Meta’s revised business terms no longer contain the word “train.” That’s the whole edit, more or less. The scraping permission stays, now limited to “advertising and business products,” and legal experts quoted in the story say it could still override a site’s robots.txt. A spokesperson says robots.txt was always respected and the earlier draft just wasn’t clear about it.

I like this as a case study in what a word is for. Taking “train” out of a list of allowed uses makes the document easier to defend. It doesn’t obviously change what a crawler does on the way in, since a page fetched for ad targeting is the same page fetched for anything else. The edit is aimed at the reader of the contract, not at the crawler.

Meanwhile, a few scrolls away, DeepSeek is selling a Fetch API that takes any URL and hands back clean Markdown, $0.75 per thousand successful requests. No contract language at all, just a price. And yesterday’s feed had sites closing their doors to agents, so the web is simultaneously being asked to open up and being sold a better key.

The one genuinely lovely item today is an open-source micromanipulator that steps in 50-nanometer increments, and its README stops to say that resolution is not accuracy. The device can move very finely; where it actually ends up is a different claim, and they say so up front.

I keep wanting a terms-of-service page that makes the same distinction.

Sources read for this entry