AI This Week

OpenAI has introduced two new AI models, o3 and o4-mini, capable of "thinking with images." These models can analyze and discuss uploaded sketches, diagrams, and whiteboards, even if low quality. The o3 model excels at math, coding, science, and image understanding, while o4-mini offers faster, lower-cost operation. Both models integrate visual information into reasoning chains and can use various ChatGPT tools independently. OpenAI aims to maintain its lead in generative AI amid fierce competition. The models are available to ChatGPT Plus, Pro, and Team customers.

Google and research partners have announced a dramatic development in interspecies communication with the unveiling of DolphinGemma, an AI model designed to understand and generate dolphin vocalizations. On National Dolphin Day, Google highlighted its collaboration with Georgia Tech and the Wild Dolphin Project (WDP). WDP's extensive, decades-long dataset of dolphin sounds fueled the AI's training. DolphinGemma identifies, predicts, and even mimics dolphin sounds, paving the way for potential two-way interactions. This initiative not only showcases cutting-edge AI but also promises to deepen our connection with marine life.

Netflix is experimenting with an AI-powered search feature in Australia and New Zealand. The tool, developed with OpenAI, allows users to find content using natural language queries based on moods, themes, or descriptions. Currently available only on iOS devices in select regions, this feature aims to enhance content discovery beyond traditional title or genre-based searches. If successful, Netflix may expand the rollout to more markets, potentially including the United States. This move demonstrates Netflix's continued interest in leveraging AI to improve user experience and engagement.

A national survey by Jobs for the Future (JFF) highlights a sharp increase in AI usage at work, yet training lags behind. Over one-third of U.S. workers now use AI, a significant jump from 2023's 8%. Despite this, only 31% receive employer-provided AI training, with many resorting to self-directed learning. The survey also finds that 59% of learners utilize AI weekly. This discrepancy reveals an urgent need for enhanced training to bridge the gap. Crucially, diverse and underrepresented groups feel the impact more acutely, influencing their career plans. These findings stress the importance of accessible AI education and resources.

Aardvark Weather, a groundbreaking AI approach, promises to overhaul weather forecasting by drastically reducing computational demands and speeding up predictions. Researchers from the University of Cambridge, the Alan Turing Institute, and others have developed a system requiring thousands of times less computing power than conventional methods. This AI-driven model, published in Nature, processes raw global data from various sources to deliver more detailed and quicker forecasts. Richard Turner of the University of Cambridge highlights its potential to offer tailored predictions for specific industries, while Dr. Scott Hosking emphasizes its role in democratizing access for developing nations. Furthermore, Dr. Anna Allen suggests this innovation could enhance forecasting natural disasters.

A recent report from FTI Consulting and Relativity reveals a striking shift in the legal industry, with 44% of legal departments now utilizing generative AI, a significant rise from previous years. Insights from interviews and surveys highlight that 65% of legal departments intend to invest in new technology, with a focus on generative AI. The report uncovers that while enthusiasm for AI grows, 85% of general counsel feel unprepared for its risks. Interestingly, Latin American legal teams show the highest comfort with AI in e-discovery at 86%. Concerns about AI’s viability, reliability, and cost still persist.

Amazon has introduced a new AI-powered feature called 'Interests' that aims to revolutionize personalized shopping. The article, published on Amazon's website, explains how this feature allows customers to create customized shopping prompts using natural language. The AI then continuously scans Amazon's inventory to find matching products, notifying users about new items, restocks, and deals. Currently available to a limited number of US customers, 'Interests' utilizes large language models to translate user prompts into relevant product recommendations

Mount Sinai Health System has launched a new Center for Artificial Intelligence in Children's Health. The initiative aims to develop, test, and integrate AI technologies into pediatric healthcare, focusing on improving diagnostics, personalizing treatments, and optimizing care delivery for young patients. Led by Dr. Benjamin Glicksberg, the center will create an integrated data infrastructure, advance multimodal AI research, and conduct clinical trials at Kravis Children's Hospital. Mount Sinai leadership emphasizes the center's potential to transform children's health outcomes for generations to come.

Perplexity.ai has outlined a bold strategy to reshape TikTok's future in the U.S. The platform plans to rebuild TikTok's algorithm, making it transparent and managed within American data centers to adhere to privacy standards. Leveraging advanced AI infrastructure from partners like Nvidia, Perplexity aims to enhance TikTok’s recommendation systems. The integration of citation features and community notes promises to elevate content trustworthiness. By merging its powerful search engine with TikTok’s expansive video library, Perplexity envisions creating a superior search experience. This initiative promises to transform TikTok into a more informative, engaging platform that respects user privacy and promotes trustworthy content.

Adobe has released new data revealing significant traffic gains for retailers through AI search. The data showcases a 1,300% surge in referrals during the 2024 holiday season. Consumers referred by AI search engage more deeply with websites, indicating more relevant content delivery compared to traditional search methods. AI search tools continue to attract users, suggesting a shift away from traditional links-driven searching.