All Updates
Page 1736 of 1959
March 5, 2026
FlexAttention Adds FlashAttention-4 Backend
PyTorch now supports FlexAttention with a FlashAttention-4 backend on Hopper and Blackwell GPUs for faster attention computation. It introduces automatic generation of CuTeDSL score/mask modification functions. Additionally, JIT-instantiation enables custom FlashAttention-4 configurations.
Oracle Job Cuts Amid AI Cash Crunch
Oracle Corp. is set to cut thousands of jobs as part of measures to address a cash crunch. The financial pressure stems from massive investments in AI data center expansion.
Anthropic CEO Restarts Pentagon Talks
Anthropic CEO Dario Amodei is re-engaging Pentagon officials after a failed $200M DoD contract. The Pentagon turned to OpenAI instead. The company seeks a new cooperation framework.
Anthropic Renegotiates DoD AI Contract
Anthropic is renegotiating its AI contract with the US DoD after being labeled a supply-chain risk over ethical disputes on mass surveillance and autonomous weapons clauses. CEO Dario Amodei is engaging DoD officials and investors like Amazon and Nvidia are lobbying to resolve tensions. Industry groups warn that the risk label could limit government access to top US AI tech.
US Drafts Sweeping Nvidia Sales Controls
The Trump administration is drafting rules granting the US sweeping authority over Nvidia's global sales. Nvidia, the dominant AI chipmaker, faces potential formal regulatory oversight. This positions the government as a key player in the AI hardware industry.
Google Labels Battery-Draining Android Apps
Google is rolling out warning labels in the Play Store for apps using excessive battery due to high background activity. These target apps abusing partial wake locks, with gradual rollout over coming weeks. Affected apps may also lose visibility in recommendations.
Apple Blocks ByteDance Apps in US
Apple began blocking US users from downloading or updating ByteDance apps for the Chinese market. This action started in January following TikTok's US operations transfer deal announcement. It targets apps designed specifically for China.
Alpha School's AI-Powered Ideal Education
Alpha School leverages adaptive AI software for personalized, mastery-based learning, enabling students to achieve top 1% U.S. scores in just 2 hours daily. It reverts to Socratic tutoring via AI, distinct from chatbots like ChatGPT, using standard curricula. Incentives and project time build self-driven learners and vital life skills like critical thinking.
Pentagon-Anthropic Feud Exposes AI Surveillance
Anthropic's clash with the Pentagon highlights the US government's lightly regulated purchase of commercial data like browsing histories and location info. AI is increasingly used to analyze this data at massive scale for surveillance. The feud draws attention to ethical and oversight gaps.
NVIDIA CCCL Adds FP Determinism Control
NVIDIA introduces controls for floating-point determinism in CCCL, ensuring multiple runs with identical inputs yield bitwise identical results. This addresses non-determinism in FP operations critical for reproducible HPC and ML computations. Developers can now enforce consistent behavior across runs.
GitHub & Andela Scale Global AI Learning
GitHub and Andela collaborate to scale AI opportunities worldwide. Developers linked to Andela integrate AI tools into real production workflows. Key learnings shared on GitHub Blog.
Cursor Launches Agentic Automations Tool
Cursor is rolling out Automations, a new agentic coding tool. It enables automatic launching of AI agents in the coding environment, triggered by codebase additions, Slack messages, or timers.
Roblox Deploys AI Chat Rephrasing
Roblox is rolling out AI-powered real-time chat rephrasing to make messages more civil. It transforms offensive phrases like 'Hurry TF up!' into 'Hurry up!' while preserving user intent. All chat participants are notified when rephrasing occurs.
PNW Leader Fuels Battery Boom
Grayson Shor, founder of the Pacific Northwest Battery Collaborative, is the driving force behind uniting the region's battery community. The initiative is energizing the Pacific Northwest's growing battery industry.
Meta Sued for AI Glasses Privacy Breach
Meta faces a lawsuit over privacy violations in its AI smart glasses. Lawyers allege marketing promised user control over footage sharing, but subcontractors reviewed sensitive customer videos including nudity and sex. An investigation exposed the discrepancies.
Hidden AI Risk in Portfolios
Arjun Raghavan, CEO of Partners Capital managing $75 billion for families and foundations, identifies AI as the largest risk factor in markets currently. He appeared on Bloomberg Open Interest to discuss technology trends and opportunities in private credit.
Google Pledges $50M to Fight Superpollutants
Google commits at least $50 million through 2030 to eliminate superpollutants like methane and refrigerants, joining Amazon and Salesforce in a $100M initiative. These gases cause nearly half of planetary warming and break down faster than CO2. The pledge equals about three hours of Alphabet's annual profit, amid rising AI data center emissions.
Anthropic CEO Pursues Pentagon AI Deal
Anthropic CEO Dario Amodei may still seek a deal with the Pentagon. A prior $200 million DoD contract failed due to disputes over unrestricted military access to Anthropic's AI. Negotiations highlight tensions in AI safety and defense use.
OpenAI Bolsters Safety with Police Alerts
OpenAI agrees to enhance safety protocols post-high school shooting, including notifying law enforcement of suspicious ChatGPT use. Canada will involve privacy, mental health, and police experts in high-risk case reviews. The company pledged a report on new protocols and potential retroactive reviews.
Google AI Search Adds Yoruba, Hausa Support
Google has expanded AI search to support Yoruba and Hausa for Nigerian users. Speakers can now access AI Mode and AI Overviews in their native languages, enhancing local usability.