The First Step to Using AI (LLMs) well is not Prompt Engineering.

Khan said to Captain Kirk after waking up after a 2 centuries long nap...

"I am very surprised how little advancement has been made. . . Improve a machine you may improve productivity, but improve MAN, your increase is a thousand fold."

AI seems to give us all a taste of being a bit super human. And if that's true, are newest emerging super hero might be Prompt Engineers. As we a observe the emerging capability (almost magic) of GPT and it's ilk, we think we might need this new skill to unleash great new horizons of value. But something struck me as off about that logic. Are we expecting Large Language Models to be still at their current level for years on end while AI-educated linguists translate our wishes into the oracle?

This would create yet another inefficiency and bloated function that drains the effectiveness of the promise of AI. Problem solving for YOU. It seems to me that later models are going to wise up to this AI-interaction limitation and get better and better at deciphering what we humans really want. Similarly, we humans are capable of learning, one would hope.

As we all learned how to make Google's search engine do our bidding in the 90's, we will learn how to make AI models dance after some mere practice, improving incomplete answers and sub-optimal responses with human or other AI or machine interventions. Our magic.

So as we invest in prompt engineering in the short term, it should be with that mindset. A short term skill that is designed to be absorbed to near nothing in the next 2 to 5 years.

What Should Be our Focus?

But what do we focus on to get the most out of AI's broader accessibility? Now that we have answers on tap to all questions, it might be nice if we answered the right ones. A nice piece from our friends at HBR on problem formulation is a good entry point to problem solving, which if practiced to even a low level of mastery, should help you know if we are working the right problem.

Why This Focus?

A more enduring issue we are going to have to contend with is the rampant human-ness embedded in AI models. That's good because finding out when we are being fed a line or if a key data point is just made up, should always be our super power, in any business or serious life endeavor. So, the world of QA marches on with old school controls, validation exercises and rigorous human and machine testing. Trust almost, nothing.

This is a healthy road forward. It's not an easier road.

I know we were all hoping to slide into the AI pod, hook up, and let the AI do all the drudge work. But doing the grind qualifies us for the high level thinking. A world of top down thinkers is a world where there is no mastery. No sacrifice. No ladder to distinguish true believers from charlatans and true passion from mere passing interest.

Which do you want to rise the top?

A perfect AI would elevate us all equally and without merit or prejudice. The best Power-Pointer wins. This confuses our merit-based view of life and work.

So roll up your sleeves and get out those problem solving models. They were mildly interesting before, but now, with the ability rapidly frame issues, create 10X more noise (ok, content) and with everyone having the power to code automation into everyone else's work and life, we are all going to need to be much faster, effective and revered problem formulators to use these tools effectively.

It's not the conversation with the oracle that is the magic. Once again, the real magic was improving man and woman (not as much though!). That work happened in the quite hours of the day, with a pen and paper, keyboard and screen, or whiteboard, and a great conversation with a caring colleague, grinding past symptoms, aftereffects and signals, to what is really going on with us and our businesses.

K