TECH NEWS – OpenAI’s o3 and o4-mini models can be a bit of a privacy nightmare, as they can easily guess where a photo was taken…

OpenAI’s latest artificial intelligence models, released last week, have sparked a new craze as users have begun testing how good they are at geo-guessing. OpenAI’s new o3 and o4-mini models are both capable of image inference. In a broader sense, this means extensive image analysis capabilities. The models can crop and manipulate images, zoom, read text… and if you add the agent’s web search capabilities, you theoretically have a killer, accurate image search tool.

OpenAI says that for the first time, these models can integrate images directly into their thought processes. They’re not just looking at an image, they’re thinking with it. This opens up a new problem-solving class that combines visual and textual thinking. Some books on the shelf – University of Melbourne; cars with left-hand drive but left-hand steering – Suriname… Models are also able to extend all of their reasoning, including the cues they perceive and how they interpret them. However, research published earlier this year suggests that models’ explanations of how they arrive at their answers do not always reflect the actual cognitive processes of artificial intelligence, if they can be called such at all.

The privacy concerns are clear. The posts of a frequent social media user could be enough for an AI model to predict future movements and locations accurately. No wonder TechCrunch asked OpenAI about this, to which the company responded: “OpenAI o3 and o4-mini bring visual reasoning to ChatGPT, making it more helpful in areas like accessibility, research, or identifying locations in emergency response. We’ve worked to train our models to refuse requests for private or sensitive information, added safeguards to prevent the model from identifying private individuals in images, and actively monitor and take action against abuse of our privacy policies.”

This sounds very scary, and it could be an interesting future…

Source: PCGamer, OpenAI, TechCrunch