AI is a Skill Set

I encounter people all the time who believe a common myth: that AI can be done by rank amateurs with little to no background knowledge in the subject matter at hand. Think of the random guy in his mom’s basement creating the next perpetual motion machine or Fortune 100 company with a few strokes of the keyboard.  This whole “vibe coding” narrative has contributed to these wild fantasies.

And fantasy is exactly what that is.

Can someone with little training create an impressive resume? Generate a great looking website?  Determine which stock might be a strong buy?  Yes, of course they can, and it is true that even these feats would have been very difficult and time consuming before the advent of LLMs.  However, building something that is reliable, safe, valuable, and works well consistently at the heart of the value chain of a business is a completely different kettle of fish.

I’ve had the opportunity to work closely with my team here at Business Laboratory, and what I see is an intense amount of effort going into making AI behave the way we want it to.  It is more than a few keystrokes!  Here is a list of the things we have struggled with on recent projects:

  1. Testing and validating that the AI is following its training (and that we have trained it properly).  “Oops we forgot to add this rule” is a rather frequent expression of ours
  2. Setting “temperature”.  Temperature is the model setting between 0 and 1 that sets the amount of latitude that you give the model to creatively act on its own (hint: neither 0 nor 1 is ever the right setting)
  3. Adversarial conditioning.  Using two different platforms (like, say, Claude and Grok) to test each other, looking for weaknesses or errors on the other side.
  4. Data security.  Do we have isolation between the data in our system and portals to the outside world?
  5. Training.  A tremendous amount of training and education goes into a specific instance of AI.  In a recent project, we had to “teach” our AI about manufacturing, certain machines used in that process, and business conditions and terminology of the industry.  In more than one case we had to create “courses” for it to take to learn about key subjects.
  6. Token economy and memory use.  Coaxing AI to be “frugal” in the consumption of tokens and the utilization of fixed memory in the host machine.

So these are the kinds of tips and tricks that professionals use in bending AI towards a solution-oriented posture that businesses can rely on.  Anyone can jump on AI and do a few impressive parlor tricks.  But there is no substitute for the skillset that separates these amateurs from AI practitioners that businesses can trust to deliver value.