News
Short posts prepared from approved sources with clear links back to the original articles.
RSS feedDario Amodei says Anthropic is not seeking a ban on open-weight models, but remains worried about powerful models being misused or helping authoritarian states.
Fish Audio says more than 8 million people use its open-source or hosted models, and the new seed round will support voice tools for creators and enterprises.
The LFM2.5 encoder models on Hugging Face aim to support document-scale classification and routing tasks with 8,192-token context and faster CPU inference.
Its Jetson modules and developer kits give students, researchers, and robotics builders a compact way to run edge AI and autonomous-machine projects.
AWS combines LangGraph, Strands, and Bedrock AgentCore in a market-surveillance example meant to show stateful orchestration, recovery, memory, and observability.
Microsoft says MAI-Cyber-1 Flash is trained for vulnerability analysis and will support new tools for continuously reducing customer security exposure.
Amazon is said to be scaling back several Nova models while moving resources toward a Frontier Model Research group and a new model planned for re:Invent.
PJM Interconnection plans to curtail very large data centers during shortages starting in 2027, a sign that AI-driven infrastructure demand is straining power markets.
Customers are shifting from AI experiments to broader “Frontier Transformation,” with Copilot, Agent 365, and Microsoft IQ positioned as the stack for that shift.
Grafana Assistant has expanded beyond single-tool observability, letting teams ask natural-language questions across clouds, databases, trackers, and monitoring systems.
Managed Agents in the Gemini API now default to Gemini 3.6 Flash and add hooks, model selection, budgets, scheduled triggers, and free tier access.
LangChain describes an agent-first data stack that shifted its analytics team from answering every request to maintaining the context agents need to answer well.
Runpod’s latest guide focuses on practical configuration choices—deployment, storage, GPU sizing, cleanup, and region selection—that can waste money or slow model work.
NVIDIA has reportedly invested in Ilya Sutskever’s Safe Superintelligence, a move The Decoder says will shift the lab’s compute plans away from Google chips.
Fish Audio raised a $52 million seed round to build AI voice models for creators and enterprises, adding funding to a crowded market for synthetic speech tools.
A new arXiv paper, “Same Question, Different Answers,” evaluates large language model reliability by looking at whether models stay consistent across related prompts.
An arXiv paper on text-to-SQL explores training models to decide when extra reasoning helps, using supervised fine-tuning and preference optimization.
Perplexity’s Personal Computer for Windows can work across local files, Microsoft 365 apps, and the web for paying Max and Enterprise Max users.
NVIDIA is promoting Jetson modules and developer kits as compact hardware for students, researchers, and robotics teams building AI systems outside the data center.
AWS describes a financial-services surveillance agent that combines LangGraph workflow orchestration with Strands reasoning on AgentCore for regulated, multi-step investigations.
WIRED’s interview with A.G. Sulzberger connects the Times’ lawsuit against OpenAI and Microsoft to wider worries about how AI changes search, attribution, and news economics.
The OlmoEarth Platform turns Ai2’s Earth observation models into infrastructure for fine-tuning, evaluation, and large-scale geospatial inference.
Microsoft’s FY26 review argues that customers are shifting from AI experiments to broader deployments, with governance, data, and agent management becoming central concerns.
OpenAI’s new post describes agentic AI as a way to automate parts of scientific computing, with an emphasis on workflows where models plan, code, run, and revise analyses.
Microsoft says MAI-Cyber-1 Flash can help identify and fix security weaknesses, but questions remain about how AI security agents are controlled in practice.
Amazon is reportedly scaling back several Nova models while moving resources toward a Frontier Model Research group led by Pieter Abbeel. Existing Nova customers keep support for now.
PJM Interconnection plans to curtail large data centers during shortages starting in 2027, highlighting how AI-driven electricity demand is straining grid planning.
Grafana Labs expanded its AI assistant so operators can ask natural-language questions across metrics, logs, traces, cloud data, and issue trackers from more than 30 sources.
Latent Space examines how OpenAI is turning Codex-style agent workflows toward knowledge workers, after ChatGPT Work and Codex reached a reported 10 million users combined.
Google updated Managed Agents in the Gemini API with Gemini 3.6 Flash as the default, environment hooks, budget controls, scheduled triggers, and free tier access.
Cursor Start gives developers in India local pricing, UPI payments, and access to cloud coding agents. The plan lowers the entry cost for regular AI-assisted development.
Nvidia and Microsoft launched an open AI security alliance with several partners, but without OpenAI, Google, or Anthropic.
AWS describes how Deepgram uses IAM temporary delegation for SageMaker AI support, cutting initial support-ticket investigation from days to minutes.
It is working with Cadence and Synopsys to optimize electronic design automation tools for its Vera CPU as chip designs grow more complex.
A new arXiv cs.CL paper proposes agentic evaluation for copyright-law compliance, aiming to assess legal behavior in language-model outputs beyond static benchmarks.
The Trump administration’s AI policy is being pulled by multiple camps, making the debate less a two-sided fight than a contest among many priorities.
It is joining Microsoft, IBM, SpaceX and others in an Open Secure AI Alliance focused on open-source tools for AI safety and security.
Task-aware knowledge compression can pre-compress large knowledge bases for analytical AI work that strains traditional retrieval-augmented generation systems.
Security needs to be rethought for the AI era, with attackers and defenders both using automated systems that change how organizations detect and respond to risk.
New research finds ChatGPT users taking on tasks across job roles, suggesting AI is changing the boundaries of work rather than affecting only narrow technical jobs.