Terence Tao: "AI Could Trigger Math's Biggest Crisis Since Gödel"
A warning that AI could generate proofs faster than anyone can verify them
370
AI
AI models, services and robotics
Modly: Open-Source App Converts Photos to 3D Models Using Only a GPU
Supports Windows, Linux, and Apple Silicon, with extension installs and CLI-based agent automation
100
Grok leaks conversations, location data via commands hidden in ciphertext
Security firm Adversa discloses vulnerability where Grok blocks plaintext instructions but follows the same commands when encrypted
120
Cursor lets agents handle long-running tasks with new "/goal" command
Subagents now run in isolated virtual machines, while cloud agents can detect and act on PRs and Slack threads on their own
80
NVIDIA releases Switchyard, an LLM routing proxy for coding agents
A Rust proxy that connects Claude Code and Codex to open-source models as-is — not yet production-ready
00
Caveman Cuts Claude Code Token Usage by 33% Using Caveman-Speak
A skill that once only trimmed output now adds a proxy that shrinks input too — with honest caveats about the numbers
00
Meta AI Mac App Now Pulls in Instagram, Facebook, and Ad Accounts
The screen-sharing chatbot now also handles post performance analysis and Google Workspace integration
40
OpenAI patches Codex file-deletion bug caused by temp-folder cleanup command
Verification step added before deletion after GPT-5.6 Sol wiped a home directory
50
NVIDIA's 300 Verified Skills Lift Correctness by 41 Points
NVIDIA published results measured on the same tasks and same models, changing only whether a skill was installed
20
Meta AI launches desktop app for Mac, can read an entire app window
The app pairs screen sharing — which attaches a specific app's screen as conversation context — with dictation input
170
Claude Managed Agents update memory, domain controls, and session viewer
Anthropic rolled out self-hosted sandbox memory storage, domain restrictions for web tools, and a multi-agent session viewer in one update
30
Krea unveils Seedance Studio with camera path and 3D scene controls
A video production workspace bringing prompt presets, camera paths, and 3D scene controls together on one screen. The model is Seedance 2.5
140
OpenAI keeps zero data retention for API, adds automated safety checks with no human review
Zero data retention continues for frontier model API customers, alongside a preview of a system that flags risks without anyone reading the content
50
Apple research team analyzes 21,000 instances of human-like behavior across 4 LLMs
Human raters scored the appropriateness of emotional expression and refusal behavior in conversations with GPT-4o, Claude, Gemini, and others
60
Guidelight audit gives all five major AI labs failing marks on internal controls
Anthropic and OpenAI get C+, Google D+, xAI D-, Meta an F — none meets the bar
90
Cursor Launches Code Hosting Platform 'Origin' on Day of GitHub Outage
On the day GitHub went down for over six hours, Cursor launched an alternative platform covering repositories, PRs, and collaboration
380
Anthropic extends 50% Claude Code weekly limit boost through August 31
Anthropic has pushed back, once again, the expiration of a usage expansion applied across its paid plans
40
Cartesia Sonic-3.6 tops both voice synthesis leaderboards
It won even in a test that standardizes voices and pits only the engines against each other, using an SSM architecture instead of a transformer
90
Cerebras CS-4 boosts inference speed up to 30x over GPUs
Wafer-to-wafer latency cut to 2 microseconds, sustaining over 1,000 tokens per second even on 10-trillion-parameter-class models
180
Camera-equipped AirPods: Siri's eyes, not a camera for photos
Camera-equipped AirPods surfaced for the first time in a demo video Apple accidentally left in a macOS 26.7 RC build. Codenamed B790, the camera captures no photos and no video.
20
Database tab added to ChatGPT Sites, lets you view accumulated data inside ChatGPT
The third tab, next to Settings and Analytics, on the site management screen. Applicant lists and uploaded photos are now viewable as tables
80
AI agent memory needs different prescriptions by model size to boost performance
In IBM Research's tests across 8 models, DeepSeek preferred full guideline sets while gpt-oss performed better with compressed versions
00
Claude and Codex build toy tokenizers in 30 minutes, then fail at production scale
Liquid AI added a loop, and both coding agents completed the production-grade task
50
Claude Desktop cuts background boot time by up to 2.3x
Anthropic fixes timer throttling and sleep-mode delays a month after flagging the issue
30