The Daily AI Briefing - 03/07/2025
"Welcome to The Daily AI Briefing!" Hello and thank you for joining us today. I'm your host, bringing you the most significant developments in artificial intelligence. From AI-generated music climbing the Spotify charts to breakthrough collaborative algorithms, we've got a packed show covering the rapidly evolving AI landscape. Let's dive into today's stories. In today's episode, we'll explore an AI band that fooled Spotify listeners, examine new expressive AI voices from Rime, discover how competing AI models are learning to collaborate, look at Google's new Gemini command-line tool, and much more. Our top story today focuses on "The Velvet Sundown," an AI-generated music band that recently reached 500,000 monthly listeners on Spotify before being revealed as an "art hoax." The music was created using Suno's AI generator, with albums appearing in June having no digital footprint. While Deezer flagged potential AI usage, Spotify had no disclosure requirements in place. After initially denying AI involvement, an "adjunct member" named Andrew Frelon later admitted to using Suno's "Persona" feature to maintain a consistent vocal style. This case raises an interesting question: does the origin of content matter if consumers genuinely enjoy it? Moving to voice technology, Rime has introduced remarkably realistic AI voices with distinct personalities for creating agentic experiences. These voices include human-like features such as laughing, breathing, and sighing, making interactions feel more natural. The platform supports multilingual text-to-speech via API or on-premises solutions and has reportedly driven double-digit conversion increases for brands like Dominos. In a fascinating development for collaborative AI, Japanese lab Sakana AI has introduced AB-MCTS, an algorithm enabling competing AI models like ChatGPT, Gemini, and DeepSeek to work together on complex problems. This approach solved 30% of ARC-AGI-2 puzzles versus just 23% for the top individual models. The system dynamically allocates different models based on their strengths, with some handling strategy while others excel at code. Their underlying framework "TreeQuest" has been released as open-source. For developers, Google has launched a new command-line AI tool that brings code analysis, app generation, and workflow automation directly to terminals. Installation requires Node.js 18+ using "npm install -g @google/gemini-cli." Users get 60 requests per minute and 1,000 per day with a Google account. It can explain codebase architecture, analyze PRs, resolve git conflicts, and even generate apps from sketches. On the enterprise front, ReMarkable is scaling with AI agents to prevent human team burnout. CTO Nico Cormier shared how they use Agentforce to automate customer support with an agent named "Mark" and internal IT support with "Saga" in Slack, finding the right balance between AI and human interaction. In research news, scientists from Helmholtz Munich have developed "Centaur," an AI model simulating human decision-making. Fine-tuned on Meta's LLaMA using data from 60,000 participants across 160 psychology experiments, it accurately predicts human choices in various tasks. Centaur outperformed 14 traditional cognitive models on nearly all tested tasks, including gambling, memory, and problem-solving scenarios. Researchers plan to use it as a "virtual laboratory" to test theories and understand cognitive processes. Several AI tools are trending today, including Huawei's open-source reasoning model Pangu Pro, FreePik with unlimited image generation for Premium+ accounts, Perplexity with its new Max tier, and Higgsfield Soul offering advanced image AI with free daily generations. For those seeking AI career opportunities, positions are available including Designer at The Rundown, IT Systems Technician at xAI, Applied Data Scientist at Deepmind, and Research Scientist Manager at Meta. In other news, Perplexity has launched a $200 per month Max tier, Open