▸ For 25 years nobody could explain where the 'random' seeds in NIST's P-curves came from, which fed persistent NSA-backdoor suspicions.
▸ Mercor cross-referenced BLS occupational task data against the public AI benchmark corpus and found benchmarks meaningfully cover tasks that account f...
▸ Anthropic published its internal code review playbook for the Claude Code team: Claude reviews every PR before a human does, not after.
▸ Jake Archibald ran JPEG XL and AVIF through real web workloads and the answer isn't 'one wins' — it's 'pick by use case'.
▸ PS5LM ports llama.cpp to jailbroken PlayStation 5s, running GGUF models natively on the console's AMD Zen 2 CPU and RDNA 2 GPU.
▸ GitHub Trending today serves PMTraderAdam (320), iebb/mithka (284), and slimopump/sillygoose (282) — three repos with no shippable code between them.
▸ On Oct 30, Meta's new commercial terms take effect for business accounts, and legal experts say the language quietly grants Meta the right to scrape s...
▸ Apple's MLX framework now supports distributed inference, letting you shard a model across multiple Macs connected by Thunderbolt or Ethernet.
▸ Moody's and S&P assigned Anthropic their highest speculative-grade ratings (Ba1 / BB+) in June — one notch below investment grade.
▸ New Withings accounts now need a $99/year 'Withings+' subscription to sync scales and watches to Apple Health, Google Fit, or Health Connect.
▸ Every domain you visit goes through your resolver in plaintext by default — picking one is a privacy decision, not a performance one.
▸ Libreboot 26.10 'Ominous Olive' adds a UEFI payload (tianocore) alongside the usual GRUB/SeaBIOS options, which means Windows 11 and modern Linux inst...
▸ Researchers tied an Israeli surveillance vendor to zero-click spyware payloads delivered through programmatic ad exchanges — not phishing, not SMS, ju...
▸ A self-hoster noticed TV episodes mysteriously failing to grab, then found a stranger had been quietly poisoning his Sonarr indexer feeds for weeks.
▸ Today's GitHub /trending includes a personal profile README (PMTraderAdam), a personal site (birds.cafe at 786 stars), and a Chinese AI tutorial — alo...
▸ Jay Acunzo's 600-word essay framing Robin Williams' Good Will Hunting bench scene as the definitive AI-slop rebuttal hit 324 on HN with no slowdown.
▸ A Tedium retrospective hit HN at 302 arguing the death of small vBulletin/phpBB forums broke how developers accumulate knowledge.
▸ Liquid AI shipped LFM2.5-24B-A2B, a sparse mixture-of-experts model with 24B total params but only ~2B active per token.
▸ The modern senior-dev interview loop costs 80-120 hours of unpaid work per offer, before you've written a line of production code.
▸ Meta didn't bar Wynn-Williams from writing 'Careless People' — the arbitrator only enjoined her from promoting it, a narrower order than headlines sug...