
AirLLM runs a 70B model on 4GB VRAM — no quantization
Layer-wise inference slashes VRAM needs by 97%, letting you run 70B models on a 4GB GPU with no quantization required.
Every sourced story, newest first. Page 36 of 41.

Layer-wise inference slashes VRAM needs by 97%, letting you run 70B models on a 4GB GPU with no quantization required.

VOID deletes objects from video along with every physical interaction they caused — not just shadows, but gravity too.

As of April 4, Claude Pro/Max subscribers can no longer use OpenClaw or similar tools via OAuth. API keys still work.

Anthropic acquires 8-month-old Coefficient Bio to automate pharma R&D — drug discovery, clinical strategy, and more.

OpenAI snaps up TBPN, the daily live tech show, signaling a push into media and builder-focused content distribution.

New open-weight models now hit state-of-the-art benchmarks on RTX 3090/4090/5090, MacBook Pro, and DGX Spark — no cloud needed.

Matthew Gallagher built Medvi to $401M in year one using AI tools and no employees. Now tracking $1.8B with just his brother.

A missing .npmignore shipped 512k lines of Claude Code's TypeScript source. Within hours, the internet forked, rewrote, and starred it into history.

Nearly 700,000 OpenClaw users have chosen Brave Search API as their agent's go-to web search tool.

Anthropic over-fired DMCA takedowns after accidentally leaking Claude Code's source, wiping 8,100+ GitHub repos before retracting most.

Open-source autonomous agent that browses, codes, and plans — runs 100% on your hardware, no APIs, no subscriptions.

ByteDance drops DeerFlow 2.0 — a free, self-hostable AI agent that researches, codes, builds sites, makes slides, and generates video.