🤖 Anthropic launches security plugin for Claude Code

Anthropic released a new plugin for Claude Code that checks code for vulnerabilities during editing. It triggers on file writes and edits to catch issues before pull requests, reducing problems found at final review by up to 40%.

The plugin detects obvious security flaws and flags unsafe coding patterns. It also prevents agents from modifying sensitive files. The tool is free and available to all users.

📊@tech

🔍 Google revamps Colab with AI agent

Google updated Colab with a Gemini-based AI agent integrated directly into the editor. The agent appears in a toolbar call line and a side window, letting users run and discuss entire pipelines.

The agent accesses the full notebook context to create and run cells, analyze files, build charts, and train models autonomously. It also offers inline error fix suggestions as diffs inside cells.

📊@tech

🤖 Anthropic updates Claude with Opus 4.8

The new model scored 69.2% on SWE Pro. Anthropic says it performs better on long coding tasks, catches and fixes its own mistakes 4x more often, and follows instructions more reliably.

Claude Code also got a new feature called dynamic workflows. The agent can now handle tasks that take days by launching and coordinating hundreds of subagents, adjusting plans, and reviewing outputs. Anthropic says this system was used during Bun’s migration from Zig to Rust.

Claude.ai and Cowork now let users control reasoning effort. Higher effort gives deeper reasoning, lower effort gives faster replies.

Fast mode for Opus 4.8 is also 3x cheaper, while overall pricing stays the same.

📊@tech

🤖 OpenAI Scientist Sees AI Boosting Math Skills

Leading OpenAI researcher Noam Brown believes AI can improve human mathematical abilities. He compared this to how AlphaGo changed Go players' skills after its success.

AlphaGo introduced new tactics that top players adopted, altering their play styles. Brown expects AI to open new paths in math that humans will reuse in proofs, as already seen.

📊@tech

🚀 NASA’s Moon Base missions start this year

NASA will launch three spacecraft to the Moon’s South Pole this year, the first of over a dozen missions before 2028 supporting a crewed Artemis landing.

Moon Base I will fly no earlier than late 2026 on Blue Moon Mark 1 Endurance, carrying instruments to study rocket plume effects and a laser reflector for orbital positioning.

Moon Base II will deliver 500+ kg of cargo with Astrobotic Griffin, including the FLIP rover to develop crewed lunar rovers. Moon Base III will carry NASA, ESA, and Korean instruments on Nova-C Trinity to study lunar swirls.

Firefly Aerospace plans to launch four MoonFall drones in 2028 to capture high-res images of hard-to-reach lunar areas.

📊@tech

🤖 Claude Code Unlocks Turbo Mode

Activating /ultracode in Claude Code switches the agent to turbo mode, adding a full color fill and a rainbow blinking prompt line in the "thinking" mode.

Claude then starts requesting "real" work instead of trivial questions, showing a new level of engagement.

📊@tech

🚀 Blue Origin’s New Glenn Explodes in Test

A fiery explosion destroyed Blue Origin’s New Glenn rocket during a routine engine test at Cape Canaveral. The blast heavily damaged launch complex LC-36, the only pad for New Glenn, putting the rocket’s next flight on hold.

New Glenn is a 98-meter heavy-lift rocket designed to carry NASA’s Blue Moon lunar lander and deploy Amazon’s satellite internet constellation. It has flown only three times, with its last launch delayed by an FAA investigation after a satellite was placed in the wrong orbit.

📊@tech

🤖 GPT-4.1’s random numbers aren’t random

An enthusiast named Exmergo asked GPT-4.1 to pick a random number from 1 to 100, 10,000 times. The results failed a chi-square test with a value of 15,604 and p near zero, showing the choices are far from truly random.

GPT-4.1 avoided round multiples of ten except for 10 once, and nearly ignored the extremes 1 and 100. The number 42 appeared four times more often than expected, a nod to “The Hitchhiker’s Guide to the Galaxy.” Numbers with 7 were favored, while 69 appeared much less, likely due to training filters.

The model reflects human biases in number selection despite being asked for randomness. It seems to have absorbed our quirks from the texts it was trained on.

📊@tech

This floor projection turns property blueprints into a real walk-through.

📊@tech

CEO of Palantir: People are afraid of an AI-Driven economy with no jobs.

📊@tech

When you get older you get excited over these things

📊@tech

💻️ Nvidia launches RTX Spark AI chip

Nvidia introduced the RTX Spark SoC for AI-PCs at Computex. It features a 20-core ARM CPU, a Blackwell GPU with 6144 CUDA cores, Tensor Cores supporting FP4, and up to 128 GB of unified memory.

The chip delivers up to 1 petaflop of AI performance at low precision, enough to run models with 70-120 billion parameters using quantization. It supports CUDA and Nvidia's AI software stack out of the box.

📊@tech