All Updates
Page 714 of 1669
May 6, 2026
Meta AI Upgrade Thwarts Age Bypass Tricks
A child tricked Meta's online age-verification using a fake mustache. To counter such bypasses, Meta is revamping its tools with an AI system. The AI analyzes images and videos for visual cues like height and bone structure.
Google Shuts Down Project Mariner
Google has discontinued Project Mariner, an experimental AI feature for autonomous web tasks. Shut down on May 4th, 2026, its technology is now integrated into other Google products like Gemini Agent. First revealed in December 2024, it was updated to handle up to 10 tasks simultaneously.
Robot Monk Gabi Joins Seoul Temple
A robot named Gabi has become the newest monk at a temple in Seoul, South Korea. It humorously vowed not to overcharge while seeking enlightenment. The story blends robotics with Buddhist tradition, questioning if it's meditating or rebooting.
Validating Non-Deterministic Agentic Behavior
GitHub shares a method using dominatory analysis to validate agentic behavior in Copilot Coding Agents, where 'correct' outcomes aren't deterministic. This 'Trust Layer' avoids brittle scripts and black-box judgments. The approach is detailed in a GitHub Blog post.
Musk Considered Altman for Tesla Board
Elon Musk once considered offering Sam Altman a seat on Tesla's board of directors. This was revealed to jurors during the trial over the feud between the two AI leaders. The disclosure highlights past potential collaboration amid their rivalry.
Snap-Perplexity $400M Deal Dead
Snap's $400 million deal with Perplexity AI has fallen through. The planned AI search feature for Snapchat never fully rolled out.
Pentagon Botches Alibaba-Baidu Blacklist
The Pentagon mishandled the blacklisting of tech giants Alibaba and Baidu. This error provides insight into internal US administration conflicts on China policy. It underscores the fragility of the Trump-Xi truce.
Apple's $250M Siri Delay Settlement
Apple is paying a $250 million settlement for delaying Siri features promised years ago, citing AI capabilities unavailable at the time. Affected iPhone users may receive up to $95 per device as compensation. The case highlights challenges in delivering advanced AI integrations on schedule.
Anthropic-xAI Partnership Doubles Claude Limits
Anthropic has partnered with SpaceXAI to double the 5-hour rate limits for Claude AI. This extension allows teams to use advanced models more extensively for engineering and operations tasks. Higher usage limits enhance productivity for AI-dependent workflows.
Microsoft May Shelve 2030 Clean Energy Goal
Microsoft is considering dropping its 2030 clean energy target due to surging power demands from AI. The company's AI initiatives are significantly lifting overall power usage. This highlights tensions between AI growth and sustainability.
Anthropic Launches Dreams for Claude Agents
Anthropic has debuted Dreams, a new feature for Claude Managed Agents. The dreaming capability reviews past sessions to restructure agent memory. It includes outcomes for setting task criteria and multiagent orchestration for complex tasks.
Qwen3.6 Quantized Excels in VS Code Coding
User reports excellent results using Q5_K_XL quantized Qwen3.6-35B-A3B on AMD RX 7900 with llama.cpp Vulkan server. Integrated with VS Code Copilot for generating full React apps and Playwright tests from large prompts. Shares detailed server command, sampling params, and chat config.
Core42 Leases 20MW Minneapolis Data Center
Core42, the AI infrastructure arm of UAE's G42, has leased 20 megawatts in a converted Minneapolis office turned data center. This marks G42's expansion into US AI compute infrastructure. The deal positions Core42 as anchor tenant, bypassing Silicon Valley hyperscalers.
Atlas Open-Sourced for GB10 Ultra-Fast Inference
GB10 Solution's Atlas inference engine is now fully open source in Rust+CUDA, delivering 100+ tok/s on Qwen3.6-35B-FP8 with MTP. Optimized for DGX Spark Blackwell GPUs, it supports OpenAI/Anthropic APIs and runs via simple Docker command.
Google Speeds Gemma 4 3x via MTP Drafters
Google has made Gemma 4 models three times faster using MTP Drafters. This employs speculative decoding, pairing a heavy main model with a lightweight drafter for token pre-generation. The models now run efficiently on consumer GPUs and edge devices.
Maket Frees Draw from Scratch for All
Maket has made its Draw from Scratch canvas free for all users. This tool allows creation of residential layouts and 3D visualizations without a paid plan. It democratizes AI-powered design tools for broader access.
Google Favors AI Licensing Over Consultancy
Google is pursuing licensing agreements instead of building consultancies like OpenAI ($10B) and Anthropic ($1.5B). This strategy targets enterprise distribution through private equity portfolio companies. It may prove smarter for capturing the next big AI channel post-cloud.
Scale AI Lands $500M DoD Contract
The Pentagon awarded Scale AI a $500M contract for AI-assisted data analysis and decision-making. Scale is backed by Meta Platforms. This advances US military's AI integration efforts.
White House Eyes Pre-Release AI Reviews
The New York Times reports the White House is considering government reviews of AI models before release, a reversal from Donald Trump's pro-AI stance. David Sacks, the AI and Crypto Czar, appears to have failed amid this policy shift. The story highlights tensions in Trump's AI agenda.
Microsoft May Delay Clean Energy Goal for AI Data Centers
Microsoft is accelerating global AI data center expansion but considering delaying or scaling back its ambitious 'hourly clean energy matching' target. This goal is one of the industry's most aggressive emissions reduction commitments. Bloomberg reports internal discussions on the matter.