🎭 Resemble AI's free voice cloning 🎭
Open-source model outperforms industry leaders

Resemble AI has open-sourced Chatterbox, a MIT-licensed voice cloning model that consistently outperforms ElevenLabs in blind tests. The model requires just five seconds of audio for voice cloning and features real-time generation with approximately 200ms latency. What sets Chatterbox apart is its emotion control parameter, allowing users to adjust intensity from monotone to dramatic expressions. The system includes built-in watermarking via PerTh, Resemble's imperceptible neural embedder, and is production-ready with simple pip installation, full documentation, and availability on both GitHub and Hugging Face.

🧠 Google releases MedGemma medical AI models 🧠
Open-source AI for healthcare applications

Google has released MedGemma, a new family of open-source medical AI models built on the Gemma 3 architecture. The suite includes a 4B-parameter multimodal model and a more powerful 27B text-only variant, both designed specifically for healthcare applications. These models aim to democratize access to medical AI technology while maintaining high standards of accuracy and reliability for clinical decision support, medical research, and patient care applications.

🤖 Hugging Face's affordable humanoid robots 🤖
Open-source robotics for everyone

Hugging Face is expanding beyond AI models with two new open-source humanoid robots: HopeJR and Reachy Mini. HopeJR is a full-body humanoid with 66 degrees of freedom capable of walking, gesturing, and grasping, priced under $3,000. The more affordable Reachy Mini, a desktop companion for testing voice-based and interactive AI applications, costs between $250-$300. Built in collaboration with The Robot Studio and leveraging expertise from Hugging Face's acquisition of Pollen Robotics, both robots are designed to be hackable and rebuildable, with support from Hugging Face's LeRobot ecosystem of open datasets and model libraries.

INTERESTING NEWS

📊 Meta research challenges AI reasoning assumptions 📊

Meta's FAIR team and The Hebrew University have overturned a core assumption about large language models: longer reasoning chains don't necessarily mean better answers. Their study showed shorter chains improve accuracy by up to 34.5%, reduce compute requirements by 40%, and cut inference time by a third. The "short-m@k" method runs parallel short reasoning paths and uses majority voting to select the best result, outperforming traditional chain-of-thought techniques while requiring less computational resources.

🧪 Intology AI makes scientific history 🧪

Intology AI's Zochi has become the first AI to independently achieve peer-reviewed publication at ACL 2025, a prestigious natural language processing conference. The system autonomously completed the entire research process, from analyzing thousands of papers to designing experiments and writing the manuscript. Zochi's paper on multi-turn jailbreaking achieved a 4.0 meta-review score, placing it in the top 8.2% of all ACL submissions. Operating with minimal human intervention, the AI identified research gaps, implemented new methods, and validated results.