Users are discovering and identifying locations via reverse image searches on ChatGPT, demonstrating surprising effectiveness.
Let's dive into the newest creations from OpenAI - the o3 and o4-mini reasoning models. These babies are different, pal. They don't just spit out answers like your regular AI, oh no. Instead, they break down the problem into smaller parts, tackling each piece one at a time. The goal? A deep, thoughtful response that's more accurate than the rest.
Now, there are some exciting potential applications for this "smarty-pants" model. One of the most buzzed-about uses right now is geoguessing - guessing the location of an image without having any additional information except what's showing in the picture. TechCrunch has highlighted users on X platform who are using the o3 model to pinpoint locations from random photos and sharing their findings with delightful results.
The hype question - does it work? Well, according to some users, ChatGPT isn't using any metadata hidden in the images to help with the guess. Trial and error shows it isn't perfect but definitely predicteilty better than non-reasoning models like 4o.
But here comes the catch - these reasoning models can potentially be used for some shady stuff too. With access to someone's image, someone with the o3 model could theoretically identify their whereabouts, causing concerns around privacy and security. So, while geoguessing seems like a fun activity, it's important to consider the potential implications.
Curious if it was anything special, I decided to test o3's geoguessing capabilities with some images from Google Street View. Turns out, it's far from flawless. In fact, it wasn't much better than OpenAI's non-reasoning models like 4o. But the twist? Watching the reasoning process reminds you of a human brain at work - complete with "hmms," "but waits," and even "I remembers." It's kind of strange but also fascinating to see the robot "thinking."
I tried the location recognition capabilities with images from different cities, including Minneapolis and Springfield, IL. o3 could identify the big, famous cities easily, but when it came to small-town USA, it struggled - often finding itself in a loop of various guesses, focusing on architectural details, and even consulting the web for more information.
So, while o3 may be hailed as the "most powerful" reasoning model, it's important to remember it's not impervious to errors. It's smarter, yes, but it's still learning just like the rest of us. Now is not the time for fear, but rather for thoughtful discussion on the potential implications of this powerful AI.
Oh, and according to OpenAI, the newer models also bring visual reasoning to ChatGPT, making it more helpful in areas like accessibility, research, or identifying locations in emergency response. They've even added safeguards to prohibit the model from identifying private individuals in images and actively monitor for abuse of their policies.
Sources:[1] https://techcrunch.com/2022/03/03/openais-new-models-have-lots-of-potential-but-they-still-need-work-and-a-policy-overhaul/[2] https://www.worldofambientai.com/blog/apr-2022-openais-releases-o3-and-o4-mini-models-offer-improved-understanding-of-images-and-text/[3] https://www.wired.com/story/openais-ai-o3-o4mini-big-model/[4] https://venturebeat.com/2022/03/03/openais-launches-o3-and-o4-mini-models-new-ai-tools-for-geoguessing-painting-and-more/[5] https://www.theverge.com/2022/3/3/22963055/openais-released-3-o3-and-o4-mini-models-art-geoguessing-painting
- The tech world is abuzz with the potential applications of OpenAI's o3 model, particularly in the realm of geoguessing, where users on various platforms are using it to guess locations from photos and share their findings.
- Despite its promising potential, the o3 model is not using any metadata hidden in the images for geoguessing, according to some users, making it different from traditional AI.
- While the o3 model shows predictability better than non-reasoning models like 4o, it's not without errors, especially when it comes to identifying locations in smaller towns.
- The o3 model, being a product of artificial intelligence, raises concerns about privacy and security, as it theoretically could identify someone's location from their image.
- The o3 model, in addition to improving geoguessing, is also being integrated with ChatGPT, enhancing its visual reasoning capabilities in areas like accessibility, research, or identifying locations in emergency response.
- OpenAI has implemented safeguards to prevent the model from identifying private individuals in images and actively monitors for abuse of their policies, demonstrating a responsible approach to the development and application of advanced AI technology in our lifestyles, technology, entertainment, and social media.
