🤖 AlphaEvolve: Google's self-coding AI agent 🤖
Autonomous code optimization and discovery

Google DeepMind has unveiled AlphaEvolve, a groundbreaking AI coding agent powered by Gemini models that autonomously writes and optimizes code to solve complex problems. Unlike previous AI systems that use reinforcement learning to optimize model weights, AlphaEvolve modifies, evaluates, and optimizes code directly. The agent has already demonstrated impressive capabilities, accelerating AI model training, enhancing chip performance, and solving challenging mathematical problems. In tests with over 50 open math challenges, AlphaEvolve matched the best-known solutions 75% of the time and improved upon them in 20% of cases, leading to genuine mathematical breakthroughs.

🎵 Stable Audio Open Small launches 🎵
Text-to-audio model for on-device use

Stability AI and Arm have released Stable Audio Open Small, a lightweight text-to-audio model designed specifically for on-device deployment. This new model enables real-world applications of AI audio generation with minimal computational requirements, making it accessible for mobile devices and edge computing scenarios. The release includes both the model weights and implementation code, allowing developers to integrate audio generation capabilities into their applications without requiring cloud processing or expensive hardware.

 🧠 Psyche Network: Distributed AI training 🧠
Community-powered model development

Nous Research has launched Psyche, a Solana-based distributed training system that allows anyone with compatible hardware to volunteer their GPUs for AI model training. The project's first initiative, "Consilience," aims to create a 40B parameter model using 20 trillion tokens, making it the largest community-powered AI training effort conducted to date. This decentralized approach could democratize AI development by tapping into unused computing resources across the globe, potentially accelerating innovation outside traditional research institutions.

INTERESTING NEWS

⚖️ Claude AI hallucinates legal citation in court ⚖️

In a Northern California courtroom, Anthropic's legal team faced embarrassment after Claude AI completely fabricated a legal citation that was then used in court. The AI generated a made-up title and nonexistent authors that somehow passed the team's "manual citation check." The incident highlights the ongoing risks of AI hallucinations in high-stakes environments and serves as a reminder that human verification remains essential, especially in legal contexts where accuracy is paramount.

🎭 FBI warns of deepfake scams targeting officials 🎭

The FBI has issued an alert about sophisticated scammers using AI-generated voice deepfakes to impersonate senior US government officials in targeted attacks. This warning follows several high-profile deepfake incidents, including a LastPass attack using a CEO's synthesized voice and a political robocall that impersonated President Biden last year. As voice synthesis technology becomes more convincing, organizations are being urged to implement additional verification protocols for sensitive communications and financial authorizations.

H-FARM EVENTS ON AI

FutureShots25(May 23-24)

While we wonder how the world will change, AI is already rewriting it. Will we lead the way or just watch?
Come and meet people who turn prompts into voice, prototype humanoids, design systems used by millions, restore mosaics with AI, make robotic arms, and bring art into space.