracy) and also how the Doctors were using the AI more as a search engine and consequently struggling to get the value out of it.

This, notes Ethan, is a common problem and the answer is not necessarily to learn complex ‘prompt engineering’. Framing it as a complex learning task will simply discourage people from trying to understand how to work with AI, and anyway there is no clear science yet on how to do this (‘researchers are still arguing over the most basic foundations of good prompting. This is because AIs are inconsistent and weird, and often have different results across different models’). The need to know how to do complex prompting may be less of an issue over time anyway – AI’s are getting less sensitive to nuances in prompts, but also better at interpreting intention, and there are tools that can help (I like this prompt generator from Anthropic which Ethan mentions). 

He suggests two pathways to get started…good enough prompting for tasks, and good enough prompting for thought. For the former, try it out on lots of different tasks that you do in your day to see how it does. He says to ‘treat the AI like an infinitely patient new coworker who forgets everything you tell them each new conversation’. For example, give it good and bad examples, step-by-step directions, feedback and so on. When it comes to good enough prompting for thinking, you should simply use it as a thinking companion and converse with it using natural dialogue (maybe using the voice-driven interfaces) and not overcomplicate it. I’ve also personally noticed that politeness gets you better answers, spookily enough. You can use it, says Ethan, as a rubber duck: ‘…the popular idea in computer programming that, if you explain an issue to an inanimate rubber duck on your desk, you will work through the problem on your own by talking it out.’

Useful advice.

I write a weekly Substack of digital trends, transformation insights and quirkiness. To join our community of thousands of subscribers you can sign up to that here.

To get posts like this delivered straight to your inbox, drop your email into the box below.

Photo by S. Tsuchiya on Unsplash