Gemini AI Music Generator & Meta Smartwatch Revival

Google has launched LIIA 3, an AI music generator developed by DeepMind, expanding its presence in multimodal AI. This latest version allows users to generate 30-second music clips using text, images, or video inputs, a unique capability compared to text-only generators like Suno. Lyrics can be generated in eight languages, including German, French, Spanish, and Hindi.

The feature is accessible directly within the Gemini app by switching to a musical output and is also being integrated into YouTube's Dream Track tool. This integration enables creators to quickly generate soundtracks for YouTube Shorts. Each generated track is accompanied by custom cover art created by Nano Banana. While previous LIIA versions were limited to Google Cloud's Vertex program, this release significantly broadens access.

LIIA 3's primary limitation is its 30-second clip length, as the model cannot build upon initial generations to create full songs. Google intends for these clips to serve as background music for YouTube Shorts or fun, interactive personal song messages, emphasizing expression over musical masterpieces. The music includes Synth ID audio watermarks, clearly identifying them as AI-generated. While some compare it to Suno, LIIA 3 is positioned more as a social feature, similar to how Suno is often used for interactive fun rather than professional music production.

Anthropic recently triggered a wave of complaints by updating its terms of service, specifically regarding the use of OAUTH tokens. The new policy stated that using OAUTH tokens obtained through Claude's free, Pro, or Max accounts in any other product, tool, or service, including the agent SDK, is not permitted. OAUTH tokens function similarly to API keys for regular Anthropic subscriptions, allowing users to access AI models via third-party applications.

This change particularly affected users powering OpenClaude agents with their Claude Max accounts, leading to significant backlash. Critics accused Anthropic of self-destructive behavior, citing previous actions like blocking token use in non-Claude code apps and pursuing developers for branding infringement. Many users expressed frustration, questioning the company's approach to competition and developer relations.

Anthropic's Head of Research, Shahipard, clarified that the update was a "docs cleanup" and not intended to block personal tinkering. He stated that the goal was to compel third-party businesses to pay for usage through the API, rather than using personal subscriptions. However, this clarification only added to the confusion, leaving users uncertain about the permissible scope of use for personal agents. The incident highlighted broader concerns about walled gardens in AI and the future of modular AI use cases, with some users finding similar restrictions already in place with Google's OAUTH tokens.

Meta has revived its plans to release a smartwatch, integrating it into its broader AI device lineup. Rumors of a Meta smartwatch, internally codenamed Malibu, first emerged in late 2021 with leaked prototypes featuring two cameras: one on the dial for video conferencing and another on the underside for quick photo capture after removing the watch. A key design concept was the ability to read nerve signals in the wrist, enabling the device to function as a controller, a technology later seen in Meta's haptic control wristbands for the Orion smart glasses prototype.

Project Malibu was initially canceled in mid-2022 as Meta shifted its focus to smart glasses as its primary wearable strategy. However, The Information now reports that the smartwatch project has been revived under the code name Malibu 2. This new iteration is expected to include health tracking features and a built-in Meta AI assistant. The decision to revive the smartwatch reportedly stemmed from a project strategy meeting late last year, driven by concerns over a bloated product lineup for augmented reality glasses.

Meta is now concentrating on a limited number of strategic bets, including a new version of Ray-Ban smart glasses expected later this year and AR glasses potentially arriving in 2027. The Malibu 2 smartwatch is planned for release this year, positioning Meta in direct competition with Apple and Google in the smartwatch market. This move underscores Meta's commitment to the wearable category as a central component of its AI strategy, contrasting with Apple's decision to pass over a camera-equipped Apple Watch due to practical issues like clothing obscuring the camera.

xAI has announced Grok Heavy, an advanced version of its Grok 4.2 public beta, which significantly scales up its sub-agent architecture. While Grok 4.2 introduced four sub-agents to debate responses before providing a final answer, Grok Heavy increases this count to 16. This aims to either improve answer quality or, at minimum, generate highly detailed outputs, albeit at a higher token cost.

An example query, "How does chaos birth cosmic order?", demonstrated Grok Heavy's capabilities. The 16 agents debated the response for over a minute, producing a 700-word report that cited almost 900 references. While the accuracy and usefulness for such a subjective question are hard to definitively judge, the output showcased an impressive level of detail and depth, highlighting the experimental nature and potential of this multi-agent approach.

Separately, discussions have emerged regarding the real-world performance of Chinese AI models compared to their benchmark evaluations. Lindy founder Flo Crell noted that despite high benchmark scores, Chinese models consistently underperform in agentic behavior and non-coding use cases. He suggested that these labs might be distilling frontier models, training specifically for evaluations, or potentially even stealing weights. The consensus indicates that Chinese models are still at least one generation behind leading Western models like Sonnet and Opus, emphasizing the need to critically evaluate benchmarks against practical application.