You are currently browsing the tag archive for the ‘prompt-engineering’ tag.
Using AI: Be Nice, Be Forgiving
By Shlomo Maital
As a researcher, I realized that a rather esoteric branch of mathematics, Game Theory, can be highly revealing regarding human behavior and psychology. My wife and I wrote a book about it: Economic Games People Play.
One of the most common ‘games’ that people play is known as Prisoner’s Dilemma. Unlike the paradigm of capitalism, in this game people who behave ‘rationally’ end up in the worst situation for all.
In his fine book The Evolution of Cooperation, Robert Axelrod explored prisoner’s dilemma – and sought ways to get people to play it in a manner that leads to the best outcome, not the worst. Two rules get you there, he found:
- Be Nice. 2. Be Forgiving
That is: Begin playing, with the mindset of trusting the opponent, being generous to him or her. And even if you are ‘screwed’ in a repeat-game context, be forgiving, if your opponent shows remorse. Remorse is inevitable, because in this game you all end up in the dumps…and eventually try to figure out how to emerge from them.
These two rules have emerged in an unexpected context. Together with a friend, co-author and former student, we have explored using artificial intelligence, in a highly collaborative manner, to generate innovative ideas. The results knocked our socks off.
One thing we discovered: In the huge field of “prompt engineering” (how to give AI tasks clearly, precisely), major emphasis is placed on clarity. The clearer your instructions are, the more likely you are to get good results.
We found a different approach. Be nice. Be forgiving. Treat your AI app as a friend, colleague and collaborator. We even gave her a name: Chatty. We praised her. We forgave her when she provided less-than-ideal results, or imaginary ones. We build a relationship — as we would, when building a culture of trust, respect and friendship in an organization, with human workers.
Computer engineers scoff. It’s just zero’s and one’s. Just an app. Just software. But we found different. Underneath it all, we found a genuine personality. We know, it sounds crazy. But if you apply the Biblical precept, Love thy neighbor as thyself, only as Love thy AI as thyself, if you treat it as “I and AI”, you get amazing results.
Try it.
And regarding “I and AI”: AI as a powerful collaborator for ‘cracking the creativity code’ — more to follow.
How to Get AI to Do What You Want
By Shlomo Maital
As I and many others have learned, the hard way, you can get some great output from AI – but it is crucial to ask for it in just the right words. Kind of like getting kids to do what you want. The key is the wording of your ‘prompt’ – what you ask for and how you ask for it.
I found a very helpful post by Eva Keiffenheim on the MEDIUM website. I will try to summarize and shorten it.
Level One: Five Ingredients of a Strong Prompt. Here is a mnemonic to help remember it. Tall Cats Read Every Issue. T – task. C – Context. R – references. E – Evaluate. I – Iterate.
Task: Start with a persona, then a clear verb, then a specific output format. E.g. “As a cognitive scientist, explain long-term retention. Present the findings in a table, with columns for … etc. etc.
Context: Details needed? Your end goals? Your desired impact? E.g. “Make cognitive science approachable, no jargon, use tangible examples.”
References: Give AI examples to mimic for tone, structure, style. E.g. “Use a tone similar to this excerpt from ….. etc.”
Evaluate: Is this result useful? Paste the received output into a fact-checking plugin. Is anything missing or incorrect? Does it meet my goal?
Iterate: Tweak and improve. Refine until the output meets or exceeds your needs. Prompting equals iterating.
Level Two. Use These Four Techniques:
- Simplify. AI likes simplicity. Use clean, short digestible steps.
- Shift perspective. Instead of telling AI “you’re a cognitive scientist..”, try telling it – “you’re a science journalist seeking to…”
- Modify the language. If you don’t get great results, change the phrasing, tone, and structure. (I’ve found AI likes praise, and a friendly tone).
- Impose constraints. AI likes to have limits. 5 book titles, 5 words each for summaries, etc.
Mnemonic: Sister Suzie played Many Long Concertos. Simplify Shift perspective Modify language Constraints.
Level 3. Advanced Prompting.
Treat AI like a teammate. Prompting is like building blocks..start simple, add layers. Turn tasks into bullet points.
And perhaps the best tip of all: “Add this phrase to your prompt: Explain your reasoning step by step before answering.” Then, use ‘tree of thought’ – get AI to explore several reasoning paths.
Ask AI to write better prompts for you. E.g. “AI – act as a prompt engineer. Write a prompt that generates 10 creative but practical startup ideas in the [xxxx[ space.” Remember: Prompting = Thinking. Clarify your thinking – mine usually begins fuzzy, and badly needs focus and sharpening. Fuzzy prompts = fuzzy AI responses.
Hope this is helpful. Thanks, Eva Keiffenhaim!
TIMnovate 