MIT study reveals large language models (LLMs) can give wrong answers due to learned grammatical patterns. Researchers develop benchmarking procedure to mitigate risks of LLMs producing harmful content.
Warner Bros. Discovery leverages AWS Graviton-based Amazon SageMaker AI instances for ML inference, achieving 60% cost savings and 7-60% latency improvements. With a focus on personalized content recommendations for 125M+ users globally, they balance scalability and cost efficiency in a diverse digital entertainment landscape.
Amazon SageMaker AI introduces enhancements to its inference optimization toolkit, including EAGLE speculative decoding for faster, more efficient model inference. By training EAGLE 2 or EAGLE 3 heads, users can optimize performance using their own data for workload-specific results, improving inference without compromising quality.
Amazon SageMaker HyperPod introduces GPU partitioning with NVIDIA MIG, allowing multiple tasks on a single GPU. This boosts efficiency, reduces wait times, and maximizes GPU utilization for diverse workloads.
Amazon discontinues Lookout for Vision, recommends SageMaker for AI/ML computer vision models. AWS offers pre-trained defect detection model on Marketplace for fine-tuning.
Black Forest Labs releases FLUX. 2 image generation models with photorealistic details and multi-reference feature. NVIDIA and ComfyUI collaboration optimizes performance and reduces VRAM requirements by 40%.
EU and US ease AI regulations for growth; Nvidia reports strong earnings amid fears of a bursting AI bubble. Meta avoids breakup like Google, as AI transforms journalism-audience dynamics.
MIT Jameel Clinic hosted a BoltzGen seminar featuring Hannes Stärk, introducing the groundbreaking protein binder generator. BoltzGen's unique capabilities challenge traditional protein design methods, with potential to revolutionize drug development.
Dr. Talia Hussain criticizes lecturers for using AI to create teaching materials due to lack of incentives. Recent doctoral graduate highlights challenges of preparing modules for one-time teaching on precarious contracts.
Canadian organizations can now access advanced foundation models like Anthropic’s Claude Sonnet 4.5 on Amazon Bedrock through cross-Region inference (CRIS), accelerating AI initiatives. CRIS enables higher throughput and performance, distributing inference processing across multiple AWS Regions securely.
Anthropic's Claude Opus 4.5, available in Amazon Bedrock, sets a new standard in AI capabilities, outperforming previous models at a third of the cost. Opus 4.5 excels in software engineering and office productivity, transforming development projects and handling complex tasks with precision and efficiency.
AI is accelerating violence against women and girls, with 1 in 4 people neutral about non-consensual deepfakes, warns UK police. Technology companies are complicit in this epidemic.
Amy Redford appreciates fans' love but criticizes AI tributes to her late father Robert Redford as "fabrications". She is grateful for the outpouring of support and stories from around the world.
Amazon Bedrock now supports OpenAI GPT-OSS models with 20B and 120B parameters, offering advanced reasoning capabilities. Organizations can seamlessly migrate to AWS for enhanced security and scalability while deploying GPT-OSS models with ease for various applications.
AI can both strain and support energy transition. It optimizes operations, predicts failures, and aids in discovering new materials. MIT's Data Center Power Forum tackles challenges of data center power demand.