AI News

Impact Score

OpenAI’s new LLM exposes the secrets of how artificial intelligence really works

14 Nov 2025

OpenAI built an experimental, weight-sparse LLM designed to be far easier to inspect than typical models. The work aims to reveal the internal mechanisms behind hallucinations and other failures, and to improve how much we can trust artificial intelligence.

artificial intelligence, machine learning, mechanistic interpretability, openai

Impact Score

Us2.ai advances cardiac imaging with multimodal Artificial Intelligence

14 Nov 2025

Us2.ai highlights studies and partnerships showing how Artificial Intelligence applied to echocardiography can surface hidden cardiac amyloidosis and support broader structural heart disease workflows.

artificial intelligence, cardiology, research, us2.ai

Impact Score

MMCTAgent enables multimodal reasoning over large video and image collections

14 Nov 2025

MMCTAgent enables dynamic multimodal reasoning with iterative planning and reflection. Built on Microsoft's AutoGen framework, it combines language, vision, and temporal understanding for complex tasks such as long video and image analysis.

computer vision, microsoft-research, multimodal

Impact Score

NVIDIA sweeps MLPerf training v5.1 for artificial intelligence

14 Nov 2025

NVIDIA swept all seven tests in MLPerf Training v5.1, posting the fastest training times across large language models, image generation, recommender systems, computer vision and graph neural networks. The company was the only platform to submit results on every test, highlighting its GPUs and CUDA software stack.

machine learning, nvidia, performance

Impact Score

Panmnesia samples first PCIe 6.0/CXL 3.2 fabric switch for Artificial Intelligence infrastructure

13 Nov 2025

Panmnesia has begun sampling a hybrid PCIe 6.0/CXL 3.2 fabric switch that supports port-based routing. The switch targets Artificial Intelligence infrastructure and remains backward compatible with earlier PCIe and CXL generations.

cxl, hardware, panmnesia, pcie

Impact Score

Tachyum unveils 2 nm Prodigy universal processor for Artificial Intelligence rack efficiency

13 Nov 2025

Tachyum announced a 2 nm Prodigy Universal Processor aimed at running much larger Artificial Intelligence models at a fraction of current cost, and claimed dramatic rack-level performance gains versus NVIDIA Rubin systems.

artificial intelligence, machine learning, semiconductors, tachyum

Impact Score

NVIDIA sweeps MLPerf Training v5.1 benchmarks

13 Nov 2025

NVIDIA posted the fastest times across all seven MLPerf Training v5.1 tests, demonstrating new gains in Artificial Intelligence training from GPUs to networking and software. The company submitted results on every benchmark and introduced Blackwell Ultra and NVFP4-backed optimizations.

benchmarks, high performance computing, machine learning, nvidia

Impact Score

Designing for artificial intelligence agent compatibility in the MCP era

13 Nov 2025

Microsoft Research examines how agentic Artificial Intelligence and expanding tool sets are driving system convergence and rising complexity, and explores the Model Context Protocol as a standard to improve agent collaboration.

agents, artificial intelligence, microsoft-research

Impact Score

When Artificial Intelligence meets biology

13 Nov 2025

Microsoft researchers disclosed a confidential study that explored how open-source Artificial Intelligence tools could bypass biosecurity checks and said their work helped produce fixes now influencing global standards.

biosecurity, ethics, microsoft-research

Impact Score

more Artificial Intelligence-resilient biosecurity with the Paraphrase Project

13 Nov 2025

Microsoft researcher Eric Horvitz and collaborators discuss the Paraphrase Project, a red-teaming effort that exposed and helped secure a biosecurity vulnerability in Artificial Intelligence-driven protein design. The episode frames the work as a practical model for mitigating dual-use risks in Artificial Intelligence applications.

artificial intelligence, biosecurity, microsoft, research

Impact Score

RedCodeAgent: automatic red-teaming agent against diverse code agents

12 Nov 2025

RedCodeAgent automates and improves red-teaming attack simulations against diverse code agents to help uncover real-world threats other methods overlook. The discussion appears in a Microsoft Research post.

code agents, microsoft-research, red teaming, security

Impact Score

AMD positions Instinct MI400 against NVIDIA Vera Rubin, MI500 artificial intelligence series due in 2027

12 Nov 2025

At Financial Analyst Day 2025, AMD outlined the Instinct MI400 artificial intelligence accelerators arriving in 2026 and confirmed an annual refresh cadence with the MI500 set for 2027. AMD says the MI400 will roughly double compute over the MI350, add HBM4 memory and broaden data format support.

amd, hardware, machine learning