Over the last few weeks ChatGPT’s native image generation abilities had been taking over the internet that allowed the chatbot to create Studio Ghibli style images among other effects from real life pictures of people. However, OpenAI recently rolled out its latest o3 and o4 Mini language model which are now being used to power another trend on social media which may not be as harmless.
What is the new ChatGPT update?
OpenAI explained in its launch blog that o3 and o4-mini are visual reasoning models which means that they can think with images in their chain of thought allowing them to crop, zoom in, rote and use other image processing techniques to analyze the data on the picture submitted by the user.
Much like GPT-4o, the new visual reasoning capabilties of o3 and o4-mini are built natively, meaning they do not rely on an external model like DALL-E 3 to process the pictures.
What are users doing with ChatGPT’s new abilities?
Social media users have started using the new abilities of OpenAI’s new model especially top of the line o3 model in order to identify the geographical location of that place. A number of users on X (formerly Twitter) and Reddit have shared the remarkable accuracy shown by ChatGPT in order to identify the exact location of an image, some even going to the extent of getting an exact geometrical location.
We didn’t have access to OpenAI o3 given that it’s a paid model but we tried running a few images via o4 mini and the results were shocking to say the least. ChatGPT thought for a few seconds and zommed and cropped the image in various direction after giving the correct answer in most occassions. However, we weren’t able to get an exact coordinate despite various attempts.
While a game geoguessr based on ChatGPT’s new model may seem innocent fun at this point, that may not be the case moving forward. It is easy to see this new tool being used to find the locations from pictures from random individuals, which could put these people in harm’s way and risking their privacy.
OpenAI risking safety to rollout new models faster?
Ever since the first public rollout of ChatGPT in late 2022 there have been concerns about AI being used for nefarious purposes and many of them are now coming to fruition recently. Although geoguessing the location of images in itself is very alarming, the chatbot was earlier also used to generated fake Aadhaar and PAN cards among other identification marks.
While ChatGPT has been known to block such requests in the past, amid rising competition from Chinese AI companies, OpenAI has been rolling out one AI offering after another and this may be coming at the cost of safety.
A recent Financial Times report had revealed that OpenAI has significantly cut the time and resources it spends on testing the safety of its AI models and the new models are being rushed to the market without ‘sufficient safeguards’.
Reportedly, staff and third party groups were given just days to assess the risks and performance of the new models compared to the several month timeframe for previous models.
One OpenAI staffer told the publication that LLMs (the foundation models running ChatGPT) have become more capable of ‘potential weaponization’.
“But because there is more demand for it, they want it out faster. I hope it is not a catastrophic mis-step, but it is reckless. This is a recipe for disaster.” FT report quoted the OpenAI staffer as saying.
