Gemini Can Now Write You a Song

2026-03-08 · Source: The AI Daily Brief: Artificial Intelligence News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Novice, medium

Summary

Google has launched LIIA 3, an AI music generator from DeepMind, allowing users to create 30-second music clips from text, image, or video inputs, with lyrics in eight languages. Accessible via the Gemini app and YouTube's Dream Track, it also generates custom cover art with Nano Banana and includes Synth ID watermarks. While not for full songs, it targets background music for YouTube Shorts and personal messages. Separately, Anthropic faced a brief controversy over changes to its terms of service regarding OOTH token usage for third-party apps, which was clarified to permit personal tinkering but restrict commercial use without API payment. Meta has revived plans for a smartwatch, codenamed Malibu 2, featuring health tracking and a built-in Meta AI assistant, aiming for a 2024 release. XAI introduced Gro Heavy, an enhanced version of Gro 4.2, increasing its sub-agent count to 16 for more detailed responses. Finally, an analysis of Chinese AI models suggests their real-world performance, particularly for agentic behavior and non-coding tasks, lags significantly behind benchmark claims, indicating they are at least one generation behind leading Western models.

Key takeaway

For CTOs and VPs of Engineering evaluating AI model adoption, you should prioritize real-world performance testing over benchmark scores, especially for agentic behaviors and non-coding use cases, as claims from some models, particularly Chinese ones, may not translate to practical utility. Additionally, consider how evolving terms of service from major AI labs like Anthropic and Google could impact your integration strategies for third-party AI agents, potentially pushing you towards direct API usage for commercial applications.

Key insights

Multimodal AI expands into music generation and wearables, while model evaluation and terms of service remain critical.

Principles

• Multimodal input enhances creative AI applications.
• Real-world AI performance often diverges from benchmarks.
• Terms of service can impact AI ecosystem development.

Method

XAI's Gro Heavy employs 16 sub-agents to debate and refine responses, aiming for more comprehensive and detailed outputs from complex queries.

In practice

• Use LIIA 3 for short-form content soundtracks.
• Test AI models beyond benchmarks for real-world utility.
• Monitor AI service terms for agentic application compatibility.

Topics

• AI Music Generation
• AI Wearables
• AI Model Benchmarking
• Large Language Models
• AI Terms of Service

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Product Manager, AI Engineer, Tech Journalist

Related on AIssential

• AI Music Generation Goes Consumer with Google’s MusicFX DJ — Google's MusicFX DJ brings real-time, interactive AI music generation to consumers via the Lyria RealTime diffusion model.
• Anthropic Makes TIME Cover — AI tools are rapidly democratizing creative and professional tasks, accelerating innovation across industries.
• What We are Missing in Multimodal LLM Evaluation? — Current MLLM evaluation benchmarks fail to assess cross-modal information integration, hindering true progress measurement.
• The handshake refusal heard around the AI world — Intense competition and rapid advancements define the current landscape of leading AI development.
• Any Headphones = Live Translation — AI adoption faces regulatory and ethical considerations while simultaneously advancing in practical applications and autonomous agents.
• not much happened today — Architectural innovation, training efficiency, and agentic orchestration are driving AI model advancements and competitive differentiation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The AI Daily Brief: Artificial Intelligence News.