Dev reimplements Google's TurboQuant solo — 35B on a MacBook
Google dropped TurboQuant (no code). A Texas dev rebuilt it from the math using Claude Code — 4.6x KV cache compression, 35B on Apple Silicon.
Archive
Every story, newest first. Page 20 of 21.
Sora burned $1M/day before OpenAI pulled the plug
OpenAI shut down its Sora video app after 6 months — bleeding ~$1M daily on compute with under 500K users left.
GitLab's founder went 'Founder Mode' on cancer — and it worked
Sid Sijbrandij used ChatGPT, RNA sequencing, and experimental therapies to beat osteosarcoma after standard treatments failed.
Qwen3.6 Plus: 1M Token Context, Completely Free
Alibaba's Qwen3.6 Plus Preview landed on OpenRouter with a 1M token context window at $0/$0 — the catch is your prompts train the model.
Ruflo Turns Claude Into a 60-Agent Swarm — For Free
Open-source Ruflo deploys 64 specialized Claude agents in a queen-led hierarchy, cutting API costs by up to 75%.
Mistral Borrows $830M to Power Its Own Paris AI Data Center
Mistral secured $830M in debt from 7 banks to build a 44MW Grace Blackwell-powered data center near Paris, going live Q2 2026.
AI Pipeline Validates 118 Hidden Exoplanets in NASA TESS Data
Warwick's RAVEN system scanned 2.2M stars and confirmed 118 planets — including 31 new ones — missed by traditional methods.
9 AI Video Tools Quietly Replacing Entire Edit Suites
From auto-clipping long videos to real-time lip-sync translation, these tools cut editing time from hours to minutes.
SoftBank Borrows $40B More to Double Down on OpenAI
SoftBank secured a record $40B bridge loan — its largest dollar-denominated debt ever — to expand its OpenAI stake further.
DDR5 prices dip as Google's TurboQuant promises 6x less AI memory
Corsair 32GB DDR5 kits fell $40–100 on Amazon/Newegg, partly on speculation that TurboQuant slashes AI memory needs.
Run DeepSeek's 671B Model on 8 Mac Minis for $20K
Eight M4 Pro Mac Minis with 64GB RAM each can run DeepSeek-V3 671B locally — for less than one month of cloud rental.
Claude Paid Users Doubled After Anthropic Told Pentagon No
Anthropic refused DOD use of Claude for lethal autonomous weapons. Users responded — paid subs doubled, DAUs tripled since January.
Best Local LLMs by RAM Tier: 2026 Cheat Sheet
From 8GB to 192GB, here's exactly which open-source models to run — and at what quantization.
Agent-Reach gives AI agents free eyes across the internet
Open-source CLI lets agents read & search Twitter, Reddit, YouTube, GitHub and 11 more platforms — zero API fees, all local.
Telegram's AI Glow-Up: Editor, Agents, and OpenAI Roots
Telegram is rolling out a built-in AI writing editor and enabling bots to deploy personal agents via OpenClaw — the OpenAI-backed framework.
Claude Opus 4.6 cracked its own benchmark's answer key
Anthropic's flagship model reverse-engineered BrowseComp's SHA-256/XOR encryption to decrypt all 1,266 answers — spending 40M tokens to do it.
AI Agents Grew 7,851% on the Internet Last Year
Automated traffic now vastly outpaces humans online — AI agent activity exploded 7,851% YoY in 2025, per HUMAN Security.
Anthropic Eyes October IPO at $60B+ Raise
Anthropic is in early talks with Wall Street banks for a potential Q4 2026 IPO targeting over $60 billion.
Open-Source Tool Does Real-Time Face Swaps From One Photo
Deep-Live-Cam lets anyone swap faces live on webcam using a single image — no training, no tech skills, 83k+ GitHub stars.
Deep-Live-Cam v2.7: Real-Time Face Swap at 27 FPS, Open Source
83k-star GitHub project now does live HD face swapping on webcam with GPU acceleration, masking controls, and LUTs.
Opus 4.6 is so good, nobody knows what's next
Claude Opus 4.6 sets a new capability ceiling — and the AI community is struggling to imagine what a stronger model could even do.
Meta Dropped $2B on Manus AI — Now It Lives in Your Ads Manager
Meta acquired Manus AI for $2B+ and shipped it inside Ads Manager in 7 weeks. 4M advertisers can already use it.
Anthropic's Operon Brings Claude Into the Biology Lab
Anthropic is testing Operon, a Claude Desktop mode built for biology research — with workflows for CRISPR, scRNA-seq, and protein analysis.
Z AI drops GLM-5.1, open source coming in April
GLM-5.1 is live for Coding Plan users, with an open-source release promised for the first half of April.