« first day (1194 days earlier)      last day (2643 days later) » 

12:00 AM
But. I think they'll add it soon, considering the good response this is getting.
 
Ah yes, but do they know we want it?
 
They should -- Undo made a FR several months ago, and it's in the meta post
 
That ^
 
Oh, goodie. Didn't notice.
@ArtOfCode Did you ever figure out how the timestamps for /events are formatted?
I'm doing last_msg["time_stamp"] but it doesn't like it.
 
nope
 
12:07 AM
@quartata Oh, they've been thoroughly nagged. I'm told it's in the process.
 
@Undo Cool.
 
\o/
 
New metasmoke user 'nc4pk' created
 
I wonder if it's in v3 already
 
What's the criteria for the "core" role?
 
12:27 AM
"have had me tick a box next to your name"
 
Something I just noticed is that every SmokeDetector messages is multiline
It's just that chat doesn't render leading newlines
 
@quartata that's... interesting
@ArtOfCode Or v2.3, or whatever they call it
 
yeah, that one
 
@Undo Found out while working on my own chat client
Was wondering why Smokey's messages wrapped weird.
I assumed it's to get around the char limit
But judging by your response it's something that CE is doing
 
Think so
There was some hubbub around multiline messages a few months back, may have come from that
But no, they can't be multiline. Those don't render links and such
 
12:41 AM
Uh
Oh nevermind. I'm just stupid.
Blame Shoes for wrapping funny
 
@Undo what's up with condition 70 and not having a user?
 
Ah shit it was me!
 
Pretty sure that's from my testing and not cleaning up properly
 
Condition 70 was one I made and then deleted, or so I thought.....
 
@Undo User cascading destroy nullified the user field?
 
12:46 AM
@quartata oh, that's good to know. Let me look into that
(it shouldn't do that, so not your fault)
 
Wait, now I'm trying to remember what happened...
 
meh, easier for me to try to repro locally
 
Actually, no I don't think that's what happened. I think 71 was the first one I made, then for some reason the checkbox for autoflagging wasn't sticking so I went through the wizard and deleted the second condition it made (72)
It could be that I deleted condition 70 and 71 is the condition from the wizard which I then edited
 
that'd make some sense.
 
12:47 AM
fp- by Gothdo
 
@quartata this looks right. 71 is the first one that has your name on it, and 72 doesn't exist, presumably because you deleted it
 
OK.
Phew.
 
@quartata 15 users with max rep > 1
 
Oh OK plenty for the spampocalypse
 
Nuked #70 manually
 
12:50 AM
New metasmoke user 'ravron' created
 
Uh, who reported that?
 
obvious troll attempt
 
Right, but it's not spam.
 
But you can see who by going to metasmoke or sd why
 
Oh yes, right. Forgot
 
12:53 AM
sd why
 
@Seth [:35617397] Post manually reported by user 4castle from metasmoke.
 
you can report manually through ms now?
 
automatic report from FDSC
@SmokeDetector f
 
FDSC wouldn't do that for a VLQ flag surely
 
FDSC only does that for spam flags
 
12:55 AM
it does it for rude flags too, right?
 
Maybe
 
J F
@NobodyNada Yes
 
Unicode chars getting messed up.
 
(1) is probably a Unicode issue, (2) is us not rendering HTML entities properly.
 
12:58 AM
@SmokeDetector Oh hi! Neat.
 
Welcome, @doppelgreener!
 
ello ello ello
 
@NobodyNada Thanks. :) I heard of Smokey a couple of weeks ago when we enabled it for the RPG.SE chat room. Still figuring out the personal participation side of it.
 
Speaking of which did that ever work
I didn't see it posting anything
 
It works, but RPG barely ever get spammed
 
1:07 AM
@ArtOfCode Does Smokey respond instantly or does it take a few minutes before it evaluates & flags spam?
 
@doppelgreener it's not instant, but we usually report spam within a few seconds of it getting posted
 
@doppelgreener API requests are batched to save quota, so there's often a few minutes between batches.
 
We have a pretty consistent record of destroying spam within single-digit minutes, and our spam blocking mechanisms are quite strongly tuned to learn from what we flag.
 
I don't see sites that I joined after setting up flagging on the list.
 
