I guess this is one of those things where you actually expect an AI model to work really well (in their current embodiment). As highlighted by other comments, it takes a lot of pattern matching to make a good guess at this game.
Watching o3 guess a photo's location is surreal, dystopian and entertaining
411–420 of 453 posts
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#412I'm embarrassed that I live in HMB and didn't recognize it. I should go to the harbor more.
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#413Earlier quoted context omitted.
I live in Belltown, Seattle. Oh no! The world knows my neighborhood!
I'm not particularly fond of the whole "privilege" discourse, but this comment is a great example of somebody completely failing to understand a privilege they have. Which they share with many other people, sure,[0] but there are many people who, through no fault of their own, do need to worry about others learning about their location. [0] Which is probably one reason why the discourse grates some. Privilege still s…
It’s a fair point, and we should be sensitive to ill effects on the less privileged, but at some point it’s unreasonable to decry tech than is neutral or beneficial to the vast majority of people because it could possibly harm a small number of people.
The majority of people clutching pearls over the “privacy” implications of photo geolocation are at least as privileged as me, and have even less concept of what e.g. stalking victims go through than I do.
It’s just “think of the children” all over again; “I am uncomfortable with random people knowing my general vicinity” sounds weird, and “I am deeply concerned for the vulnerable people this could harm” sounds noble.
The reality is that those with serious privacy concerns aren’t posting random photos to the internet. I mean, jesus, there’s EXIF data. Hand wringing over AI is entirely performative here.
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#414Earlier quoted context omitted.
I find this type of problem is what current AI is best at: where the actual logic isn't very hard, but it requires pulling together and assimilating a huge amount of fuzzy, known information from various sources They are, after all, information-digesters
“best where the actual logic isn’t very hard”? yeah, well it’s also one of the top scorers on the Math olympiads
LLMs struggle with context windows, so as long as the problem can be solved in their small windows, they do great.
Humans neural networks are constantly being retrained, so their effective context window is huge. The LLM may be better at a complex, well specified 200 line python program, but the human brain is better at the 1M line real-world application. It takes some study though.
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#415Earlier quoted context omitted.
I don’t know if most software engineers build toy CRUD apps all day? I have found the state of the art models to be almost completely useless in a real large codebase. Tried Claude and Gemini latest since the company provides them but they couldn’t even write tests that pass after over a day of trying
Agreed in general, the models are getting pretty good at dumping out new code, but for maintaining or augmenting existing code produces pretty bad results, except for short local autocomplete. BUT it's noteworthy that how much context the models get makes a huge difference. Feeding in a lot of the existing code in the input improves the results significantly.
This might be an argument in favor of a microservices architecture with the code split across many repos rather than a monolithic application with all the code in a single repo. It's not that microservices are necessarily technically better but they could allow you to get more leverage out of LLMs due to context window limitations.
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#416Earlier quoted context omitted.
Agreed in general, the models are getting pretty good at dumping out new code, but for maintaining or augmenting existing code produces pretty bad results, except for short local autocomplete. BUT it's noteworthy that how much context the models get makes a huge difference. Feeding in a lot of the existing code in the input improves the results significantly.
This might be an argument in favor of a microservices architecture with the code split across many repos rather than a monolithic application with all the code in a single repo. It's not that microservices are necessarily technically better but they could allow you to get more leverage out of LLMs due to context window limitations.
if your microservices become more verbose overall, now you have handicapped your ability to cram the whole codebase into a context window.
I think AI is great but humans know the why's of the code needs to exist AI's don't need stuff, only generate it
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#417I play competitive Geoguessr at a fairly high level, and I wanted to test this out to see how it compares. It's astonishingly good. It will use information it knows about you to arrive at the answer - it gave me the exact trailhead of a photo I took locally, and when I asked it how, it mentioned that it knows I live nearby. However, I've given it vacation photos from ages ago, and not only in tourist destinations eit…
I was absolutely gobsmacked by the three minute chain of reasoning this thing did, and how it absolutely nailed the location of the photo based on plants, the color of a fence, comparison with nearby photos, and oh yeah, also the EXIF data containing the exact lat/long coordinates that I accidentally left in the file. https://bsky.app/profile/matthewdgreen.bsky.social/post/3lnq...
This is super easy to test though (whether EXIF is being used). Open up Geoguessr app, take a screenshot, paste into O3. Doing this, O3 took too long (for the guessing period) but nailed 3 of 3 locations to within a kilometer.
Edit: An interesting nuance of modern OpenAI chat interface is the "access to all previous chats" element. When I attempted to test O4-mini using the same image -- I inspected the reasoning and spotted: "At first glance, the image looks like Ghana. Given the previous successful guess of Accra Ghana, let's start in that region".
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#418Earlier quoted context omitted.
Agreed in general, the models are getting pretty good at dumping out new code, but for maintaining or augmenting existing code produces pretty bad results, except for short local autocomplete. BUT it's noteworthy that how much context the models get makes a huge difference. Feeding in a lot of the existing code in the input improves the results significantly.
This might be an argument in favor of a microservices architecture with the code split across many repos rather than a monolithic application with all the code in a single repo. It's not that microservices are necessarily technically better but they could allow you to get more leverage out of LLMs due to context window limitations.
this is a short term issue though. The available context window has been increasing exponentially over the past 2 years
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#419Tried the same, results made me laugh. Completely clueless. I've seen passing prompts 8 about how it's not in the city I am and yet it tries again and again. My favourite moment was when it started analysing piece of blurry asphalt. After 6 minutes o3 it was confidently wrong: https://imgur.com/a/jYr1fz1 IMO not-in-US is actually great test if something was in LLMs data and the whole search is a for show.
Huh, I've been very impressed. I've given it photos I took in a Nairobi slum, a random non-iconic street in Bath, a closeup of a road in Tuscany, and a small playground in Jakarta, and it got them all perfectly.
Re: Watching o3 guess a photo's location is surreal, dystopian and entertaining
#420Earlier quoted context omitted.
You don't need a LLM to do this. A dedicated image->coords model would likely perform much better, and that's old school ML at this point.
Have you seen a description of one of those? I didn't know that those existed.
It would obviously be classified CIA/NSA technology.