They will mislead you. Bad. (But not on purpose.)
I recently became enamored with AI models for my searches/inquiries, and have all but replaced Google as my primary search engine. It’s great for anything you can ever imagine asking unless you need something like a website or to pull up something quickly. Those times are good for a basic Google search, but that is rare for me these days. I’ve made ChatGPT my default search, and have dived in head-first.
And I’m learning a lot about a lot. Anything I can imagine wondering, I ask it. I’ve also learned how to ask questions properly, what to expect based on my inquiry, and also learned that AI will lead you down the wrong path. With a lot of confidence.
Many times, you don’t discover its errors until AFTER it has cost you. That is why I’m writing this. Now, new users may expect unflappable, concrete answers. To both some pretty complex questions and some very easy ones. I’ve been surprised and disappointed, but I realized something.
You must learn to ask correctly. This takes being able to elucidate and articulate your thoughts into cogent words that a computer can understand.
Many people aren’t capable of this. I know because they’re the same people who are afraid to post online. After all, they cannot write. That doesn’t stop some, but others decide not to participate if they aren’t allowed to just talk, then allow the words to vanish into the ether. (Just like attorneys do to manipulate. They put nothing in writing unless given money, and then their assistants write it, they review it, and sign it. Same with judges. They don’t write a word. Young staff attorneys do all the work. But they’ll have expensive meetings, of course, just none ever “on the record.”)
But for the past two weeks, I’ve been doing a lot of work on my truck and restorative work around the house. That meant I needed a lot of specific, technical, and often complex answers and information.
I would ask in several ways, writing in my best manner. I would even question the answers and have learned it often makes glaring mistakes. ChatGPT admits it right on the page. But it’s egregious sometimes and can be very expensive in money and time. At least it’s cost me, which is why I want to warn you. Set proper expectations for pop AI in July 2025. Many, many times it will simply tell you to call an expert. That’s what you’re asking IT about, but hey, thanks for the obvious.
For example, I would ask about parts for my GMC truck. I would receive a comprehensive answer that was very convincing. I would verify the answer, or part number needed, with Amazon, Google and specific car part stores. And many times, the part was wrong. Very wrong.
For example, ChapGPT said I needed a certain transmission fluid drain pan, so I ordered one from Summit Racing. A pretty common item for a very common transmission. A 4L60E. I also ordered a drain plug from another store. The pan was $60 and the plug, which made my skin crawl to order because it’s just a magnetic bolt, was over $15. I waited days for them to arrive and then could do nothing with them. After I gooped up the pan with grease to hold the gasket on, so I also had to clean it just to return it.
They arrived, and I prepared to put the pan on, with the gasket and long story short: it wouldn’t fit. It was the wrong one. Too thick by 1/2 inch. Which meant the plug wouldn’t fit, so I didn’t need either one. I managed ot return the pan, with much struggle, and the place I got the plug from said don’t bother because the shipping charges would be as much as the plug cost. That isn’t true (although close), but they made it clear they didn’t want to deal with it.
Then it told me the oil filter I needed was incorrect. Not just once, but twice. This ultimately means I wasted 3 filters, and had to order ANOTHER from Amazon before getting the right one. It contradicted Amazon and when I asked why, it told me there are variations and Amazon’s search engine is wrong. It wasn’t.
Then it told me I needed a bridge rectifier, which I ordered when I actually needed a new transformer. So I then ordered one of those. See the increasingly expensive pattern? And the transformer’s on backorder, so it’s going to take forever. Even though shipping costs me the same. And the rectifier was “only” $5, so there was no use in returning the unneeded part. I have a growing pile of money/items that were “unneeded.”
This continues through tonight. It tells me that a cheap multimeter I have can check continuity, and it can’t, and on and on. It told me an OBDII scanner I have wouldnt test transmission codes.
I rely on traditional search engines to source items I need in life, usually as inexpensively as possible and delivered fast. (Amazon, as a Prime member, usually is my answer, but not always. I have to find obscure, old parts a lot, which are only available at specialty stores. RARELY local.)
However, that isn’t what AI is designed for. It will list possible vendors, but when you click through, you’ll find they don’t have the item, it’s the wrong item, it’s out of stock, there’s a massive shipping fee, etc… In other words, it is a waste of time. ChatGPT won’t provide discount codes or similar information, either. So, depending on a chat model for shopping is sort of futile. Not entirely, but typically, you aren’t going to find exactly what you’re searching for the first time. Or second. Or ever, maybe. So Google is better for that, I think. I tend to nearly always wind up on Amazon, anyway, but lately, I have had to order a lot of specialty parts.
So I almost bought a THIRD one, for a lot of $, to get one that did. And it turns out the scanners I have WILL read those codes. I discovered that after spending about an hour researching scanners, watching videos and doing research. Until I learned I wasted my time, and almost money. Again.
So, while AI is awesome and getting better, it is often wrong.
Trust but verify. A lot.
