Auto-download daily
Issue · June 6, 2026

Top of Hacker News

25 stories of the day: AI, technology, business and simply interesting discussions — recapped and unpacked.

Not just a headline — the discussion unpacked

The top HN threads: facts from the discussions, both sides, insider voices. Email, Telegram, or RSS.

He tested every IP KVM in his homelab — one of them caught fire (jeffgeerling.com)

Jeff Geerling ran the cheap remote-control boxes that let you reach a headless machine's BIOS over the network. The thread out-tested the test: a YC founder wire-debugged the USB signal, and one commenter's NanoKVM burst into flames 15 minutes after plugging in.

🚀Tech ★ 260 · 💬 70 · Jun 6

He built a Hacker News with every AI story filtered out (elijahpotter.dev)

Elijah Potter shipped a stripped HN front page with the AI content removed. Half the commenters first misread the title as a font announcement — "Sans" — which says something about how saturated the feed has gotten.

🚀Tech ★ 158 · 💬 83 · Jun 6 · #AI

He fine-tuned a 7B model on 37 million words of 1990s Microsoft manuals for $50 (passo.uno)

Fabrizio Ferri Benedetti fed a small LLM three decades of out-of-print docs from Bitsavers and taught it to write like a 1995 SDK reference — full cross-refs, platform requirements, the works. The thread's best objection: the magic of old docs was never the prose, it was the depth, and you can't fine-tune that in.

🚀Tech ★ 181 · 💬 64 · Jun 6 · #big tech#Microsoft

ISS crew sheltered over an air leak NASA has been chasing since January (bbc.com)

Astronauts were told to take cover during repair work on a leaking Russian segment, then — within the same news cycle — told to return to normal operations. The leak isn't new: NASA reported in January that pressure had stabilized, while admitting it couldn't be sure the air wasn't just escaping somewhere else.

🌍Other ★ 372 · 💬 241 · Jun 6 · #space

Gemma 4's 12B model now fits in 6.7GB of VRAM with almost no quality loss (blog.google)

Google's quantization-aware training bakes 4-bit compression into Gemma 4 during training instead of bolting it on after. The result: a 12B model that runs on an 8GB laptop GPU, and a 3.2GB E2B that does audio, image and text on a Mac. Simon Willison ran it; the thread argued whether Google's own quant is even the best one.

🚀Tech ★ 310 · 💬 92 · Jun 6