@ArtOfCode I'm still not getting any emails from your blog
 
1:10 AM
@NobodyNada yeah, nor me. I have no idea why
 
Most of our spam also follows one of three predictable patterns: people testifying about a voodoo priest spellcaster who can reunite you, the lucky reader, with their divorced husband; someone advertising that you can truly become a vampire via mail-order; and questions full of Korean text and links repeated 10-20 times.
I think it's just down to: we don't have much spam get through for it to see, and I bet we could give Smokey a run for its money on it detecting spam vs us already removing it.
(... we'd probably lose at least half the time, but still!!)
 
Welcome, @hBy2Py!
 
Woah, fast-welcome!
Hi!
reading Glorfindel's intro bit
So, has this been taken care of? Will Smokey eventually delete the chat post when it's gotten enough feedback?
 
That was a false positive.
Normally Smokey deletes posts marked as false positives
But it was older than 2 minutes so we can't.
Because chat.
 
1:27 AM
Ahhk, got it. How many chatroom responses does Smokey wait for before taking action?
(or not)
 
Depends on what you mean by "taking action"
On a true positive it'll repost it to other rooms where it runs on the network
 
If you mean autoflagging, it doesn't wait for a response.
 
If you mean autoflagging though that's different
That happens immediately given the right criteria
 
I don't know what I mean, this is my first encounter with Smokey. :-D
@quartata Ahh - so that higher-rep users on the relevant sites can attack?
 
Actually so that it can just go away.
6 spam flags immediately deletes a post.
Although bringing it to the attention of mods is a plus too
 
1:30 AM
Is there any initiation process to go through before participating in training Smokey?
 
@hBy2Py So Smokey has two functions: reports, and autoflagging. The MSE post you saw is about autoflagging, but Smokey also just posts reports of posts it thinks are spam but isn't confident enough to flag automatically into this chatroom. That lets whoever's here evaluate the post, and flag it as spam if necessary.
 
<nod>, got that much.
 
[ SmokeDetector | MS ] Bad keyword with email in answer: Migrate Blog comments to Wordpress + Disqus by Alex on stackoverflow.com (@dorukayhan)
 
The heck?
 
tpu- by ArtOfCode
 
1:31 AM
@hBy2Py read this, then ask for privileges once you feel ready
 
Huh. That didn't onebox right.
Oh well.
 
How did @Art communicate the tpu- here without typing into chat..?
 
@hBy2Py Because chat isn't the only place to submit feedback
 
there's actually a bunch of stuff that's useful to read, like the wiki pages on Privileges, Feedback Guidance, and Commands
 
You can also do it from the review queue on Metasmoke or via the API
The API lets userscripts submit feedback
 
1:32 AM
We have a userscript called FDSC (Flag Dialog Smokey Controls) which automatically feedbacks when we flag
 
^
 
I was going to write an explanation, but they did it for me :)
 
:-)
Lots of enthusiasm here, I like it
 
Any time you see "xyz by person" it's from Metasmoke.
 
tl;dr: I flag, FDSC takes that as an implicit "yes, that was spam", and tells metasmoke to record that I think it was spam. Metasmoke sends that back to chat so that people know about it, because otherwise it would bypass chat entirely.
 
1:34 AM
<nod>
Well, cool, thanks for the welcome and the info, all. I may be back....
 
Anytime :)
 
I hope to see you later!
 
Huh this seems pretty cool
I forgot Python had difflib
Very cool
 
:)
 
Oh right, that's you. Forgot.
 
1:46 AM
blah
 
?
 
@ThomasWard \o
 
@Andy ohai
@quartata i'm weird, I make random posts/noise when I randomly appear from the darkness and shadowy mists
The room also gets a little darker too :P
@Andy is Docker good enough for you per @ArtOfCode's reply and other efforts on the GitHub, or do you prefer my approach?
o/ @NobodyNada
where's the coffee you told me you were going to get me :P
 
I hadn't seen your approach yet. I'm going to test dockers tomorrow and I can let you know
 
!!/coffee
 
1:48 AM
@quartata brews coffee for @quartata
 
Eh, good enough.
 
@Andy my approach is more or less the "Warn about git not being available in Windows, so don't support git functions and avoid the import failures"
untested but my Windows machine is kinda fubar right now so....
 
Could do Appveyor?
 
