OpenAI and Apple Partner Up, Zoom Wants AI Clones in Meetings, OpenAI's Sora Soars at Tribeca ... and more

Welcome to AI Weekly Breakthroughs, a roundup of the news, technologies, and companies changing the way we work and live.

OpenAI and Apple announce partnership to integrate ChatGPT into Apple experiences

Apple is integrating ChatGPT into iOS, iPadOS, and macOS, allowing seamless access to its capabilities, including image and document understanding, directly within these systems. Siri can utilize ChatGPT’s intelligence to provide answers, with user consent required before any data is sent. ChatGPT will also enhance Apple’s Writing Tools, aiding content creation and image generation in various styles. Privacy protections ensure requests are not stored by OpenAI and users' IP addresses are hidden. The integration, using GPT-4o, will be available later this year for free, with additional features accessible to ChatGPT subscribers who connect their accounts.

Apple Unveils Apple Intelligence at WWDC 2024

At WWDC 2024, Apple unveiled Apple Intelligence, a generative AI system integrated across its ecosystem, including iOS, macOS, and VisionOS. Emphasizing safety and personalization, Apple Intelligence is designed to understand users' routines and relationships while maintaining privacy, processing much of the data locally on Apple silicon. The AI also enhances Siri, now allowing typed queries and deeper integration into apps through App Intents. Features include Smart Replies in Mail, Genmoji for custom emojis, and Image Playground for on-device image generation. Apple Intelligence will be available on iPhone 15 Pro, M1 Mac, and iPad devices with the latest OS updates, rolling out later this year alongside a partnership with OpenAI to integrate ChatGPT capabilities.

Apple's macOS Sequoia Features Enhanced AI Capabilities

Apple previewed macOS Sequoia, the latest version of its desktop operating system, introducing transformative features and enhanced intelligence capabilities. Key updates include iPhone mirroring for seamless continuity, Safari's Highlights feature for easy information discovery, and a new Passwords app for better credential management. Gaming also sees improvements with more immersive experiences and new titles like Assassin’s Creed Shadows and Frostpunk 2. MacOS Sequoia will feature Apple Intelligence (see above).

Zoom CEO Wants AI Clones in Meetings

Zoom CEO Eric Yuan discusses plans for AI-driven "digital twins" to attend meetings on behalf of users, highlighting Zoom's evolution from a video conferencing service to a comprehensive AI-integrated workplace platform. This initiative is part of Zoom's broader strategy to enhance user efficiency and redefine professional communication.

AMD Asserts Dominance with Ryzen AI 300 Series

AMD launches the Ryzen AI 300 series, claiming industry-leading AI capabilities with 50 TOPS from its NPU. The new chips, aimed at powering thin, light Copilot+ laptops, mark AMD’s entry into the Zen 5 processor family, featuring significant CPU and GPU advancements. With partners like Acer and HP, AMD sets a high benchmark for AI performance in PCs, asserting dominance over competitors like Qualcomm and Intel.

Tribeca 2024 Debuts AI-Generated Films by OpenAI's Sora

AI-generated short films are gaining traction at film festivals, highlighted by the 2024 Tribeca Festival’s announcement of Sora Shorts, a new program featuring five original films created using OpenAI’s text-to-video model, Sora. This marks the first showcase of Sora-generated films at a major festival. The festival will also include a panel discussion with these directors. While Sora promises advancements like 60-second videos and interactive elements, it lacks audio capabilities and has sparked controversy, including concerns about job losses in traditional filmmaking sectors.

Sirion Acquires Eigen Technologies to Boost AI-Driven Contract Intelligence

Sirion has announced the acquisition of Eigen Technologies, enhancing its capabilities in the Contract Lifecycle Management (CLM) market through advanced document intelligence. Eigen’s AI-driven platform specializes in extracting data from diverse document types, expanding Sirion's reach beyond contracts to include financial and operational documents. This integration forms a substantial part of Sirion’s strategy to dominate the AI-native CLM space, positioning it at the forefront of leveraging AI for contract intelligence on a global scale.

Unitree Robotics Unveils $16k G1 Humanoid

Unitree Robotics has introduced its new G1 humanoid robot at ICRA 2024, marking a significant evolution from its predecessor, the H1. The G1 is smaller, standing at 127 cm, compared to the adult-sized H1, and is priced significantly lower at $16,000 versus the H1's $90,000. Despite its reduced size, the G1 maintains a similar weight due to optimized design efficiencies such as fewer wires and custom-built hardware components. The G1 targets research applications, especially in university labs, given its affordability, nimble design, and advanced features like articulating hands, 3D LiDAR, and Intel RealSense cameras. While it lacks the power and reach to replace human labor in practical settings, it serves as an excellent platform for developing AI algorithms.

Kling AI Emerges as Strong Rival to OpenAI's Sora

Kling AI, a new AI video generator developed by the Chinese company Kuaishou, is emerging as a strong rival to OpenAI's Sora. Currently available in China through a waitlist, Kling AI has quickly gone viral with its ability to create 1080p videos up to two minutes long. Early demo videos showcase impressive results, although they exhibit typical AI-generated artifacts like smoothing. While these early clips are promising, their true capabilities are yet to be fully assessed due to potential post-production enhancements. As OpenAI plans to publicly release Sora later this year, Kling AI's broader availability will be crucial if it aims to dominate the AI video generation market.

