GPT-OSS-20B reasoning model thinks in English, struggles with multilingual reasoning. Fine-tuning with SFT and GRPO on Amazon SageMaker HyperPod shows promising results.
Canadian organizations utilizing Amazon Bedrock can now access the latest OpenAI models, including GPT-6, from Canadian AWS Regions with cross-Region inference. The models available in Canada offer a range of capabilities, from strong reasoning to fast, low-cost inference, supporting text and image inputs with a 1-million-token context window.
Establishing cryptographic identity through X.509 PKI is crucial for AI agents accessing sensitive data. Keyfactor EJBCA and AWS services provide secure authentication for multi-agent AI frameworks, preventing unauthorized access and ensuring trust in machine identities.
German Chancellor Friedrich Merz and NVIDIA CEO Jensen Huang discuss Germany's AI roadmap at Kraftwerk, emphasizing AI applications in manufacturing and automotive. Germany aims to merge engineering expertise with AI technology, collaborating with NVIDIA to advance industrial AI models and establish a cutting-edge ecosystem in Europe.
Amazon Bedrock now supports structured outputs for all Anthropic Claude models, ensuring schema-compliant JSON responses. This eliminates the need for custom parsing and post-processing, making tasks like data extraction and contract analysis more efficient and reliable.
vLLM introduces push-based NIXL connector for efficient KV transfer, reducing TTFT and improving TPOT in LLM inference. Benchmark shows up to 3x better TTFT and 30% better TPOT with push mode on Amazon EC2 P5en instances using Qwen3-32B model.
Amazon SageMaker JumpStart offers a curated catalog of AI models, templates, and notebooks, streamlining ML model selection to deployment. Leading providers like Alibaba, NVIDIA, and Google contribute diverse models for model discovery, deployment, and customization.
Deepgram leverages AWS AI chips to enhance efficiency, processing 11-70% more audio per dollar for customers. Deepgram and AWS collaborate to optimize Nova-3 on Trainium2, improving inference for speech-to-text models.
Researchers at MIT and Boston Children’s Hospital developed E(3)-Pose, a deep learning model that tracks fetal brain motion during MRI scans to improve image quality and accuracy. The model embeds physics principles to enhance brain orientation identification, aiding in early detection of abnormalities in fetal brain development.
TwelveLabs releases Pegasus 1.5 on Amazon Bedrock, a video language model transforming video into structured metadata. Pegasus 1.5 offers time-based metadata, multimodal prompting, and on-the-fly processing, enhancing video analysis capabilities.
MIT researchers developed a multimodal AI model combining 3D structures with language to improve material design. The model outperformed text-only models in identifying promising materials for applications like solid-state lithium-ion batteries.
NVIDIA RTX Spark laptops, with powerful AI capabilities, now available for $2,399 from top PC makers. Jensen Huang surprises fans at launch events in Berlin, sparking excitement worldwide.
Reflection AI introduces Beam, an open-weight model for agentic coding and long-horizon tasks. Beam's 501 billion total parameters and 23 billion active parameters shape its deployment on AWS, offering developers control and customization options.
AI helped improve JavaScript decision tree regression, finding edge cases and suggesting performance enhancements. New version created after 16 hours of challenging work, with synthetic data demo showcasing improved accuracy and efficiency.
AI agents must maintain quality to avoid broken relationships with users. Amazon Bedrock AgentCore ensures accurate, fluent responses for business decisions at scale.