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
Meta is set to deploy a new custom AI chip, "Artemis", in its data centers in 2024, aiming to enhance its generative AI capabilities in various applications and devices while reducing dependency on Nvidia's chips. Reuters highlights that Artemis, which focuses on the inference aspect of AI, will complement the substantial number of Nvidia H100 chips Meta has acquired. Although similar to the previously announced MTIA chip and primarily aimed at AI decision-making processes rather than training, Artemis represents a strategic move by Meta to optimize its AI infrastructure, despite its somewhat delayed entry into the market.
Google has enhanced its Bard chatbot by integrating AI image generation capabilities using Imagen 2, aligning it with the features of its competitor, ChatGPT Plus, which uses DALL-E 3. This advancement comes as part of Google's effort to position Bard, powered by the Gemini Pro large language model, as a strong competitor in the multimodal chatbot arena. In addition to image generation, Bard has broadened its linguistic reach, now supporting over 40 languages, and improved its double-check feature for better response evaluation. Notably, Bard's image generation, unlike the subscription-based ChatGPT Plus, is available at no cost and comes with robust safety features to prevent the creation of inappropriate or explicit images, reflecting Google's commitment to responsible AI deployment.
Amazon introduced Rufus, an AI-based shopping assistant designed to enhance product search and discovery for users by utilizing Amazon's product catalog, customer reviews, Q&As, and web information. CEO Andy Jassy plans to integrate generative AI throughout Amazon's businesses, with Rufus being a part of this wider AI strategy. This announcement precedes Amazon's report of its fourth-quarter earnings, highlighting the company's ongoing investment in generative AI tools and services.
Google Maps is integrating a generative AI feature to enhance discovery, utilizing large language models to suggest new places based on over 250 million locations and contributions from 300 million Local Guides. This new feature is more conversational, offering tailored suggestions for searches like thrift shops or restaurants, and can handle niche queries. Initially available to select U.S. Local Guides, with plans for broader rollout, this update is part of Google's strategy to evolve Maps from a navigation tool into a platform for location discovery.
Researchers at New York University conducted a study to determine if an AI could learn language in a manner similar to a human baby. They utilized a unique dataset obtained from a helmet camera worn by a child, Sam, which captured his daily interactions and environment. The AI model was trained with over 600,000 video frames and their corresponding spoken phrases, effectively learning to match words to objects, a crucial component of language learning. This approach, with its limited but rich dataset, challenges the prevailing notion that an innate understanding of language is essential for rapid language acquisition. The study not only offers insights into child language development but also paves the way for more advanced, human-like AI language models.