Auto-download daily
Issue · October 17, 2026

Top of Hacker News

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

Claude traced NIST's mysterious ECDSA curve seeds to a 1990s random number generator (anthropic.com)

Zero bits in the top of several published seeds were the tell: they matched lrand48, which returns only 31 bits of a 48-bit state. Claude brute-forced the hidden bits for the binary curves and used lattice reduction for the prime ones, dating the prime-curve seeding to between May 1997 and April 1998.

The HN thread skipped the cryptography and argued about who is allowed to run this kind of analysis.

🚀Tech ★ 692 · 💬 274 · Oct 17 · #AI#science#security#Anthropic#Claude#ECDSA+1

Not just a headline — the discussion unpacked

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

Home Assistant's maintainer called a password scan 'won't fix', then apologized on Hacker News (github.com)

The check lives in the Supervisor, hashes values locally and sends only a 5-character SHA-1 prefix to HIBP. It has existed for years; a recent release merely made its warning visible, which is how a user noticed it.

The maintainer's apology didn't end the argument. Readers objected to the premise: a home automation hub that phones out, even with a prefix, is not what they installed it for.

90% of US working hours fall outside what AI benchmarks measure, Mercor finds (mercor.com)

Of 18,797 tasks in the US occupational database, only 924 have at least five benchmark prompts behind them. Software engineering and financial analysis are heavily tested; truck driving and gas plant operation barely appear.

Readers argued about whether that gap means safety for physical jobs or just a longer wait for robots.

Claude writes 90% of Claude Code, so its team rebuilt code review around risk tiers (claude.dev)

Anthropic says merged PRs for Claude Code more than doubled from June to September while reverts fell by over 40%. A classifier now sorts changes into no-human, one-human or two-human review, and an agent's approval never counts as a human's.

Hacker News mostly doubted that throughput measures anything.

Waymo picks Toronto as its first Canadian market, with riders in 2028 (waymo.com)

Waymo is partnering with Toronto's Element Fleet Management to run the fleet and says the work will create hundreds of local roles. It cites 400 million fully autonomous kilometres and claims 95% fewer serious-injury crashes than human drivers.

The thread was dominated by riders from other cities saying the same thing: it just works.

Jake Archibald benchmarked JPEG XL against AVIF, and AVIF won most of the web (jakearchibald.com)

Archibald tested avifenc, libjxl, jixel and cwebp on an M4 Pro, scoring with SSIMULACRA2. JPEG XL lost on lossy photos and screenshots (30-90% larger), but dominated lossless and can recompress existing JPEGs 9-20% smaller.

Commenters disputed whether the benchmark or Chrome's earlier decision was the real story.

🚀Tech ★ 204 · 💬 144 · Oct 17 · #dev tools#web#AVIF#Chrome#Jake Archibald+1

eBay now demands FCC IDs for radio gear, but exempts ham-band-only transmitters (arrl.org)

The ARRL asked eBay to clarify, and eBay confirmed devices that can operate only on amateur frequencies are exempt. Two-way VHF/UHF radios able to transmit outside ham bands still need FCC certification.

The thread's real question: why ham radio operators police each other so hard.

🌍Other ★ 176 · 💬 138 · Oct 17 · #culture#hardware#policy#ARRL#eBay#FCC

Four M5 Ultra Mac Studios linked over Thunderbolt 5 run an 816B-parameter model at 36 tokens a second (machinelearning.apple.com)

Apple's MLX team tested tensor parallelism over RDMA on Thunderbolt 5: four machines decode at 36.73 tokens/s at 8,192 tokens of context, against 15.74 with two-machine pipelining. A single M5 Ultra is about 1.6x faster at decoding than the M3 Ultra.

🚀Tech ★ 269 · 💬 123 · Oct 17 · #AI#big tech#hardware#Apple#Mac Studio#MLX+1