Is AI Helping Us Be More Productive, Two New Robotics Partnerships, Meet Midjourney 6.1 ... and more

Welcome to the latest episode of AI Weekly Breakthroughs, a show where we explore all the key happenings from the AI world to see what really matters. This week was a big week in AI for one field in particular – healthcare. 

Virchow2 and Virchow2G Revolutionize Cancer Diagnosis

Paige and Microsoft released Virchow2 and Virchow2G, AI models designed to transform cancer diagnosis. Trained on over 3 million pathology slides from 225,000 patients, these models cover 40+ tissue types and use various staining methods. Virchow2G, with 1.8 billion parameters, sets a new performance standard. These models enhance diagnostic accuracy and support life sciences and pharmaceutical research. Virchow2 is available on Hugging Face for research.

‘Game-Changing’ AI for Detecting Hidden Heart Inflammation

Oxford University spinout Caristo Diagnostics developed an AI model that detects hidden heart inflammation, a precursor to heart attacks. This AI identifies invisible biological processes in CT scans, allowing for earlier intervention. Piloted in UK hospitals, it identified patients with a 20 to 30 times higher risk of fatal cardiac events. Nearly half were advised on lifestyle changes or medication. The AI is approved in Europe and Australia.

AI Predicts 3D Receptor Structures Enhancing Drug Development

Researchers at Uppsala University used AI to predict the 3D structure of the TAAR1 receptor, aiding drug development for mental health disorders. Supercomputers identified potential drug molecules from chemical libraries, validating the AI model through successful receptor activation tests. AI accuracy surpassed traditional methods, promising improved drug discovery.

New ML Model Boosts Disease Prediction Accuracy

A new ensemble feature selection model enhances disease prediction using Electronic Health Records. The SEV-EB algorithm combines statistical, deep, and optimal features, achieving 95% accuracy and a 94% F1-score, outperforming traditional methods. This model captures both short- and long-term health patterns, marking a significant healthcare advancement.

Nvidia Faces Scraping Allegations

A whistleblower alleges Nvidia scraped millions of videos to train its AI systems, including those for Omniverse and self-driving cars. The data collection from platforms like YouTube and Netflix raises legal and ethical concerns. Despite Nvidia’s claims of broad data-use approval, this reflects ongoing debates about the legality of such practices.

Nvidia Tools Leverage Apple Vision Pro for Robot Control

Nvidia introduced tools to advance humanoid robot development using the Apple Vision Pro. Nvidia’s NIM microservices and OSMO orchestration service enable AI and simulation workflows, translating user movements into robot actions for precise teleoperation. This approach aims to simplify humanoid robot development.

Google’s Table Tennis Robot

Google’s new table tennis-playing robot, detailed in a paper on ArXiv, uses 20 motion-capture cameras, dual 125 FPS cameras, a 6 DOF robot, and a specialized paddle. It employs convolutional neural networks trained on a large dataset. The robot excelled against beginners but struggled with intermediate and advanced players.

Tesla’s Voxel-Based Vision System for Autonomous Robots

Tesla patented a vision system for autonomous robot navigation using a single neural network to process camera data and predict 3D environmental occupancy. This method divides space into 3D voxels to understand surroundings in real time, advancing Tesla’s humanoid robot development.

Other Updates

– OpenAI co-founder John Schulman joins Anthropic

– OpenAI’s president Greg Brockman takes an extended leave

– Palantir and Microsoft team up to provide AI services for U.S. defense and intelligence

– Google Meet’s new AI feature “Take notes for me” will automatically take meeting notes