Version 0.2: seen before, how sure you are, and karma for its age
Version 0.2 of the add-on passed Mozilla’s review and is the one you get from Firefox Add-ons now. If you already have SynthStamp, Firefox updates it on its own.
The first version did one thing: it let you tag an account and showed you what others had tagged. This one adds three.
Seen before
A story that did well in one subreddit comes back a month later under another name, word for word. A picture does the same, resized. Until now SynthStamp could only tell you about that if somebody had tagged the first copy.
Now it tells you on its own. When a post, a comment or a picture was posted earlier by another account, it gets a small mark: “Seen before”. Click it and you get where it was first seen, by which account, and a link to open that earlier post.
It works the other way round too. If you are looking at the earlier copy, the mark says it was posted again later, and by how many accounts. And if several of the accounts that posted it are already tagged as bots, it says so.
One thing about the wording, because it matters. The mark says “seen before”, never “the original”. SynthStamp only knows what people with the add-on have read. The first copy it saw may itself be a repost of something nobody with the add-on ever opened. Every one of these marks carries that sentence.
Short lines are left alone. “This.” and “Came here to say this” are posted by thousands of people, and marking them would only teach you to ignore the mark. A longer text that many different accounts have posted is marked as that, a text posted by several accounts, with no claim about who was first.
Say how sure you are
The first version had two buttons, bot and human, and a third for “not sure”. Most of the time the honest answer sat between them: this reads like a bot, but I would not swear to it.
So the buttons are now a scale with four steps: human, maybe human, maybe bot, bot. Pick the one that fits. A “maybe” is counted and shown as a maybe. It does not turn into a community mark on its own, so one person’s hunch cannot label an account for everyone.
Naming the kind of bot (AI, repost, spam) is still there and still optional. So is tagging a single comment or picture without saying anything about the person behind it.
Karma for its age
This one came from a comment under a suspicious post. Someone had looked up the poster and written: “271k in 1 month”. That is the sort of thing you can check on Reddit by hovering over a name and waiting, which nobody does for every post.
The add-on now does that lookup for the poster of a thread and the first few top comments, and for any account whose menu you open. When the figures stand out, they are written next to the name, plainly: “271k karma in 1 month”.
Three things it does not do. It does not tag anybody: a figure is a figure, and plenty of real people have a lot of karma. It does not guess: both numbers are Reddit’s own. And it is not hidden: there is a switch for it in the add-on’s menu, and the privacy page lists it.
Also in 0.2
Picture tags work on real Reddit now. The first version could not read pictures from Reddit’s image hosts, so it never fingerprinted one. That story is in the post about the first day.
What it means for what is collected
Recognising a repost needs the first copy to have been seen. So with page sharing on, the add-on sends the text of the posts and comments on the Reddit pages you open, a fingerprint of each picture (never the picture), and now the karma and age of the few accounts it looks up. The full list is on the privacy page, you see it before the add-on does anything, and one switch turns the sharing off.
If you use Reddit in Firefox, get the add-on.