Curse you Mike Caulfield for being so damn insightful and hello after a long time. I am tempted to still feel this is classic Benderian parroting. Crafting the prompt to outline its methods as you do provides a clear reasoning path and leaving it to use to decide as opposed to just asking “Where is this photo?”. We might get region, city, street address? And like the movie mentioned in the comments it’s still fraught. Unless we know ahead of time, as you did, or know the location first hand, it might always be open to interpretation, which of course is always true.
I’m reminded of seeing a landscape photo recently in Mastodon. As a hobby, in movies, paintings of landscapes I often try to guess the location. This photo spoke to me as being in southern Arizona or New Mexico based much on my time spent living there. The types of trees, rocks, sky color? I cants say I can dictate a formal chain but it just happened in seconds (the alt text told me it was Ramah New Mexico).
This flows into the speculation of AI requests for facts vs rendering judgement (is this person a suspect? Is that splotch cancer) as to the machine there is no difference.
Just want to thank you for flipping some ideas on for me and also putting up with one of those old comments that should be a post. Heck it’s an excuse for a long overdue hello.
You articulated something I've been trying to communicate with AI skeptics about the opacity of human reasoning. Love this and need to think more about the testability of arguments (or of form of argument?).
Is there a chance that ChatGPT o1 recognized Yakutsk from its training set and then concocted some bullshit reasoning to make it sound convincing? Because I agree with you—the reasoning seems far-fetched. You either know it or you don’t. There are hundreds of cities in Russia with thousands of the same “khrushchyovka” buildings. In fact, the most famous Soviet New Year movie, The Irony of Fate, is about a guy who gets drunk on New Year’s Eve, boards a plane, and mistakenly ends up in Leningrad instead of staying in his native Moscow. Thanks to the uniformity of Soviet architecture and streets with identical names, he unwittingly enters an apartment identical to his own.
I think the concrete stilts and the vegetation are the most convincing element, and it is overemphasizing the other stuff because I told it to provide a variety of reasons. In general I ask it for multiple reasons knowing that I'll probably exclude many. In my experience it's not always the best at understanding the difference between additive evidence and determinative.
Curse you Mike Caulfield for being so damn insightful and hello after a long time. I am tempted to still feel this is classic Benderian parroting. Crafting the prompt to outline its methods as you do provides a clear reasoning path and leaving it to use to decide as opposed to just asking “Where is this photo?”. We might get region, city, street address? And like the movie mentioned in the comments it’s still fraught. Unless we know ahead of time, as you did, or know the location first hand, it might always be open to interpretation, which of course is always true.
I’m reminded of seeing a landscape photo recently in Mastodon. As a hobby, in movies, paintings of landscapes I often try to guess the location. This photo spoke to me as being in southern Arizona or New Mexico based much on my time spent living there. The types of trees, rocks, sky color? I cants say I can dictate a formal chain but it just happened in seconds (the alt text told me it was Ramah New Mexico).
This flows into the speculation of AI requests for facts vs rendering judgement (is this person a suspect? Is that splotch cancer) as to the machine there is no difference.
Just want to thank you for flipping some ideas on for me and also putting up with one of those old comments that should be a post. Heck it’s an excuse for a long overdue hello.
Thank you. Searching for methods to guide students towards useful applications of AI tools.
You articulated something I've been trying to communicate with AI skeptics about the opacity of human reasoning. Love this and need to think more about the testability of arguments (or of form of argument?).
Is there a chance that ChatGPT o1 recognized Yakutsk from its training set and then concocted some bullshit reasoning to make it sound convincing? Because I agree with you—the reasoning seems far-fetched. You either know it or you don’t. There are hundreds of cities in Russia with thousands of the same “khrushchyovka” buildings. In fact, the most famous Soviet New Year movie, The Irony of Fate, is about a guy who gets drunk on New Year’s Eve, boards a plane, and mistakenly ends up in Leningrad instead of staying in his native Moscow. Thanks to the uniformity of Soviet architecture and streets with identical names, he unwittingly enters an apartment identical to his own.
I think the concrete stilts and the vegetation are the most convincing element, and it is overemphasizing the other stuff because I told it to provide a variety of reasons. In general I ask it for multiple reasons knowing that I'll probably exclude many. In my experience it's not always the best at understanding the difference between additive evidence and determinative.
Also I might need to see that film.