OpenAI Hands Three AI-Made Proofs to Human Referees
OpenAI says its internal models proved results on Kakeya, the Exponential Time Hypothesis and Yang-Mills. Outside experts are now checking all three.
Webitlio: news on tech, AI and games
Featured
Latest
AI
Gemini 4 Argon Is Here: $2 Now, $4 After November 30
Google's Gemini 4 Argon is open to developers, businesses and paid subscribers. What it claims, which benchmark figures are Google's own, and what it will cost.
Mobile
Google Play Gets a Private Shelf for Apps You Vibe-Code
Google Play now lets adults install Android apps they built in AI Studio with no developer account. The apps stay private, unreviewed and visible only to you.
AI
Why So Many Enterprise AI Pilots Never Reach Production
An AI demo can impress in a week, but real data, permissions and budgets take far longer. What stands between pilot and production, and how to test for it.
AI
Google Wants Gemini to Run the Work, Whatever Model Does It
Google's Gemini agent keeps a company's tools, memory and rules in one place and can hand jobs to Claude. Swapping the model is easy; the platform may not be.
Technology
ArtCraft Adds Open-Source Word, Excel and PowerPoint Clones
ArtCraft's developer is adding WordCraft, GridCraft and DeckCraft to his free Rust suite. The official page lists them as in development. What that means.
Technology
Torvalds Uses Claude Code on a Linux Page Cache Speedup
A new branch in Linus Torvalds' repository speeds up small reads of one file from many threads, and its cover letter credits Claude Code. What it means.
AI
Why Cheaper AI Models Can Still Drive Compute Demand Up
OpenAI and Anthropic just cut prices on their newest models, yet total demand for chips and power may keep climbing. Here is how both can be true.
Security
Containing AI Agents: NVIDIA's OpenShell and Sentry
NVIDIA says agent safety rules must live outside the model. We look at OpenShell, Sentry and the Hugging Face breach behind them, and at what is still unproven.
AI
Midjourney Adds a Thinking Option to Rerun Images
Midjourney is testing a "Rerun (Thinking)" button on its Alpha website to improve prompt accuracy, text and coherence. How it works, and what is still unknown.
AI
OneRail Uses Nvidia AI to Pick the Cheapest Delivery Mode
OneRail's OmniSTAR uses Nvidia software to choose, order by order, the lowest-cost delivery option that meets the promised service. The speed claim is 10 times.
Technology
Torvalds Uses Claude Code on a Linux Page Cache Speedup
A new branch in Linus Torvalds' repository speeds up small reads of one file from many threads, and its cover letter credits Claude Code. What it means.
Games
Your First Prestige: What to Expect and When to Reset
Resetting your progress on purpose sounds wrong. Here is what prestige really does, what you keep, and how to know when the time is right.
Why So Many Enterprise AI Pilots Never Reach Production
An AI demo can impress in a week, but real data, permissions and budgets take far longer. What stands between pilot and production, and how to test for it.
Security
Containing AI Agents: NVIDIA's OpenShell and Sentry
NVIDIA says agent safety rules must live outside the model. We look at OpenShell, Sentry and the Hugging Face breach behind them, and at what is still unproven.
Mobile
Samsung Reportedly Cuts Q4 Phone Output as Memory Costs Soar
Korean reports say Samsung asked suppliers for 20–30% fewer phones in Q4 as memory prices climb. Samsung has not confirmed the numbers. Here is what we know.
