Google Photos' AI Magic Editor won't change pictures of IDs, receipts, faces, or bodies
Plus: Amazon is reportedly training a two-trillion-parameter LLM, and more
AI in brief Google's AI-powered Magic Editor will not work if you try to alter images of ID cards, receipts, human faces, or body parts.
The feature, now available in the Google Photos app on the latest Pixel 8 smartphones, uses generative AI to edit images. Users can do all sorts of things like removing unwanted objects, like people in the background, repositioning the focus of a photo, or changing its lighting.
But it won't touch up everything you might want it to. The software has been designed to avoid editing documents that contain personally identifiable information, such as IDs or receipts, or the faces and body parts of humans, Android Authority reported. If you try to highlight any of these things to try and change in Google Photos, the app will likely show you an error message.
Google refers to its policy detailing what its generative AI technologies should and shouldn't do. Changing IDs, for example, could allow people to impersonate others or create content for deceptive or fraudulent activities, like helping underage teenagers buy alcohol, for example. Whereas altering faces and body parts could be used to harm others, like creating non-consensual deepfakes or cyberbullying.
The safety guardrails on Magic Editor, however, aren't always perfect. "Google already blocks plenty of these edits, but some go through. On an older Google Photos version, we tried editing faces and IDs but could not perform those edits as Google blocked the edits with a generic error message," according to Android Authority.
"We could edit invoices, though. We do not know if Google will be improving the guardrails, too, alongside presenting the specific error messages mentioned above."
Amazon is secretly training a giant large language model Olympus
Amazon is reportedly training a huge large language model codenamed Olympus, with the hopes it could compete with rivals OpenAI, Google, and more.
Olympus will reportedly contain a whopping two trillion parameters, making it one of the largest systems to date, Reuters reported. Amazon hopes the model will attract new customers to AWS, and woo existing ones away from using rival large language model Anthropic or Meta it is already hosting on its cloud computing platform.
The effort to develop the model is being led by Rohit Prasad, head scientist of AGI at Amazon and the former lead for Alexa.
Like most Big Tech companies, Amazon is investing heavily in AI and said in a previous earnings call that it would cut back resources in other areas like fulfilment and transportation for retail but step up for large language models and generative AI.
- Google introduces phone-shaped housing for its AI tech
- Desperately seeking GPUs? AWS will let you reserve instances in advance – no refunds
- Judge bins AI copyright lawsuit against DeviantArt, Midjourney – Stability still in the mix
Intel leads latest round of funding for Stability AI
Intel is leading a near-$50 million round in Stability AI, makers of the popular text-to-image tool, Stable Diffusion.
The funds are a convertible note, a loan which Stability will have to pay back in the form of future equity for lenders, Bloomberg first reported. The investment comes after several senior employees, including its head of strategy and human resources, and members of its board left the company.
Intel and Stability already have a close working relationship. In September, the chipmaker said the AI startup would be the first customer using its AI supercomputer, a system made up of Xeon processors and 4,000 Gaudi2 AI accelerators.
At the height of the AI image frenzy, Stability was valued at $1 billion following a $101 million seed round, and reportedly wanted to raise even more money at a higher valuation but has struggled to attract investors. Despite the initial buzz, however, the British upstart has found it difficult to monetize its AI products as it faces more and more competition amid rising operational costs.
There are numerous visual generative AI tools on the market right now, including Midjourney, OpenAI's DALL-E, Adobe's Firefly, and more. ®