Go 1.27's experimental simd package ran a float32 dot product 10.6x faster than a plain loop on our M4 Pro. Forced into emulation, the same code ran 1.9x slower than the loop it replaced.
6 min
An open decision model went from 2,824 to 26,781 stars in under eight days. We ran it over 140 of our own articles: 64.3% right in 45 ms each, and the confidence score was the best part.
A Claude Code engineer floated killing plan mode on X and drew 3,122 replies and 1.6 million views. We ran plan mode in a scratch repo: it blocked a file write and let a sqlite DELETE straight through.
Signal's Android repo has 20 commits building phone number free registration. We read them, and the one that matters wires the whole thing to a Google Play one-time purchase.
Homebrew 7.0.0 landed on 13 September and moved Intel Macs to Tier 3. Buried in the Hacker News thread is the more interesting admission: the maintainers built the Rust rewrite everyone asks for, measured it, and found it slower.
Armin Ronacher spent $1,200 and 35 hours on a GPT-6 Astra experiment that produced 75,000 lines of code he called worthless. A researcher measured AI code as twice as verbose as human code. Hacker News hit 746 points asking to limit AI news. The backlash has numbers now.
bzip3 hit the Hacker News front page with 412 points and shipped version 1.5.4 six hours later. We built it, benchmarked it against xz and zstd on 100MB of Wikipedia text, and reproduced the silent data loss bug the release quietly fixed.
A claim that 89.6 percent of CDN-using European companies sit behind Cloudflare drew 242 points and 205 comments in a day. We fingerprinted 24 large European company sites ourselves and found Cloudflare on exactly one of them.
The VMware Virtual Disk Development Kit vanished from Broadcom's developer site around 25 August. We checked all three documented URLs and the SDK catalog ourselves: 44 SDKs still resolve, VDDK returns 404 on every path and is not listed anywhere.
Terence Tao's 6 post thread on AI-generated proofs drew 126 favourites. The Hacker News thread that followed it ran to 94 comments, most of them about whether orchestral musicians have side gigs.
Zero replies, and it describes the lock-in lever already in use. Only 2 of 418 comments mention the AAGUID.
The most replied-to comment in the thread, and the metadata we checked backs it: 54 of 62 passkey managers are not in FIDO's list.
94 of the 418 comments argued about attestation. FIDO's own metadata does not list Apple's or Google's passkey managers at all.