The engineering world revolves around expertise. We are looking up to people who know stuff or can do things. And for a good reason. Those are the people who can figure things out, build cool tools, and share their knowledge. We are all better off because these people are around.

But there is a darker side to the expertise. All the years of school, experience, time spent on honing one's skills, create invisible walls in expert's mind. Deeply learned patters, preferences, rules, the dos and the donts. You no longer see the problems as they are, but you see them through the filter of what you already know. The hypothesis for solving problems, consciously or subconsciously, get rejected, not based on merit, but based on ingrained patterns. This will never work or this is not how it's done.

In many situations, this is great. This IS the expertise: the ability to reject the dead-ends without spending days down unsolvable rabbit holes, and find the solution quicker.

But sometimes, every once in a while, the experts tend to reject the ideas before fully exploring them, leaving some good solutions on the table. Often for reasons, that they themselves would be hard-pressed to explain. They may have simply learned long ago that this is not how things are done.

"In the beginner's mind there are many possibilities, but in the expert's there are few"
-- Shunryu Suzuki

Notice, how often it is the junior engineers who are first to try new approaches and new tools or combining them in some unusual ways. Sure, their ideas are often proven dead-ends and rabbit holes. But again, every once in a while, they manage to put something together that suddenly clicks, in an unexpected ways, despite the common wisdom. Every so often, they discover a good path that the experts rejected long ago. Simply because they weren't limited by the prior assumptions and were willing to explore, honestly, from the first principles.

We are often reminded to "think outside the box." But if you think outside the box, you are still in the box and all you see around you is walls.

Which brings me to AI, or more specifically the current LLMs, which are most certainly Artificial, but most definitely not Intelligent. I recently started to refer to them as Artificial Ineligibles.

LLMs are "the ultimate experts." They absorbed vast amounts of data and published rules, and they seemingly know everything, or short of that, respond as if they do. Works great when they asked for the solutions to the problems that are well known. They can spin up a fresh website in a couple of minutes or port some code to another language (somewhat). But when faced with an unusual problem, they quickly hit the wall.

This is not surprising. If the question wasn't answered on a website somewhere, why would LLM have an answer for it? So it spins in circles, trying to apply known (to it) patterns to the problem, wasting your tokens and time, and just keeps coming up short. Often this is not just in-the-box thinking, it's in-the-box-buried-six-feet-under-ground thinking.

So if we blindly accept LLM's solutions, not only we are going to continue banging our heads against the walls. But we will also end up spiraling into a homogeneous world, where everyone is using the same five tools and three frameworks. Not because better alternatives for the task do not exist, but because they statistically do not exist for LLMs.

If all you have is 10,000 hammers, everything still looks like a nail.

LLMs are tools. They are great for some things. Not so much for others and definitely not good at creative problem solving.

Don't forget to use your Beginner's Mind.