To listen to the discussion on AI you’d be forgiven for thinking that it is simultaneously the best, most innovative idea to come along in a century and also an electricity eating toy that’s somewhat amusing but not good for much. Reflecting the second perspective, I’ve seen plenty of articles about people trying to use AI systems—whether ChatGPT, Microsoft CoPilot, Gemini, or any of the others—and quickly losing interest. 

That behavior is not terribly surprising. Although AI looks like it’s easy to use, that is something of an illusion: AI is easy to use badly.  

Ersatz Intelligence 

AI Large Language Models (LLMs) do an impressive job of mimicking human intelligence. They appear to carry on conversations, they help inspire creative ideas, and they are remarkably good—when prompted correctly—at identifying unexpected and surprising connections between different concepts. And since we humans have only encountered one creature that can do all that—other humans—we reflexively respond to AI as if it were actually human. It’s not. 

AI has no consciousness, no self-awareness. It uses mathematical magic to generate its responses, and it has no actual awareness of what it’s saying.  

This can lead to a great deal of frustration for the user.  

When we communicate with another person, we automatically share a lot of assumptions about the world: ask a friend if they know a recipe that uses a particular ingredient, and our friend likely assumes we have that ingredient and will try to give us a recipe that uses it. An AI might ignore the ingredient completely when it gives us a recipe (I’ve had this happen many times). And sure, you can chat with the AI and correct it, but that can quickly get tiresome. Keeping track of and correcting answers that are wrong in less obvious ways (aka hallucinations) is another big issue. 

Ultimately, talking to an AI is still talking to a computer. It’s programming in everyday language.  

Which only raises the question: how can you benefit from AI without being a programmer, being incredibly patient, or—frequently—both?  

How To Use AI Successfully 

One approach is to really work on prompt writing, or prompt engineering as it’s more accurately called. Given enough time, practice, and willingness to write complex prompts, you can get good at it.  

Another approach is to use AI infrastructure software to manage your interactions. AI infrastructure software, such as SWIRL, picks up a lot of the heavy lifting involved in AI. SWIRL enhances prompts to reduce or eliminate hallucinations (using a technique called RAG, or Retrieval Augmented Generation).  

A particular benefit of SWIRL is that it radically simplifies gaining insights from your own data. SWIRL leaves data in place—no copying to a vector database or uploading to a server—so it’s particularly good for combining data only you can access (such as Slack, Teams, WebEx, email, and more) with information stored in company databases, giving you personalized information and insights. This alone can save users approximately 8 hours a week, or the equivalent of one full day of work.  

And because SWIRL is leaving data in place, your personal information isn’t being uploaded to someone else’s server. 

SWIRL eliminates much of the frustration involved in using AI, so you can reap the benefits: more accurate responses, fewer irritating “conversations,” and more time available to do the things that matter. Because, let’s face it, isn’t that the whole point of having a tool as powerful as AI at your fingertips? 

To find out more about how SWIRL can turn AI from toy to productive tool, contact us.