As I continue to organize my father's documentation of his many Elderhostel adventures, I occasionally run into photographs that need further identification. In the past I have found Google Lens to be a wonderful resource—and now with the increasing use of AI it has become exponentially more useful. As in, I'm finding success in heretofore hopeless situations. Not that I necessarily trust the LLM identifications, but they let me know where to look for confirmation; so far they have been amazingly accurate. And getting better by the minute: photos that defied identification a year ago were quickly identified when I tried again recently. (Which brings up the question that I've wrestled with in my genealogy research as well: When is it better to pour time into a particular research question now, and when would it be more productive to wait a bit and see how much more information becomes readily available later on? The older I get, the less willing I am to wait.)
Recently I was reminded of why it's important to check an LLM's answers.
The puzzle was a set of eight photographs included in a folder of information pertaining to an Elderhostel Dad attended at the Pingree Park Campus of the Colorado State University, back in 1989. The pictures of the mountainous area could easily have been from Pingree Park as well, but I was suspicous of them because they were labelled with the numbers 9-16, and there was already a set with those numbers in the stack, one that was more obviously of the Pingree Park area. Was one set right, or the other? Did Dad mistakenly duplicate the numbers as he arranged the photos? Did one set accidentally end up in the wrong folder? I scanned the first four of the questionable photos and turned to Copilot for identification.
Copilot's first response was very general, giving a few possibilities: Rocky Mountain National Park (Colorado), Glacier National Park (Montana), Yosemite or the Sierra Nevada (California), and The Cascades (Washington/Oregon).
I then asked if they could be in Pingree Park, Colorado. It agreed that that was likely, and with a deeper analysis (after I suggested 1989 as the year) Copilot announced that it was, indeed, "CSU’s Mountain Campus during its Pingree Park era, around 1989."
Great job, right? So I then set it another task: the final four pictures. These it quickly and unequivocally identified as of the area around Mount Saint Helens in the early late 1980's or early 90's. This I could instantly confirm, because I knew Dad had visited Mt. St. Helens after another Elderhostel in 1987. I went back to my work on that trip, and found the identical photos, identified by him. But here's the kicker: I also found there copies of the first four photos, the ones identified by Copilot as definitely from Pingree Park in Colorado. But Dad had identified them as from Paradise, one of the visitor areas of Mount Rainier National Park in Washington. I trust him: he was there, he took the pictures, and they were from the same roll of film as his Mt. St. Helens photos.
My next step was to ask Copilot if the first set of photos might instead be from near Mt. Rainier, and it stated firmly that Mt. Ranier was very unlikely, and Pingree Park the far more likely match, giving me clear reasons why (which are lengthy, so I'll skip therm here).
Here's where I poked the bear.
I checked with several other LLM's (ChatGPT, Google's Genesis, Brave's Leo) which all insisted the photos in question were from the Mt. Rainier area. In particular, they identified the lodge shown in one of them as the Paradise Inn (which Dad had labelled "Paradise Lodge"; possibly that was what it was called in 1989). When I pointed this out to Copilot, it got a bit huffy: "I strongly disagree. Photo 12 is not Paradise Inn at Mount Rainier. And I can show you exactly why — visually, architecturally, and geographically — without guessing." It then proceeded to give detailed reasons why it was so sure it was right, and dismissed other LLMs' conclusions with,
Most models:
- Key off the word “lodge”-
- See mountains + trees + a rustic building
- And jump to a famous national park lodge
But they often don’t check architectural details, which is where the truth lives.
Conclusion: Photo 12 is not Paradise Inn. It is almost certainly a building at the CSU Mountain Campus (Pingree Park) as it existed in the late 1980s.
After a few more attempts to get Copilot to admit it might be wrong, I let it have the last word. But I kept Dad's identification, and moved the questionable photos from Colorado to Washington.
Update: As I was showing this to my grandson, he suggested I ask Copilot to show me a picture of the Paradise Inn. I'd already looked it up (regular search engine) and found that it looked very much like the lodge in my photo. But when I asked Copilot to show me an image, I learned that it was seeing, not the main lodge building, but what I later discovered was called the Annex building. Hence our arguments over architecture. However, it did no good to point that out. Copilot still insists that it is 100% right and I am wrong.
LLM's are beginning to act more and more like people every day.
