What SynthStamp does today, and what comes next
SynthStamp has changed a good deal in two weeks, so here is the whole of it in one place: what it does now, what it will not do, where it falls short, and what is next.
What you see on Reddit
A flag next to every username. Faint when nobody has tagged the account, coloured with a count when people have. It is the same flag on every thread that account appears in.
A menu behind the flag. Click it and say what you think, on a scale: human, maybe human, maybe bot, bot. If you pick the bot side you can name the kind (AI, repost, spam, or a declared bot such as a subreddit’s own helper). If you pick the human side you can add that the person uses AI to write, reposts, or promotes something. None of the kinds is required.
Tags on a single post, comment or picture. Sometimes the account is fine and one post is not. You can tag the text or the picture alone, without saying anything about who posted it.
What is known, at the top of the menu. How old the account is and how much karma it has, by Reddit’s own figures. How many of its posts and comments SynthStamp has seen and since when. How many people tagged it, and as what.
What it recognises without anyone tagging
Text that was posted before. A post or comment that another account posted earlier is marked “seen before”, with a link to the earlier one. Small edits do not hide it.
Pictures that were posted before. Each picture is turned into a fingerprint in your browser. Resized, recompressed and mirrored copies still match. How that works.
Karma that stands out. For a thread’s poster and its first few top comments, the add-on reads the account’s karma and age from Reddit. A young account with far more karma than its age explains gets the figures written into its flag, in red. Very high karma at any age gets them in amber.
What it deliberately does not do
It never decides who is a bot. Every mark is a count of people or a figure from Reddit. When people disagree, you see both numbers.
It never hides or removes anything on its own. You see the flag and make up your own mind. There is a setting to fold away comments the community has tagged; it is off unless you turn it on, and a folded comment opens with a click.
It never says “the original”. It knows what its users have read, and says so each time.
It never uploads a picture. Only the fingerprint leaves your browser.
It does not turn karma into a tag. A high figure is shown as a figure.
Where it falls short
It only knows what its users have seen. A repost is recognised if someone with the add-on read the first copy. Today that is a small number of people, so most reposts go unrecognised. This improves with every install and with nothing else.
Crops. A picture cut down to a part of itself has a different fingerprint. Heavy edits do too.
A post that is rewritten. If a post or comment is changed a lot after it was tagged, the tag stays with the old wording and does not show on the new one. The account’s tags are not affected. Marking “this used to say something else” is on our list.
Firefox only, for now. The add-on runs in Chrome and Edge in our tests, but it is not in their stores yet.
New Reddit changes under us. Reddit’s pages are not built for add-ons, and when Reddit changes its markup something of ours can stop showing until we catch up. Two days of that are written up here.
What comes next
- Version 0.2.1 to Mozilla: the fixes and the tidier menu from the last two days.
- Chrome and Edge store listings.
- Firefox for Android, once it has been through a phone.
- Better karma thresholds. The current ones were set from the first 64 accounts looked up. They will be set again from a larger sample.
- Edited posts. Saying so when a tagged post no longer reads the way it did.
- More sites. Reddit is the first. Each new one will be announced before it is switched on.
If you want to help
Install it and read Reddit as you normally do. That alone makes reposts easier to catch for everyone. And when you see a bot, say a bot.
Get the add-on, or read what it collects first.