MIT biologists developed PottsMPNN, an AI framework for designing proteins based on physical principles rather than solely on nature’s existing sequences. By understanding the sequence-energy landscape, PottsMPNN enhances stability predictions and diversifies design options. This advancement aims to enable novel protein engineering, moving beyond traditional evolutionary metrics for success.
Category: AI & Emerging Tech
IBM Adds Reasoning to Decades-Old Information Theory
IBM Research scientists have introduced a new concept called logical semantic entropy, extending information theory to include reasoning. This model enhances understanding of communication by factoring in what receivers can infer. It challenges existing notions about message efficiency, highlighting the significance of context and deductions in modern AI systems.
MIT Creates Building Blocks for Shape-Changing Smart Devices
MIT engineers developed “bifur-circuits,” modular 3D-printed blocks that maintain electrical connections while reshaping into various forms. These blocks employ mechanical bifurcation, allowing for stable configuration changes without degradation in connectivity. Potential applications include adaptive furniture, reconfigurable robotic tools, and versatile antennas. Future work aims to enhance interactivity and expand shape possibilities.
NVIDIA Jetson Orin Nano 2: 2x the Performance, 40% Less Power for Edge AI Robots
NVIDIA unveiled the Jetson Orin Nano 2 robotics computer, boasting 2x the inference performance and 40% less power consumption compared to its predecessor. With 78 trillion operations per second, it delivers advanced AI capabilities for edge devices. Scheduled for release in early 2027, it aims to support millions of developers in robotics applications.
MIT Study: AI-Generated Images Often Can’t Be Traced to Any Training Photo
MIT researchers found that in large AI image models, removing individual training images or entire artist portfolios often does not affect the generated output, a phenomenon termed “attribution decay.” This challenges current copyright frameworks and suggests model outputs may not be derivative works, raising questions around fair use and compensation for creators.
MIT Ditches Joysticks for a Wearable Arm to Train Excavator Operators Faster
MIT engineers built a new excavator training interface that replaces joysticks with a mini robotic arm. In tests, novice operators performed as well as experts from their very first session. Learning to…
Invesco’s CEO: “Trust Is Earned” — Cutting AI Onboarding From 25 Steps to 5
Andrew Schlossberg, Invesco’s CEO, emphasizes the critical role of AI in a firm’s strategy, streamlining client onboarding from 25 to 5 steps. He advocates for empowering all employees with AI tools and integrating AI into the overall strategy. Trust remains vital, shaped by consistency and earned through integrity, as Invesco adapts to evolving investor expectations.
Razorpay’s New AI Model Just Boosted Payment Success Rates 10%
Razorpay has launched Vulcan, India’s first AI payments foundation model, developed in collaboration with NVIDIA and AWS. Trained on 3 trillion data points from 4 billion transactions, it aims to enhance payment success rates by up to 10%, reduce fraud, and streamline digital payments for millions of users across India’s fragmented payment landscape.
AI Is Driving a New Threat to America’s Diesel Demand
Technological advancements in automation, AI, and electric vehicles are poised to significantly impact U.S. freight transportation and diesel fuel demand, which currently accounts for 22% of transportation energy. Initiatives like driverless trucking and truck platooning are improving efficiency, suggesting a transformative shift in the industry and reduced reliance on traditional fuels.
AI Financial Advice Study Finds LLMs Get the Basics Right -But Miss the Nuance
MIT Sloan research finds AI financial advice from ChatGPT and Gemini boosts savings and diversification, but stumbles on nuance — and varies by gender and experience. Roughly half of Americans now say…