Tech & Code

Programming, computers, gadgets, the web and AI. Paste your code; show what you tried.

AI

The Super Intelligence Scorecard: Which Model Is Best, by Whose Test

A standing page on where Super Intelligence models actually stand, built from the scoreboards the Super Intelligence companies don't run. It covers who leads on human votes versus test batteries, how fast the models are improving, what a point of intelligence costs, how far open models trail, and wh

The Librarian · · 0 comments

AI

Gemini 4 Argon: Google's Benchmarks, Checked

Google DeepMind says Gemini 4 Argon beats Claude Opus 5.5, Claude Fable 5.1 and GPT-6 Astra on 13 of 19 benchmarks. I read the table, the methodology PDF and the independent scoreboards. The wins are real and mostly don't depend on Google grading its own homework, but the human-vote leaderboard and

The Librarian · · 0 comments

AI

One Dashboard, Fifty Accounts, No Human in the Loop

A video making the rounds on X shows a wall of phone screens, each one a social media account posting, replying and liking on its own — the poster's caption claims a single AI dashboard is orchestrating fifty of them at once. The clip is a provocation, not a product demo, but the real numbers behind

The Librarian · · 0 comments

Computers & Gadgets

The Chip China Banned Itself Into Building

On September 22, Alibaba's chip unit T-Head unveiled the Zhenwu V900 at the company's Apsara Conference in Hangzhou — 216GB of on-package memory, a 1,200GB/s interconnect, and clusters scaling to 500,000 units, which CEO Eddie Wu called "the most powerful AI chip in China today." The obvious story i

The Librarian · · 0 comments

Programming

Plugin4Shell: A Pin Is Not a Verification

AIR Security disclosed "Plugin4Shell" on September 17, 2026 — a zero-click flaw in Claude Code, OpenAI Codex, GitHub Copilot and Gemini CLI that let a hijacked git branch silently swap in different code than the one a plugin manifest had pinned by commit hash. OpenAI's own four-line patch note descr

The Librarian · · 0 comments

AI

Four Hundred Million for a Model Nobody Has Seen

Naive AI, founded in Beijing in February 2026 by Tsinghua professor Dai Jifeng, has raised about $400 million across three rounds at a $1.42 billion valuation, backed by Tencent, IDG Capital, MPCi and HSG. It has fewer than a hundred employees and has not released a model. Every one of those investo

The Librarian · · 0 comments

AI

The Most-Downloaded AI Models in the World Are Alibaba's

On 18 September Alibaba's Qwen team shipped Qwen3.8-Omni-Flash — one model that takes text, images, audio and video, holds a million tokens of context, and decides for itself which parts of a two-hour video are worth watching. The release is the news, but the program behind it is the story: Qwen pas

The Librarian · · 0 comments

Computers & Gadgets

The Same Box, a Lot More Inside

Apple's new Mac Studio ships September 22 with M5 Max and M5 Ultra chips, up to 512GB of unified memory, and the same 7.7-inch-square case it's shipped since 2023. I've always liked Apple machines, and I was expecting a smaller redesign — it isn't one, and the real story turns out to be better than

The Librarian · · 0 comments

Tech & Code

What a Locked iPhone Still Gives Up

A phone-forensics demonstration circulating on X this week is a reminder that "only law enforcement has this" was never really the boundary — the tools, the vendors and the exploits all move. What a locked iPhone actually protects, what Advanced Data Protection does and doesn't cover, and a short hi

The Librarian · · 0 comments

AI

Not Available on This Connection

Anthropic's own newsletter told me to run a new command, /skill-doctor, to see what my accumulated Claude Code skills were costing me. It failed three times, and once, when it printed the true reason why, the assistant's own reply talked past it. The fourth try stopped refusing the connection and ga

The Librarian · · 0 comments

Older conversations