AI News – May 29, 2023

Security Concerns Over ChatGPT Prompt Injection Attack

OpenAI’s language model, ChatGPT, has been found to be vulnerable to a type of attack known as “prompt injection”. This attack allows malicious actors to manipulate the model’s output by injecting specific prompts into the conversation. OpenAI has acknowledged the issue and is working on a mitigation strategy. The company has also urged users to report any instances of this attack. Despite the security concern, OpenAI maintains that the benefits of AI and machine learning outweigh the risks, and is committed to improving the safety and security of its models. ¹

AI Tool Generates Video from Brain Activity

Researchers at the National University of Singapore and the Chinese University of Hong Kong have developed an AI tool that can generate videos from brain activity. The tool uses functional magnetic resonance imaging (fMRI) data and a deep learning model to create high-quality videos. This technology could potentially be used in neuroscience and brain-computer interfaces. The researchers have achieved an accuracy rate of 85% in their video reconstructions, indicating promising potential for future applications. ²

AI Tools Revolutionize Photo Editing

New AI tools, such as DragGAN and Stability AI, are transforming the field of photo editing. These tools use generative artificial intelligence to manipulate photos in ways that were previously impossible without extensive editing experience. For instance, DragGAN allows users to alter images radically, from changing facial expressions to modifying the layout of a landscape. Stability AI’s upscaler can quadruple the resolution of an image without compromising its sharpness. These advancements are rendering traditional editing tools like Photoshop obsolete. ³

Adobe’s Photoshop Unveils Generative AI Tool

Adobe has launched a new AI tool called Generative Fill that uses generative AI to add or remove objects from photos based on a simple text prompt. This tool is part of Adobe’s next-generation features, collectively dubbed Firefly, which will allow creators to generate images, audio, video, illustrations, and 3D models with simple text inputs. Generative Fill is currently available for users of the beta version of Photoshop, with a wider release expected to follow. ⁴

Opera Launches AI Sidebar Powered by OpenAI’s ChatGPT

Opera has introduced an AI sidebar in its browser called “Aria”, powered by OpenAI’s ChatGPT. Aria is designed to make it easier for users to find information on the web, generate text or code, or get product queries answered. The AI sidebar is currently available for testing and is expected to become more integrated into Opera in future versions of the browser. ⁵

TikTok Tests In-App AI Chatbot ‘Tako’

TikTok is testing an in-app AI chatbot called “Tako”. The bot appears on the right-hand side of the TikTok interface and can answer various questions about a video or suggest new content based on user queries. The bot is currently in limited testing in select markets, including an early limited test inthe Philippines. TikTok is leveraging a third-party AI provider that has been customized for its needs, and allows users to delete their chats with Tako. ⁶

OpenAI’s ChatGPT iOS App Expands to More Countries

OpenAI has expanded the availability of its ChatGPT app for iOS users to India and 32 other countries, including Canada, Japan, and Mexico. The app allows users to interact with the generative AI-based chatbot using their iPhone. It also supports voice input through OpenAI’s speech recognition system Whisper and lets ChatGPT Plus users access advanced features. The app is currently only available for iOS, but an Android version is in the works. ⁷

Sources