Trending AI news stories + paper
Today's Top AI news stories and papers
AK’s Substack is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
AI news stories
Google's Lumiere, an advanced AI video generation model, uses a novel Space-Time-U-Net (STUNet) for creating realistic videos in a single process, surpassing competitors with its ability to generate 80 frames for smoother motion. The model notably outperforms other platforms in realism, particularly in movement portrayal, and expands capabilities to include various forms of video manipulation like image-to-video generation and cinemagraphs. However, Google acknowledges potential ethical concerns, emphasizing the need for tools to prevent misuse and detect biases in AI-generated content.
Social media platform X has blocked searches for Taylor Swift following the circulation of explicit AI-generated images of the singer on the site, leading to a temporary measure to ensure safety. The White House and US officials have expressed concern over the spread of these AI-generated photos, highlighting the need for stricter enforcement and potential legislation against such misuse of AI technology. In response, X has enforced a zero-tolerance policy, actively removing the images and taking action against the accounts responsible, with no clear indication of when the search block will be lifted or if similar measures have been applied to other public figures.
Instacart is utilizing generative AI to produce images for its AI-written recipes, resulting in photos with odd features that make the food appear unappetizing. The images often have unnatural compositions and blending of elements, like a hot dog with a tomato's texture or a lemon-lettuce hybrid, and have been confirmed by AI detection software as likely AI-generated. This move by Instacart has raised questions about the reliability and appropriateness of using AI-generated imagery in consumer-facing applications.
The University of Texas at Austin is launching the Center for Generative AI, featuring one of the largest GPU computing clusters in academia, with 600 NVIDIA H100s GPUs. This center will focus on advancing AI research in various fields including biosciences, health care, computer vision, and natural language processing, and will foster collaborations with external partners. The Center for Generative AI is part of UT's broader commitment to AI, which includes educational programs, significant research investments, and interdisciplinary efforts to develop ethical AI technologies.
Learnable.ai, a Ningbo-based startup, is aiming to raise up to $70 million (about 500 million yuan) to develop AI solutions for the grueling Chinese college entrance exam, the "gaokao." The company, led by Chairman Guan Wang, is in discussions with investors to fund the project, which seeks to automate the grading process of this nationwide test. This initiative is part of a growing trend where startups are leveraging AI to address educational challenges and improve efficiency in academic settings.