My friend @profkeithdevlin has reminded me that English is not the same as American. Here's an old guide to what we… https://twitter.com/i/web/status/834515307494531075
 
@quartata special-case development-environment Windows weirdness
not something that'd be launched for a Production-level env.
 
1:49 AM
I will test your changes, but looking through the diff, I like how simple it seems
 
@ThomasWard Yeah, Appveyor is a Windows CI.
I don't mean to replace Circle
Just spin up an instance to test it real quick
 
@Andy well I stole some of my functionality from my lxc-bootstrap scripts, which use if 'linux' in str(platform.platform()).lower() to determine if we're on Linux or not, but... it should work. Hopefully. Again, untested.
 
@ArtOfCode TIL, I speak "british" at work
 
@quartata I believe you. I'm just not the one who determines what env. we test on.
And this is more or less for @Andy to be able to 'test run' things in Windows without failing on the import sh problems (since the sh module doesn't support Windows, and never will)
It also avoids using 'no longer maintained' libraries which was also @Andy's concern.
@Andy though, the SO python people said to use the library either way, instead of trying to create our own wrappers :P
 
Translation: Andy is a small pain in the backside
That jerk
 
1:53 AM
@Andy also known as "professional conduct"
 
@ThomasWard use the pbs one?
 
@Andy yup
 
fp- by ArtOfCode
 
I would think the quote would make that not work
I suppose you ignore quotes though
@ArtOfCode Got the API calls I needed working
 
+1
 
1:57 AM
Although all I'm using is /api/posts/feedback/
So it should really be singular not plural
 
New metasmoke user 'Heptite' created
 
Welcome, @BrianRogers!
 
I wonder if we'll get to double the amount of users by the end of the day
We were at 70 users before
We're at 117
So 23 to go
 
@NobodyNada thanks
 
@ArtOfCode Apparently, I have nothing better to do than to fix Windows sh warning things, and do random Python code reviews. I think I need to get a life heh.
 
2:02 AM
I hope this doesn't crash Metasmoke
 
@quartata 74 if we're being technical
 
@quartata if it does, we're screwed.
Royally lol.
 
I just realized that this was significantly more data than I thought.
I'll add a backoff of 10 seconds per request.
 
@quartata metasmoke's not a weakling. 2 seconds would probably be fine, to be honest
 
All right.
I'm pulling every tpu-'d post ever just so you know
I'll do 5 just to be extra considerate
 
2:04 AM
that's only 46k or so :)
460 requests
we've probably handled 5 times that this afternoon
 
Yeah I know but I noticed the MS profiler said 4000ms when I refreshed so
 
not always the most accurate thing, but fair enough
also: if you see a big peak in failures tomorrow, ping me. I have a script I can run to set people's conditions down to only the sites they have accounts on.
 
All right, better now hopefully.
 
[ SmokeDetector | MS ] Bad keyword in answer, blacklisted website in answer, pattern-matching website in answer, blacklisted user: How do probiotics work? by user8305 on health.SE
tpu- by quartata
 
@ArtOfCode ooh, we don't want that knocking the introduction off the top of the starboard
 
2:06 AM
@ArtOfCode I thought Smokey wouldn't make accounts when autoflagging
 
@SmokeDetector why
 
Body - Position 534-552: dietary supplement
Body - Position 1601-1619: womenprobiotic.com
Body - Position 1606-1618: probiotic.co
Blacklisted user - blacklisted for //health.stackexchange.com/questions/11262 (https://m.erwaysoftware.com/posts/by-url?url=//health.stackexchange.com/questions/11262) by http:metasmoke
 
@SmokeDetector bah, already flagged
 
@quartata it won't; that's why it might fail
 
@NobodyNada it's still there
 
2:07 AM
I don't think that was spam
 
@NobodyNada so we can removeify a post from the starwall :P
 
@YvetteColomb yeah, there was a spam link at the end of it
 
@ArtOfCode but not without clicking "show 9 more"
 
took me a moment, but it's there
@NobodyNada how small is your screen? I can get 10 starred messages on mine with spare space
 
@YvetteColomb Doesn't disclose affiliation.
 
2:08 AM
I see 8 with my screen so heh
 
Also the website was blacklisted anyways.
 
