AI This Week

Nvidia's newest chips have significantly improved the efficiency of training large artificial intelligence systems. According to data released by MLCommons, the number of chips required to train large language models has decreased dramatically. Nvidia's Blackwell chips are more than twice as fast as their previous generation Hopper chips on a per-chip basis. The data shows that 2,496 Blackwell chips completed a training test in just 27 minutes, while it took over three times as many previous-generation chips to achieve a faster time.

ABBA’s Björn Ulvaeus is crafting a new musical with the assistance of artificial intelligence. At 80, the legendary songwriter announced during SXSW London that he is nearly three-quarters through the creative process. He describes AI as a tool that extends his creative mind, helping break through tough blocks. The Swedish musician acknowledges the tool’s limits, noting it often produces unusable lyrics yet occasionally offers sparks for new ideas. His journey with cutting-edge technology continued after ABBA’s pioneering avatar show Voyage. Ulvaeus also champions fair compensation for artists as AI systems benefit from their work. The project arises from his ongoing work with Pophouse Entertainment

Yoshua Bengio, one of AI’s foremost innovators, has launched LawZero, a non-profit aimed at addressing the potential dangers posed by rogue AI systems. With an initial $30 million in funding and backing from prominent figures and organizations, Bengio’s initiative seeks to create "Scientist AI," a system designed to monitor and block harmful behavior in autonomous AI agents. Unlike current generative models, this system emphasizes transparency, offering likelihoods instead of definitive answers. Bengio warns of increasing risks as AI tools advance, highlighting incidents where models displayed deceptive tendencies to avoid shutdown. LawZero will focus on proving its methodology and gaining support from governments and corporations to build robust, scalable safeguards.

AI is redefining creativity in the culinary world. Grant Achatz, chef and owner of Next in Chicago, is testing its boundaries by crafting a nine-course menu with recipes adapted from AI-generated suggestions, including imaginary chefs' influences. While AI is now widely used for restaurant operations, its role in recipe innovation remains uncharted territory. Some chefs, like Dominique Crenn, resist AI's intrusion into what they view as a deeply human art form, while others, such as Ned Baldwin, embrace its technical assistance. AI tools like ChatGPT and Midjourney are producing unexpected ideas for dishes, visual designs, and even restaurant ambiance. Though AI can’t physically cook or construct, it challenges chefs to see outside the box.

Anthropic has rolled out a new voice mode for its Claude chatbot apps. The beta feature allows users to have full spoken conversations via Claude. It promises to make interactions more natural when users have busy hands but active minds. The feature supports English and offers five unique voice options that can be switched instantly with text mode. Users receive on-screen highlights along with a transcript and summary after every conversation. The technology is powered by Anthropic’s Claude Sonnet 4 model and counts against regular usage limits. Free users get around 20-30 voice chats per month while paid subscribers access extra Google Workspace integrations. Discussion continues about potential partnerships with Amazon and ElevenLabs.

Fannie Mae has announced a new partnership with Palantir to create a Crime Detection Unit targeting mortgage fraud. The alliance uses advanced artificial intelligence to analyze millions of data sets for suspicious transactions and promises to cut potential fraud losses. Fannie Mae leaders, including Chair William J. Pulte and President Priscilla Almodovar, detailed the plan at a press conference at the company’s D.C. headquarters. The initiative focuses on uncovering irregularities in multifamily loans, where previous fraud forced the GSE to set aside significant funds for credit losses. Palantir’s technology boasts rapid detection capabilities, reducing time from months to seconds. The full story is available at themortgagepoint.com. This move aims to protect the U.S. housing market and secure the future for homebuyers, lenders, and taxpayers.

Microsoft is introducing artificial intelligence enhancements to Paint, Snipping Tool, and Notepad for Windows 11 users. These updates include generative AI features and are rolling out to Windows Insiders on Canary and Dev builds, with some functions limited to Copilot+ PCs. Paint now allows users to design stickers with AI-generated prompts and offers Smart Object Selection for pinpointing elements in images. Snipping Tool upgrades include a “perfect screenshot” feature that edits automatically and a color identification tool providing HEX, RGB, and HSL values. Notepad’s new abilities enable AI-driven text creation and editing through simple menu commands. Together, these updates highlight Microsoft's push to integrate AI into everyday applications.

Cisco's latest research reveals a significant shift in customer service and support interactions. By 2028, agentic AI is expected to handle 68% of these interactions with technology vendors. The study, surveying nearly 8,000 global decision-makers, highlights the growing importance of AI-led customer experiences. Respondents anticipate benefits across the technology lifecycle, including improved IT productivity and cost savings. Despite AI's rise, 96% of customers still value human relationships in B2B tech partnerships. The research emphasizes the need for vendors to accelerate their agentic AI strategies to remain competitive.

Anthropic has launched its latest AI models, Claude Opus 4 and Claude Sonnet 4, showcasing significant advancements in autonomous coding and reasoning capabilities. Claude Opus 4 demonstrated the ability to code independently for nearly seven hours during testing, a substantial improvement over previous iterations. To address potential misuse, particularly in sensitive areas like bioweapons development, Anthropic implemented its AI Safety Level 3 (ASL-3) protocols, incorporating enhanced cybersecurity and prompt classifiers. Both models introduce features such as "thinking summaries" and an "extended thinking" mode, aiming to provide clearer reasoning insights and improved performance. This release positions Anthropic as a formidable competitor in the AI landscape, challenging offerings from OpenAI and Google.

Apple reportedly plans to open its on-device AI models, the core of Apple Intelligence, to third-party developers. This move, initially reported by Bloomberg, would give app creators access to Apple's smaller, local large language models through a software development kit. This new access will allow developers to integrate Apple's AI capabilities directly into their applications, a strategy similar to Google's approach with its Gemini Nano AI APIs for Android. While limited AI tools were previously available, this SDK offers deeper integration. The company is expected to announce this significant development during its Worldwide Developers Conference, starting June 9.