Google Explores 'Memory' Feature for ChromeOS

Google is exploring a "memory" feature for ChromeOS, similar to Microsoft's Recall feature in Windows 11. This potential feature aims to enhance user experience by providing context and memory of recent activities on the device, such as rewinding or recording screen content when interrupted. John Solomon, VP of ChromeOS, emphasized the importance of user control to avoid the "creepy" factor associated with automatic, unsolicited features. The concept is still in discussion, focusing on making it useful and user-initiated. The broader interview also touches on the future of Chromebooks with AI integration, including potential benefits from NPUs and the inclusion of Qualcomm's Snapdragon X Elite.

Anthropic's Claude 3 Elevates AI with Character Training

Claude 3, the latest AI model from Anthropic, incorporates "character training" into its development, emphasizing traits like curiosity, open-mindedness, and thoughtfulness. This innovative approach aims to enhance Claude’s interactions by instilling a richer character profile, moving beyond mere harm avoidance to embody qualities found in wise and well-rounded individuals. This character training not only makes Claude more discerning but also helps it navigate complex human interactions more gracefully, acknowledging its biases and leaning towards diverse viewpoints without claiming infallibility.

Stability AI Launches Stable Audio Open for Creative Sound Design

Stability AI introduces Stable Audio Open, an open-source text-to-audio model that allows users to generate up to 47-second audio samples, including drum beats, instrument riffs, and ambient sounds from text prompts. Designed for sound designers and musicians, it also supports style transfer and fine-tuning on custom data. Unlike its commercial counterpart, which can produce full-length, coherent musical tracks, Stable Audio Open focuses on short audio samples and production elements. Trained on datasets from Freesound and the Free Music Archive, this release aims to empower creative communities while respecting creator rights.

OpenAI Unveils Advanced Security Measures for Frontier AI Training

OpenAI outlines its advanced security architecture designed for the secure training of frontier AI models, emphasizing the protection of sensitive model weights and intellectual property. Operating large-scale AI training supercomputers, the organization employs multiple layers of security strategies including enhanced identity foundations, robust Kubernetes architectures, and stringent access management systems to ensure that the infrastructure remains resilient against unauthorized access and threats. This proactive approach supports OpenAI’s mission to ensure that the benefits of advanced AI technologies are safely advanced and accessible for all.

LLMs Turn Every Question Into an Answer: Part 2

In his second installment of a series on reimagining creativity with AI, Dan Shipper explores how LLMs are revolutionizing text interaction by expanding user queries into comprehensive answers. Unlike traditional searches, LLMs like ChatGPT can dynamically generate context-rich responses tailored to user needs, effectively turning every question into a detailed answer. These models enhance creativity by providing personalized expansions that range from simple explanations to complex metaphor generation. Shipper highlights the potential of LLMs to generate unique content, making them invaluable tools for knowledge acquisition and creative ventures.

Ethan Mollick's Mid-Year AI Update

Ethan Mollick’s mid-year guide to AI updates highlights significant advancements and playful applications of AI systems. He notes that the latest AI models, such as GPT-4 and Gemini 1.5, have become more capable and easier to use, offering features like internet connectivity, image and video generation, and advanced data analysis. Mollick emphasizes the importance of exploring AI through playful interactions, such as creating songs or similes, to understand their potential and limitations. He also envisions that upcoming models will be even smarter, opening up new opportunities and risks in the AI landscape.

How Apple Fell Behind in the AI Arms Race

Apple is poised to announce significant generative AI advancements at its upcoming Worldwide Developers Conference, aiming to catch up with rivals like Microsoft and Google. Key updates include enhancements to Siri, featuring capabilities in message writing, photo editing, and text summarizing. Siri, originally reimagined under Project Blackbird but later upgraded more modestly as Siri X, will now see further improvements focused on speed, privacy, and integration with third-party AI providers. This shift comes as Apple faces pressure to innovate in the AI space, having taken a cautious approach that emphasized user privacy and seamless hardware-software integration. Recent hires, like John Giannandrea from Google, have been pivotal in centralizing and advancing Apple's AI efforts.

Extracting Concepts from GPT-4

Researchers have unveiled advanced, scalable methods to decode the complex neural activity within GPT-4, revealing 16 million often-interpretable patterns, or "features." Using sparse autoencoders, the team has been able to identify and visualize these features, aiming to enhance AI model interpretability. This breakthrough could aid in monitoring and steering AI behaviors, while also addressing challenges like spurious activations and limited capture of model behavior. The research, which includes a detailed paper, code, and feature visualizations, is now open-sourced to foster further exploration and understanding within the AI community.

AI-Generated Future Self Chat Reduces Anxiety and Boosts Wellbeing

Researchers introduced "Future You," an innovative digital chat intervention designed to improve future self-continuity, which is linked to better mental health. This single-session system lets users interact with an AI-powered version of their future selves, tailored to their goals and personal traits, and supported by a generated "synthetic memory" connecting their current life to their life at age 60. Participants reported reduced anxiety and heightened future self-continuity after the interaction, marking the first successful use of personalized AI characters to enhance users' future outlook and wellbeing.

Cybersecurity startup Seven AI raises $36M led by Greylock.

Greptile raises $4M to build an AI-fueled code base expert.

Tektonic AI raises $10M to build GenAI agents to automate business operations.

Microsoft will invest $3.2 bln in Swedish cloud, AI.

Cohere launches startup program to empower early-stage AI innovation.

Cisco launches a $1B global AI investment fund.

Apple’s Worldwide Developers Conference Cupertino - June 10 - 14

AI Engineer Summit San Francisco - June 25 - 27

World Summit AI Amsterdam - October 9 - 10

Gitex Global - Dubai - October 14 - 18

Big Data Conference Europe Vilnius - November 19 - 22