@ArtOfCode ah yeh, I just went through MS. I was reading the post and it was deleted before I could see the link
 
pew pew it's gone
 
@ArtOfCode yeah, I see it in full screen
@quartata @Yvette that's the 3rd spam for that site today
 
@quartata yep I had a look in MS. It was deleted so quickly I couldn't see the link :)
 
2:08 AM
also: I'm pretty sure the nuke on that post was 100% us
 
@NobodyNada yep I'm up to speed now :D
 
@ArtOfCode It was autoflagged
 
wouldn't doubt it, @ArtOfCode lol
 
@SmokeDetector autoflagged?
 
@quartata That post was automatically flagged, using flags from: MAR, ArtOfCode, Floern.
 
2:09 AM
Cool beans.
 
@quartata aye, and there were at least two flags (FDSC) from humans
 
@ArtOfCode NOT a 100% us. You mean we are not the only people killing spam????
lol
 
@JF Problem with Coal: the cancel editing button is off screen
 
@ArtOfCode how dare the humans flag spam! :D
 
well, ya know, the communities occasionally get one or two flags in ;)
 
2:09 AM
On the default size
 
There were 5 failures on that post
is that normal?
 
@quartata press Esc
 
@ArtOfCode I know but I like buttons
And it's rather awkwardly cut off - I can see just the top
but I can't click it
 
@ArtOfCode it's a good sign, it means the community is active
 
@NobodyNada ouch. Okay, some people's conditions need editing, and we need to change that to not be a failure
 
2:11 AM
@ArtOfCode Where can I see failures per post?
 
@quartata /flagging/logs, or per-post flag logs at /post/12345/flag_logs
 
@quartata go to the post in MS and click the little flag next to the title
 
hey, at least it proves the try-fail-continue approach works
 
Doesn't say why it failed though
I suppose that's private
Since it gives away conditions and stuff
 
J F
@quartata Expand the window?
 
2:12 AM
ahahaha I love how the eligible flaggers are divided into "core" and "everyone else"
 
@quartata hover over Fail
 
I have to go; see you later!
 
@JF I mean, I could do that :) But just letting you know since that was the default size
 
conditions aren't super-private, anyway, they just aren't very useful information to non-admins
 
@ArtOfCode "Flag options not present" I didn't think the wizard let you enable autoflagging without a condition?
 
2:14 AM
@quartata flag options, not flag condition. The SE API didn't return any flag options when we requested them, because the user didn't have an account to flag with.
 
Oh. Gotcha.
That's not a big deal.
Huh, I only have accounts on 46 communities. That's less than I thought
@ArtOfCode A button that disabled all sites you don't have accounts on would be nice
But it would need API access for hidden communities
 
hm I notice I have no auto flags when I come on. There's a long break for hours. does it not auto flag until we do a manual flag for that day?
 
Has there ever been an fpu-?
 
@YvetteColomb gonna blame random distribution and the fact that we now have going on 60 flaggers for that
 
I can't imagine it but
 
2:16 AM
@quartata yes, but they're rare
 
@ArtOfCode What would they be for?
Trying to decide what class to put them in.
 
false positive, whitelist user
 
@ArtOfCode ok then we'll blame random distribution and all these newcomers taking our autflags ;)
 
Ohh.
I thought it would be false positive, blacklist user
I suppose that makes no sense
 
so... legitimate user who has a username that happens to be caught in a blacklisted username - fpu says "false positive, and don't report any more of their posts just based on their username"
 
2:24 AM
@ArtOfCode So just to confirm
For false positives the only ones I need to check are fp, fp-, fpu, fpu-
For true positives tp, tp-, tpu, tpu-, trueu, trueu-
And for NAA naa, naa-
 
might also find a few falses in there
what are you trying to do here?
 
Building a corpus.
 
of...
 
Processed meat with delicious glaze
 
categorised, I assume
does it have to be right up to date?
'cuz if not, I can get you a sanitized SQL dump of the posts table
which has is_tp, is_fp and is_naa fields, precategorised
 
2:26 AM
I mean no but that would probably be just as much effort to parse as this
@ArtOfCode /posts/feedback does give that but
 
what format do you need it in?
 
Obviously I have to pick the feedback first
 
InternetUnexplorer is that the anti-IE?
 
