I’m a late adopter. Always have been. I was still using a flip phone a year after all my colleagues had switched to iPhones.
My skepticism of AI was therefore not unexpected. To be fair, the earliest promises of AI touted the ability to write letters and papers and emails on my behalf, a benefit I rejected out of hand. After all, I take pride in my use of the English language, my ability to express myself just so in any situation. No digital ghostwriters for me, thank you.
Puppy love
My reservations began to melt though when I discovered that AI can produce custom images on demand, six-fingered humans with three arms and all. With enough tweaking I was now able to create a blog post-relevant image rather than spend valuable time scrolling through what I hoped were royalty-free stock photos. Ooh, ahh. AI was useful for me after all!
Then came the real breakthrough: I discovered Chrome’s Gemini AI assistant quite by accident. I was searching Google for some detail using the standard two- or three-word keyword and was suddenly rewarded with—oh my garsh—paragraphs of detailed information!
What was this sorcery? Could it be the result of the enigmatic “prompt” I had spotted in overly effusive fan-boy articles about AI? Yes, yes it was!
That serendipitous start was the beginning of a growing excitement with and use of Gemini AI to solve actual problems. How to diagnose and repair a garage door opener with stripped gears! How to source and install a check valve for an air conditioner condensate drain! How to attempt a repair of a bathroom exhaust vent and replace it after giving it last rites!
I was hooked, big time. I started using AI for mundane requests and for trickier ones: which of two Ukrainian operettas spawned a common phrase my Ukrainian mother-in-law frequently used? It turns out that Gemini speaks any language on Earth as long as it’s been trained on it.
“Iceberg, right ahead!”
Then came the crash. The realization that my Pollyanna-ish love of the real-life version of SkyNet was not quite ready for prime time after all.
I was chatting away happily with Gemini about township plumbing codes concerning a thorny problem I was solving. All was going swimmingly…until it wasn’t.
I had asked AI about draining condensate into a sump pit with a sealed cover and was told sure, as long as I used a watertight fitting to penetrate the cover. I asked if that complied with the required air gap code and AI said, “my bad, you’ll have to penetrate the cover with an open-top standpipe to receive the condensate.” I persisted with, “but won’t an open top standpipe defeat the purpose of a sealed lid?” AI was extremely unctuous in admitting that yes, that’s not gonna work, you can’t drain condensate into a sump pit after all.
Really? Three attempts at advice in one chat session that all ultimately came to naught?
That was not the end of it, though. I asked about repairing an existing condensate drain line that exited through the outside of my house. I made it clear that the interior was finished and I did not want to start cutting through sheetrock. AI’s response? “Push the condensate line from the inside through a narrow PVC pipe.” Hello? And how shall I push it from the inside without opening the sheetrock wall? AI: “Sorry, my bad again.”
It didn’t end there. AI had originally approved the option of draining condensate into a sump pit, then several chat sessions later informed me that this was specifically against the New Jersey plumbing code. I then checked the code. Draining condensation from clean household air into a sump pit filled with clean groundwater is not proscribed by any section of the code.
My relationship with Gemini was now on the rocks. How can I trust a partner who lies to me? And mind you, this was not the last such offense on the part of what I had hoped to be my savior in all matters technical. On one occasion Gemini suggested that there were recessed screws in the bottom of a shredder I was trying to dismantle. When I noted that I couldn’t see anything resembling screw heads in the holes, Gemini actually informed me that the screws were hidden and required a “special tool” to remove them, describing what sounded like a Phillips screwdriver bent 90 degrees near its tip. Sure.
Redemption?
OK, I was crushed. But rather than dump the baby out with the bathwater, I continued wooing my newfound partner with caution. “Trust but verify,” as Ronald Reagan famously said.
Then a few days ago as I was busily toiling in real estate-related Stuff an idea came to me that required doing some fancy calculating to try to convince buyers that climbing mortgage rates weren’t all that bad if offset by making a correspondingly lower offer on a home. I’m a whiz at Excel calculations, but I was bored spitless and decided to ask Gemini to scare up some Visual Basic code to do the task in a far more complicated yet far more satisfying way for a geek like me. I would have the satisfaction of checking AI’s code while possibly adding some new Visual Basic tricks to my already-substantial collection.
It worked. I had heard tell that AI can write code given a comprehensive enough prompt, and now I was being rewarded with confirmation that this was indeed the case. Wow.
So here I am. I learned a lot from this whole experience. AI can be extremely useful in unearthing correct facts and methods to address complex questions. At the same time one must be aware of its hidden flaw: it learned on trillions of bytes of existing information, but that included conflicting information from well-meaning but outdated sources. Caveat emptor in a big way. If it’s important enough to matter, it’s important enough to research and cross check.
I also learned that the sentence “AI responses may include mistakes” in tiny print at the end of each AI response is there for a very good reason. It’s called “covering your ass.”
