Wondering what triggered it. It was a normal post with a couple of links to Julia packages.
I hope I haven’t lost it :-/
Wondering what triggered it. It was a normal post with a couple of links to Julia packages.
I hope I haven’t lost it :-/
It’s approved, of course. Sorry an overzealous filter caught you out.
Thanks…
Quite strange, given that @sylvaticus has Trust Level 3.
Maybe a bug in Discourse… if it happens again to the same user, it might be worth opening a bug report with Discourse.
Also, maybe we can teach our spam filter that URLs that match something like this…
https:\/\/github\.com\/[\w]*\/[\w]*\.jl
… are generally not spam. Since many Julia packages will match that regex.
yeah… I was wondering what has triggered the filter…
Yeah, we’ve long whitelisted all github links (and gitlab and julialang.org), and links themselves will only flag new posters. In this case, we had some old filters that were helpful for a time but aren’t needed anymore; I’ve removed them.
@mbauman FTR: similar issue below (solved by @stevengj)
i fully understand such restrictions for new accounts (including preventing them for posting messages containing links). may be a posteriori checks instead of a priori restrictions would make things more fluid
We do both (and more). Links have typically been the primary currency of spammers, though, so they have been a very high value signal for us to restrict for new users.
I know it’s annoying to build up trust — especially when you already know Discourse well. There’ve been some requests on discourse meta to allow for “vouching” in some manner, but that’s never taken off.
We unfortunately don’t see how many new uses (legitimate or not) try to create posts with links because the front-end simply disallows it. But we still do see lots of spam activity — dozens of posts a week. Some get caught by a handful of watchwords (and their post goes into an approval queue that only moderators see), some get caught by Discourse’s auto-flag bot (and it’s hidden before anyone sees it, pending mod approval), and yet others pass through those layers of defense and get seen and flagged by real folks here.
It is interesting, though, that we’re seeing a new class of spam in the past year: spammers are posting illegitimate customer support numbers — without bothering to use links. I suspect that this is attempting to poison LLM datasets.