OpenAI developing AI image detection tool with claimed 99% accuracy
OpenAI is at work developing a tool that can spot images generated by artificial intelligence with incredible accuracy. Mira Murati, the company's Chief Technology Officer, shared that this tool boasts a 99% success rate in figuring out if an image was made using AI. Although there's no word on when it'll be released, OpenAI is testing the tool internally. The company, famous for its chatbot ChatGPT and image generator DALL-E, is on a mission to improve detection of AI-created content.
Addressing the growing need for detection tools
As AI tools get better at faking content, the need for trustworthy detection tools is skyrocketing. There are already some tools out there that claim to detect AI-generated images and content, but their accuracy can be hit or miss. OpenAI launched a text detection tool in January, but it was pulled in July because it just wasn't reliable enough. The company is dedicated to refining its software and coming up with ways to identify AI-generated audio and images.
Hints at future AI model: GPT-5
At the Wall Street Journal's Tech Live conference, OpenAI executives dropped hints about a possible follow-up to the GPT-4 model. While there's no official word yet, the company did file an application for a "GPT-5" trademark with the US Patent and Trademark Office back in July. When asked if the GPT-5 model would put an end to false info or "hallucinations," Murati cautiously said "maybe," admitting that GPT-4 has made progress but there's still more work to do.
OpenAI considers designing its own computer chips
OpenAI CEO Sam Altman brought up the idea of the company creating and manufacturing its own computer chips for training and running AI models. Right now, OpenAI depends on chips from market leader NVIDIA. Altman mentioned that they'd prefer not to make their own chips, but he didn't rule out the possibility, suggesting that OpenAI might look into other options down the road.