Then someone actually did the research. The “please effect” turned out to be false. In fact, being a little rude proved marginally more effective.
Telling your AI to “think harder”, however, genuinely works – and the reason is pure economics. When you signal that you want a real effort, your provider is nudged to route your request to the cutting-edge models you’re entitled to use, but which their creators would rather you didn’t, because they’re ruinously expensive to run. Left to their own devices, these companies will quietly push your query to the cheapest model they think they can get away with. Fight back by selecting “thinking” mode in the settings on ChatGPT and Gemini – it’s there, and you should be using it.
What other science-backed tips and tricks are there for getting the best out of your AI?
The shine has come off “prompt engineering” – the idea that there’s a special set of magic instructions for wringing better answers out of your AI. But research shows there really are some modes of communication that get better results, and save you time and frustration in your AI interactions.
These usually leverage my own method of “Lean Prompting”. Its premise is that you’re not going to get the best answer from a single long prompt that tries to anticipate all contingencies. Instead, you should prioritise several short, back-and-forth iterations that allow you and the AI to hone in on the perfect answer
A first tip is to ask AI to make your half-formed question into a solid first prompt. Dump your ugly notes and scratchpad-type thoughts into the chatbox and say: “These are my thoughts, infer my intentions and refine this into a clean, clear prompt.”
A good refinement at this point is to provide more context. Instead of racking your brains to think what that might be, ask the AI. Say: “Good start. What information would you need from me to give me an even stronger prompt? Top three questions.”
Now run the prompt, then iterate. If the reply is in the right direction, say, “Close, but off-point here [insert the problematic part]. Fix that, keep the best bits of the orignal and be more practical/original/succinct.”
This type of dialogue will get you, incrementally but quickly, to a better answer than you’d get by asking just off the bat. But remember that these LLMs are designed to sound convincing – constantly inserting the next most plausible token.
So before you take any consequential action on the back of AI advice, :“red-team” it. You can do this by asking: “Think hard and persuade me what you’ve just said is wrong.” Another option is to enter the response into a rival chatbox and tell it to shoot it down.
The thread running through these techniques is to treat every exchange as a negotiation – pushing back, stress-testing, and demanding better.
It’s not the case, though, as some experts will tell you, that all the critical thinking in their process has to come only from you – you can get AI to improve your critical thinking, as by asking it to hone your prompts, to request the context it needs, and by asking it to re-examine its own thinking
Lean Prompting is a habit of mind. Start quick and dirty, refine ruthlessly, and interrogate the answer before you act on it. Master that loop and you won’t just get more out of AI – you’ll stay smarter than it.
Joe Smith is Founder of the AI consultancy 2Sigma Consultants. He studied AI at Imperial College Business School and is researching AI’s effects on cognition at Chulalongkorn University. He is author of The Optimized Marketer, a book on how to use AI to promote your business and yourself. Contact joe@2Sigmaconsultants.com.