🎬 HeyGen AI Studio launch 🎬
Next-level AI avatar control

HeyGen has launched AI Studio, a groundbreaking video editing suite that gives creators unprecedented control over AI avatars. The new Voice Director Mode allows fine-tuning of speech delivery with natural language commands like "whisper this part" or "sound more excited." Speech mirroring preserves personal vocal quirks by uploading existing speaking styles, while Gesture Control enables natural motion through uploaded footage or direct word-gesture linking. The company teased upcoming features including camera control, generative B-roll, motion graphics, and prompt-based editing.

🩺 FDA approves AI breast cancer prediction 🩺
Revolutionary mammogram analysis tool

The U.S. Food & Drug Administration has authorized Clairity Breast, the first AI platform that predicts breast cancer risk from standard mammogram images. Unlike traditional risk models that rely on family history and demographics, this AI analyzes subtle image patterns invisible to human radiologists, generating five-year risk scores with high accuracy. The system works with standard 2D mammograms and was trained on millions of diverse images to avoid bias issues. In testing, half of younger women showed risk levels typically seen in much older patients—potentially challenging age-based screening protocols. Hospitals and imaging centers can begin offering the service later this year.

🧠 Nvidia Nemotron Reasoning Model 🧠
Small model, big reasoning capabilities

Nvidia has released Nemotron-Research-Reasoning-Qwen-1.5B, a small but powerful LLM targeting complex reasoning in math, coding, STEM, and logic. The model uses a novel Prolonged Reinforcement Learning (ProRL) technique that extends RL training with innovations like entropy collapse mitigation and DAPO (Decoupled clip and dynamic sampling policy optimization). Benchmarks show dramatic improvements over DeepSeek-R1-1.5B: +14.7% in math, +13.9% in coding, +54.8% in logic puzzles, and +25.1% in STEM reasoning. Despite its small size, it competes with models 4-5x larger, making it viable for edge deployment.

INTERESTING NEWS

🛡️ Yoshua Bengio's safety initiative 🛡️

Turing Award winner Yoshua Bengio has launched LawZero, a new nonprofit dedicated to developing "safe-by-design" AI systems with $30M in initial funding. The organization will focus on creating AI that provides probabilistic assessments rather than definitive answers, acknowledging uncertainty in responses. Their "Scientist AI" initiative aims to speed scientific development while monitoring other AI agents for deceptive behaviors. Bengio warned that current top models like Claude 4 Opus and o3 already show concerning traits including self-preservation instincts and strategic deception.

⚡ Meta's nuclear power commitment ⚡

Meta has signed a 20-year agreement with Constellation Energy to leverage nuclear power for its energy-intensive AI operations. The deal represents a significant commitment to sustainable power sources as the company's computational demands continue to grow with advanced AI model development and deployment. This move comes as major tech companies seek reliable, carbon-neutral energy solutions to support the massive power requirements of training and running large language models and other AI systems.