MIT's SERC symposium focused on AI's impact on society, featuring talks on air pollution forecasting and ethical AI deployment. Panel discussions highlighted challenges of aligning AI with human values and governance of AI systems.
NVIDIA introduces Dynamo Snapshot for AI inference on Kubernetes, reducing cold-start latency and improving scalability during demand spikes. CRIU and cuda-checkpoint work together to checkpoint GPU and CPU states, allowing for seamless restoration and minimal downtime.
Cross-validation in machine learning is deemed ineffective by a seasoned expert due to numerous flaws in both k-fold and leave-one-out techniques. The lack of generalizability and unreliable hyperparameter tuning make cross-validation a questionable practice in real-world scenarios.
Stanford University and Lambda Labs researchers developed OpenJarvis, an on-device framework that rivals cloud models in efficiency and latency. OpenJarvis allows easy composition of models, agents, and memory, with a unique LLM-guided spec search for optimization.
Miso Labs introduces MisoTTS, an 8-billion-parameter text-to-speech model with RVQ technology for expressive speech generation. It addresses the vocabulary size problem and interlocutor tone, achieving 110ms latency.
MIT, Georgia State University, and partners launch PATH to provide industry-aligned AI training for community colleges, emphasizing hands-on learning and collaboration. Program aims to develop practical AI skills and mindsets for a workforce prepared for the future.
The NSF has renewed funding for MIT's IAIFI, focusing on AI advancing physics and physics improving AI. Collaborative research across physics and AI is leading to groundbreaking discoveries and innovative scientific approaches.
GeForce NOW offers 18 new games this month, including the highly anticipated NTE: Neverness to Everness. Explore surreal worlds and classic remakes instantly via cloud streaming, with no downloads necessary.
NVIDIA introduces Nemotron 3 Ultra, a 550 billion parameter model with hybrid Mamba-Attention architecture, offering 6x higher inference throughput. The model uses Multi-Token Prediction for faster generation and achieves stable, accurate training with NVFP4 datatype.
Researchers from MIT and the MIT-IBM Computing Research Lab developed ChartNet, a dataset and series of open-source models that outperform commercial AI models in tasks like chart interpretation. This breakthrough could empower small firms with limited budgets to leverage AI for business trend analysis and scientific figure interpretation.
Amazon Bedrock enables generative AI for 100,000+ organizations worldwide, offering comprehensive capabilities for bold innovation. Introducing Amazon Bedrock Ops Alert, a proactive monitoring solution for sustainable operational management of AI workloads, empowering teams to drive real business impact.
AI agents must select the right tools for tasks to avoid errors and delays. Learn how SFT and DPO improve tool-calling accuracy in language models for reliable automation.
In 2026, AI agents excel at tasks like customer service, but struggle with complex inquiries. MIT and Harvard researchers improved AI's ability to ask questions through a "Battleship" game, leading to significant gains in performance and efficiency.
Amazon SageMaker AI now supports Fundamental's NEXUS model for accurate tabular data predictions in days. NEXUS offers deterministic results, native tabular understanding, and non-sequential reasoning for structured data analysis.
Google DeepMind released Gemma 4 12B, an encoder-free multimodal model for text, images, audio, and video. The model runs on a laptop with 16 GB of RAM, bridging the gap between edge-friendly and larger variants, with open-source weights available for download.