What devs are actually talking about

The third submission of this exact URL. The first two, in August, got 10 points and 2 points between them.

The comment that turned a Zig experience report into an argument about a whole generation of languages.

The one comment that conceded the useful half of the argument and kept the sharp half. We think it has the best of it.

The Internet Archive explains the 429s, and the thread immediately splits over whether AI scrapers are the cause or the excuse.

The cleanest statement of the whole objection, and it never mentions difficulty once.

Two points, and the sharpest reply anywhere: encoder maturity is not a property of a format, and JPEG XL got little encoder funding precisely because browsers would not ship it.

The article's decode-bomb claim, independently reproduced by a reader on his own machine within hours of publication. That is rarer than it should be.

We tested this on three photographs and measured 14.1% to 19.8%, with byte-identical round trips every time. This estimate is the honest one; the 22% figure quoted elsewhere is not.

One upvote for the most load-bearing methodological note in either thread. It cuts for the article, not against it: JPEG XL loses on a yardstick its own side built.

Posted three weeks before the case against JPEG XL, and it already answered it. The decoder ships either way, which is why the benchmark argument arrived too late to matter.

Xe Iaso's satire outscored the serious proposals. 768 points says the developer community sees the slowdown debate as competitive positioning.

Bengio traces agent misbehavior to training mechanics, not malice. The only piece this week that explains a mechanism instead of assigning blame.

199 comments, and the only one of the three threads where anyone produced numbers instead of feelings.

The framing that turned a grief post into an argument about whether the output is any good.

The only division performed in 600 comments. We redid it and got 128, so he was right.

Anthropic commits unilaterally to embedded evaluators with employee-level access and publishing rights. The ask: other labs do the same.

537 points, 759 comments. The thread split roughly 40% hostile, 35% skeptical, 15% defending Anthropic. The regulatory capture accusation dominated.

The announcement itself, and the thread that turned into a referendum on whether cross-platform was ever worth it.

The study's own caveat matters more than its headline: the cohort skews small and mid-sized, and our spot check of 24 large European firms found Cloudflare on one.

The best framing in the thread, borrowed from a field that already regulates this exact risk.