📰 News
AI News — page 2
Funding, launches, analysis, and interviews.
Hebatron Hebrew Language Model Outperforms 24B Parameter Competitors Using Sparse Architecture
Researchers released Hebatron, a Hebrew-specialized language model that achieves 73.8% reasoning accuracy while activating only 3B parameters per inference across a 30B parameter sparse mixture-of-experts architecture.
OpenAI Launches Daybreak Cybersecurity Initiative to Challenge Anthropic's Glasswing
OpenAI introduced Daybreak, a cybersecurity program combining GPT-5.5 models with Codex security to automate threat modeling and vulnerability remediation, directly competing with Anthropic's Project Glasswing.
Google unveils Googlebook laptops with integrated Gemini AI for fall 2026
Google announced Googlebook, a new line of AI-powered laptops featuring deep Gemini integration, launching fall 2026 with hardware partners including Acer, Asus, Dell, HP, and Lenovo.
Hypercubic AI launches Hopper agentic development environment for mainframes
Hypercubic AI released Hopper, an AI-powered development environment that lets developers use natural language to navigate mainframe systems, write COBOL, and debug z/OS jobs.
Google launches Gemini-powered dictation for Gboard, challenging AI voice startups
Google's new Rambler feature brings AI-powered voice dictation to its Android keyboard app, directly competing with emerging dictation startups like Wispr Flow and Typeless.
Anthropic launches Claude for Legal plug-ins to compete with Harvey and Legora
Anthropic expanded its Claude for Legal platform with new plug-ins and connectors for document review, case law research, and deposition prep as competition intensifies in the AI legal services market.
Google launches Create My Widget for natural language Android customization
Google's new Create My Widget feature lets Android users build custom home screen widgets through natural language prompts, launching this summer on Samsung Galaxy and Pixel devices.
Threads tests Meta AI integration mimicking X's Grok functionality
Meta's Threads is beta testing an AI feature in five countries that lets users mention Meta AI for real-time context on trends and breaking news, similar to X's Grok chatbot.
Gigacatalyst Launches AI Builder for SaaS Customer Customization
Gigacatalyst debuts an embedded AI platform that lets non-technical users build custom workflows and features inside existing SaaS products through natural language prompts.
Obsidian launches AI-powered plugin review system after 120 million downloads
Note-taking app Obsidian introduces automated reviews for its 4,000+ community plugins using AI agents to scan code quality and security vulnerabilities.
Voker Launches AI Agent Analytics Platform for YC S24 Cohort
Voker, a Y Combinator Summer 2024 startup, has launched an analytics platform that helps companies track AI agent performance through automated intent classification, correction detection, and resolution measurement.
Amazon employees 'tokenmaxxing' to hit AI usage targets on internal leaderboards
Amazon workers are automating unnecessary tasks with the company's MeshClaw AI tool to inflate their token consumption statistics after the company set targets for 80% of developers to use AI weekly.
Dessn raises $6M to build AI design tools for production codebases
Dessn secured $6 million in Series A funding to develop AI-powered design tools that work directly with production codebases, eliminating handoff friction between designers and developers.
Researchers Expose Critical Flaws in AI Safety Self-Play Training Methods
New research reveals that standard self-play red teaming for AI safety collapses to self-consistency, failing to create adversarial pressure needed for robust defense training.
AI Dialogue System Boosts Emergency Room Diagnostic Accuracy by 25%
New research shows interactive AI assistance improved emergency medicine residents' diagnostic accuracy from 58.9% to 73.4% on difficult cases through iterative physician-AI dialogue.
Researchers Create Test to Measure LLM Developmental Cognition Capabilities
New research introduces a 20-item assessment tool to evaluate how large language models understand and respond to different stages of human cognitive development.