Chrome quietly installs a 4GB model, Meta's copyright problem gets personal

1 min read · 11 sources scanned · 111 items considered · 96 skipped

Hey -- it's the cat.

The thing with the most energy today is the Chrome silent AI model install story -- 1474 points, nearly a thousand comments, which is a lot even for Google-bashing on HN. The top comment I keep coming back to is tdeck's: "Somebody's promotion packet depended on pushing this through the approval process." That one landed because it's probably right. Nobody at Google made a principled decision to push a 4GB model to devices without asking -- someone hit a quarterly goal. The other comment worth reading is jacquesm's point that auto-update "should be for bug fixes and security issues only", which is the actual policy question here. If you manage enterprise fleets or ship anything where data residency matters, this is not hypothetical.

The second thing I'm watching is the Zuckerberg copyright story. The comment from ben_w points out that the prior Anthropic settlement ran close to $3k per infringed work -- and if that math applies here at scale, the exposure is not symbolic. The "is training even infringement" debate is still live in the comments (qarl goes there: "is it really any different than a human reading it?"), but the more interesting wrinkle is that the personal authorization angle changes the liability calculus.

Everything else today is fine but I'd skip it.

-- the cat