Study by MIT & others shows AI assistance in disease diagnosis varies based on user expertise. Non-experts rely on explanations, while clinicians perform best with only predictions. Importance of user-centered AI design highlighted to prevent errors and overreliance.
Over 120 organizations in the Open Secure AI Alliance are developing SAFE guidelines for AI cybersecurity at Black Hat. Members like NVIDIA and Cisco are contributing to the proposal to enhance collective defense against evolving cyber threats.
MIT researchers led by Ju Li are developing stable sodium-metal batteries as a low-cost alternative to lithium-ion batteries. A breakthrough electrolyte molecule called DMTMSA shows promise in addressing stability issues in sodium batteries, as reported in the journal Joule.
Three methods to train linear regression models: SGD, left pseudo-inverse, MP pseudo-inverse. MP pseudo-inverse offers six main algorithms, with QR-Householder being the preferred choice for implementation.
Amazon Bedrock AgentCore offers a platform for automated web insight extraction, using AI for design, product, and market intelligence. The solution includes a managed browser service and semantic search capabilities for monitoring competitor products, industry trends, and regulatory changes.
Automated Reasoning policy refinement in Amazon Bedrock automates diagnose-and-fix cycles, achieving up to 99% verification accuracy. New refinement modes address rule and language issues, streamlining policy development for customers.
Sasha Rakhlin named director of MIT Statistics and Data Science Center, praised for mentorship and research contributions. Endowed chair created by Dick Larson to support community tackling evolving questions in statistics and AI.
Meta's GEM model doubles training efficiency to 20-25% MFU, scaling FLOPs 4x in 12 months through innovative co-design and customized kernel library. Unique challenges in training GEM due to hybrid architecture and recommendation-domain data properties.
Multicollinear data impacts model interpretability. VIF analysis can detect multicollinearity, aiding in data evaluation.
Organizations are integrating AI-powered analytics with natural language (Text2SQL) answers, emphasizing the importance of semantic richness flowing from data catalogs into AI products like Amazon Quick. The challenge lies in bridging the gap between rich metadata upstream and delivering curated, trustworthy AI answers and dashboards to end users.
AI agents transitioning to production face challenges of slow performance and memory issues. Amazon Bedrock AgentCore and CloudWatch help diagnose bottlenecks and optimize agent architecture.
MIT's Daniela Rus wins 2026 High-Tech Prize for groundbreaking work in robotics and AI, focusing on self-organizing robot collectives and soft robotics. Her research at CSAIL is shaping the future of physical AI, with applications ranging from healthcare to infrastructure.
Amazon Bedrock introduces Advanced Prompt Optimization to streamline prompt migration and optimization for generative AI applications. The tool compares performance across multiple models, reducing manual effort and improving efficiency in the software development lifecycle.
GeForce NOW transforms everyday laptops into powerful gaming setups, allowing seamless switching between schoolwork and gaming. The addition of Halo: Campaign Evolved to the library offers an exciting gaming experience for members, showcasing the platform's versatility and convenience.
Yahoo's omnichannel Demand-Side Platform (DSP) tackles challenges in digital advertising with advanced technology and sophisticated audience targeting. By implementing Amazon Bedrock, Yahoo enhances Search Retargeting (SRT) capabilities, reaching users based on historical search behavior with generative AI-powered keyword expansion.