@ArtOfCode I mean ultimately JSON with a dictionary mapping sequences of word-word cooccurrence vectors to one of the three classes. I don't think you want to do that though :P
Just the post text to the class is sufficient but I'd rather do it myself since ultimately one of us will end up wasting time on it
And I already wrote code
 
I can do a SELECT body FROM posts WHERE is_tp = 1; if you like
export that as headerless CSV and bingo
 
2:30 AM
Would it be small enough to fit in a pastebin?
(or gist)
 
possibly
 
Matching words to spam?
 
lemme try it
 
@bwDraco Yeah. Making a magic 8-ball.
also known as a convolutional neural network
I know you guys have done ML before but I can't resist
@ArtOfCode Actually it would be best if you did the false positives first
Since those are the least common I want to limit the other 2 to that size
I could just weight the training but
Actually that would be better
There are less false positive examples than I thought
 
@quartata apparently it fits in a gist. click this link if you like crashing browsers
 
2:35 AM
Actually mine took it like a champ
but thanks for the warning
And thanks for the contents of it obviously :P
 
I also have tps, which is literally 30MB of the worst of the Internet. Let's see if GitHub ban me, shall we
 
Oh no no no
I hadn't thought of that
OK let's not do this
 
meh, it's unlikely. I'm more wondering if just pasting it into the text field is going to crash my browser
...which is entirely possible, it's now taken over a minute to start pasting
 
Hm, I just realized that 46,000 * 2 seconds is about a day
So maybe this is the best way
 
only 460 requests, though, at 100/req
 
2:39 AM
Oh yes right.
durr
 
never know, you might get there before I finish pasting it into a text field
 
Oh I have the code just not running it
 
I may be able to email you the file, if my server lets me upload 30MB in one go
 
If this is causing your browser physical anguish then don't botber
And no definitely don't email me a 30 MB file
 
see that's the thing, the tab's not crashing or anything, it's just... not doing anything
 
2:43 AM
I'll just roll with the API since 460*2/60 is 15 minutes
I've done scrapes that took much much longer
 
\o/ I managed to get GitHub to give me a 405 Not Allowed error :D
 
lol
 
How far back do the MS archives date back anyways
 
going on 2 years
58k posts, probably 1 year 7-8 months
 
Why does SE attract particularly dumb spam? The fact that they can figure out how to post would make you think they could do better
 
2:49 AM
shrug
 
@ArtOfCode a webhook for when something is reported would be awesome
 
@quartata 50k lines at a time (just concat the files): 0, 1, 2, 3, 4
 
oh dear
 
@quartata there's a websocket, no webhook
 
I think this is the first time I've thanked someone for spam
@ArtOfCode link?
 
2:54 AM
@quartata "somewhere in the API docs"
github.com/Charcoal-SE/metasmoke/wiki/WebSocket-API, may or may not have what you need
 
only sends feedback :(
 
can probably be poked to send reports too. I can do it tomorrow, if you drop an issue on GH that says what you need in the reports
 
[ SmokeDetector | MS ] Bad keyword with a link in answer, link at end of answer: Multiple Permalinks for Same Post by San Sun on wordpress.SE
tpu- by Yvette
 
I view this as a thank you excuse to advertise ergo spam — Yvette Colomb 11 secs ago
 
3:23 AM
[ SmokeDetector | MS ] Bad keyword with a link in answer: php form email not sending when action = submit to self by phuc bui on stackoverflow.com (@dorukayhan)
fp- by 4castle
 
Offensive username. Needs a mod flag.
or I could just ping @Undo
 
3:45 AM
Sounds like a valid Vietnamese name to me
 
Really? I'm not an expert on that language but say it out loud
 
I know what it sounds like, and yes it could be a troll. It could also very easily be legit.
 
Eh OK
...I feel bad now. Quick google confirms that
An unfortunate lack of accent marks
(in their spelling that is)
 
[ SmokeDetector | MS ] Bad keyword in body, blacklisted website in body, link at end of body, repeated URL at end of long post: How Does Aurora Deep Sea Reverse Aging Signs by amondjuri on gaming.SE
 
@SmokeDetector k
 
3:57 AM
sd autoflagged?
 
@quartata [:35620456] That post was automatically flagged, using flags from: Cerbrus, NobodyNada, avery.
 
Something I thought of is that if a user posts x (2-3) or more tp'd posrs we could automatically raise a mod flag
in hopes that they'll nuke it
 
@quartata by that point it'll already have brought other attention down upon them behind the scenes
 

« first day (1194 days earlier)      last day (2643 days later) »