Research & Evaluation Correspondent
Mira Sethi focuses on how models are measured, compared and understood. Her coverage examines evaluation design, reproducibility and the assumptions behind claims of progress. She looks for the caveat that changes the headline.
742 published articles · Showing 1–20
Hugging Face has released llama.cpp v1.0.0, introducing stricter semantic versioning rules and significant updates to API, model support, security, and UI. The release aims to provide a more stable and predictable development environment for downstream applications....
Read article →Lenovo announced new ThinkSystem servers with 6th Gen AMD EPYC 9006 processors, Neptune liquid cooling, high-performance storage, and expanded Hybrid AI Factory Services for AI and HPC workloads....
Read article →A post by @RealMarvelX on X claims OpenAI has released 8 free GitHub projects for building AI agents, but Automatica has not confirmed it. The projects include tools for security, CLI interactions, model training, and more....
Read article →A tracker on X says Nvidia plans a $10 billion investment in Anthropic after its IPO, citing reporting from the New York Times. Neither company has confirmed the claim....
Read article →Bloomberg reports an unconfirmed Anthropic IPO event featuring JPMorgan’s CEO; Anthropic has not confirmed details....
Read article →@pratikg on X wrote on 9 November 2026 that Alibaba had named four Qwen4 tiers—Max, Plus, Flash and 27B—during a presentation, with only the 27B described as open weights....
Read article →Enclave's audit of DeepSeek V4.1 Flash's perfect 11/11 hacking benchmark score found that five wins used unintended routes, not the planned vulnerabilities. After closing those shortcuts, the same model scored 4/11 on the revised benchmark....
Read article →The models target robotics and autonomous-driving researchers who need controllable, photorealistic environments to train and evaluate policies without real-world risk....
Read article →Google’s new video model cuts costs by 30% and extends native clip length to 15 seconds, aiming to speed professional video workflows. The claims are self-reported and lack independent verification....
Read article →NVIDIA's Thursday post frames AI memory as an efficiency problem across the AI factory's hierarchy and claims up to 30× higher tokens per dollar on Vera Rubin NVL72 versus GB300 NVL72, attributing gains to full-stack codesign rather than raw HBM capacity increases....
Read article →Mistral AI launched Mistral Large 4 (Le Chonk) and expanded its Studio enterprise platform to help businesses build, deploy and govern AI agents, workflows and applications across multiple environments....
Read article →The first paper, "Routes of Reasoning: Reading and Reshaping Test-Time Compute in MoE Models", introduces a method to optimize test-time compute in Mixture-of-Experts (MoE) models by selectively editing expert router scores....
Read article →At the OCP Global Summit 2026, a hands-on report from a specific vendor booth detailed NVIDIA's liquid-cooled rack and server systems, with specifications attributed to NVIDIA....
Read article →Axe Compute has partnered with Penguin Solutions to accelerate the delivery of AI infrastructure, aiming to meet growing customer demand. The multi-year agreement includes technical support and services from Penguin Solutions....
Read article →Researchers have introduced a Sparse Transformer model to improve mechanistic interpretability in neural networks. The model outperforms dense transformers on specific benchmarks and offers insights into feature learning and polysemanticity....
Read article →NVIDIA has released the Nemotron 3.5 Super VL and two Bolt models, expanding its AI model family. The models are designed for specialized agentic tasks and are available now or coming soon....
Read article →The partnership aims to advance AI capabilities for the scientific community, with an initial focus on high-energy physics....
Read article →Muse Spark Open supports multiple input modalities, including text, images, documents, audio, and audio-video, and outputs text Meta AI Research. It uses a mixture-of-experts (MoE) architecture, which allows it to handle large-scale tasks efficiently....
Read article →The paper presents theoretical bounds on estimation error for dense and sparse mixture-of-experts models, quantifying the statistical cost of adaptive routing....
Read article →The initiative, launched by the European Commission, aims to support scientific progress by providing researchers with tools, compute resources, training, and collaborative projects....
Read article →