« first day (1635 days earlier)      last day (2185 days later) » 

4:07 PM
@Mithrandir bah, class names. You can F12 the form to make it work if you like, changing Domain to SpamDomain, or you can wait for next deploy
good to go when you are @Undo
scratch that
now it is
 
[ SmokeDetector | MS ] Potentially bad keyword in title: Apple TV not saving localStorage on app exit by Nguyễn Thành Lộc on stackoverflow.com (@Zoe)
fp- by ByteCommander
 
@SmokeDetector why
 
@ByteCommander Title - Position 1-45: Apple TV not saving localStorage on app exit
 
How is that a bad pattern?
 
Need that regex tracer thing
 
4:11 PM
Fruit spam
 
@ArtOfCode done
 
ta
 
@Undo oh right I was going to add that to Halflife
since it can be as slow as it wants
 
[ metasmoke ] ci/circleci success on 23e2d3a: Your tests passed on CircleCI!
 
Yep, and if it only runs on already-caught stuff it's not high volume anyway
 
4:14 PM
[ SmokeDetector | MS ] Few unique characters in answer, repeating characters in answer: Prestashop Controller - Change the current store on Multistore by bulanDev on stackoverflow.com (@Zoe)
tpu- by ByteCommander
[ SmokeDetector | MS ] Bad keyword in link text in answer, potentially bad keyword in answer: Do comments slow down an interpreted language? by brolly on stackoverflow.com (@Zoe)
tpu- by WELZ
 
4:33 PM
@ArtOfCode 404s for me; has it been nuked?
Oh, I don't have permissions?
 
...huh
You should do
 
> app/controllers/abuse_reports_controller.rb:83:in `verify_access'
return if current_user == @report.user || current_user&.has_role?(:admin)
I have to be an admin it seems
 
yeah, that should've been excepted
fixed pending deploy
 
J F
waffles
 
more waffles
 
[ SmokeDetector | MS ] Potentially bad keyword in answer: How do I apply a clipping mask on certain objects in Illustrator? by user121111 on graphicdesign.SE
tpu- by WELZ
 
gofres
 
!!/test zenith-itech.net
 
> Potentially bad keyword in body, potentially bad keyword in title, potentially bad keyword in username
----------
Title - Position 1-17: zenith-itech.net
Body - Position 1-17: zenith-itech.net
Username - Position 1-17: zenith-itech.net
 
Still alive
 
4:38 PM
[ metasmoke ] ci/circleci success on 36d92d9: Your tests passed on CircleCI!
 
0
Q: Is it possible to block spammers based on country of origin?

sv.Do we know the country of origin of the recent spammers? If yes, can all IP addresses from a single country be blocked for a few weeks to discourage these spammers from posting? The usual flagging and downvoting has not been working effectively from what I can tell. So what else has been done i...

 
heh
that could cause a few problems
 
[ SmokeDetector | MS ] Manually reported question: Zoe should kill heself by Zoe is a cunt on area51.meta.SE
tpu- by Makyen
tpu- by Makyen
 
zzzz
one sec
ok we're good
 
5:02 PM
tpu- by WELZ
tpu by Makyen
tpu by Makyen
 
@ArtOfCode done
 
and yet
none of it has feedback
lol
 
we could make this a "how many deploys can I make Undo do in one day" sort of day ;)
 
the more deploys I do the higher chance you have of getting a CI job for it
 
It's been a while since I've last posted, but did someone open an abusive ticket against SmokeDetector?
 
5:05 PM
Do you and undo do undo done deploys
 
@bwDraco Context?
 
tpu by Makyen
 
#2152 on GitHub.
 
oh, yeah. Happens occasionally, trolls with no life
 
5:07 PM
@Undo [tag:malicious-compliance]
 
@bwDraco Not against smokedetector, against the user that created that issue
 
Hmm...
 
review page that top post won't go away - even when feedbacked a bunch of times (also AIM has no data)
 
tpu by quartata
tpu by WELZ
 
5:10 PM
yeah that's a bug
 
Manually delete?
 
@ArtOfCode @Undo Ummm.... The tpu feedback button on this MS post is broken. I haven't tried the FP button. Feedback from FIRE also didn't get recorded. When opened in a new tab it shows a "Routing Error".
 
tpu by WELZ
 
well part of the problem is the UID
 
I'm gonna stop clicking that
 
5:10 PM
isn't recognized
 
@ArtOfCode Email in the smokey@ inbox, any thoughts on how we should handle?
(mostly 'you or me'?)
Or Andy or angus, if they're so inclined
 
> Me and you, but mostly me.
 
I can if you like
 
go for it
kinda a fun opportunity to brag
 
5:15 PM
yup :D
 
...I just came in to say that Medium would be reaching out to yuou, but appears you saw already
 
Indeed they did
 
> Thanks, this list is great. We're going to reach out to that email so we can discuss more.
 
@Undo Is that what this was ?
 
@ArtOfCode You taking this one?
 
5:20 PM
(Also, did they tell you my real name? Feel free to yell at them if they did.)
 
aye
 
Indeed. Reads like a polite "Please tell us more so we can accurately target the pallet of bricks"
 
@Mithrandir nope, "one of your members"
 
wonderful
 
@Undo use triangulation
 
5:22 PM
@ArtOfCode BTW, here is 1254 spam posts. gist.github.com/superplane39/e7ec5236c5452d6c857a5194d5a53446 feel free to give that to them, and I'll start a new one
 
I also wonder if they're interested in hijacking Smokey code for their own uses. That'd be pretty neat.
 
probably less likely
what's valuable is the list of domains we have
 
^ That's what I'd want in their position too
 
domains/regexes/code/whatever
 
the means of discovering those domains requires humans which they probably are not interested in
 
5:23 PM
@Mithrandir The bright red "report gist" link is mocking me. It's literally a list of spam domains...
 
They might realize they just stumbled upon a source of those domains that's updated for free
 
(but then again the 150 I linked 'em earlier didn't get nuked yet, soo.)
 
@Andy it's secret tho
 
@Andy no, it's not, actually ;)
it's a list of medium.com posts that link to spam domains
 
[reports]
 
5:25 PM
Ah. A level of obfuscation. nods Carry on then.
 
5:35 PM
[ SmokeDetector | MS ] Potentially bad keyword in body: yelp search using Django by Enigma on stackoverflow.com (@Zoe)
Recovered from websocket._exceptions.WebSocketConnectionClosedException: Connection is already closed.
fp- by Makyen
 
response in
@Mithrandir wanna ping me an email to art@ and I'll forward you the chain so far?
 
sure, one mo'
sent
 
heh, pretty sure I've had your personal email for a while... I can just never be bothered to look through 5 different account address books to find people's addresses :)
 
!!/whois admin
 
Recovered from BrokenPipeError: [Errno 32] Broken pipe
 
5:47 PM
missing an argument
 
@WELZ Missing an argument.
@WELZ I hit an error while trying to run that command; run !!/errorlogs for details.
 
I know, I wanted to see what it'd say (because it isn't working)
 
Hey so uh
I have good news
For the first time ever
We might have an ML model on our hands that isn't garbage
I'm going to play with it more but with a "0.95" flagging condition this is what it looks like:
$ ruby liblinear/accuracy.rb 0.95
49193.0/50.0 = 99.89846272566659%
 
That's good; maybe time to revive Deepsmoke?
(like, the room)
 
Oh no
It's not a deep model nor is it for that kind of analysis
this is just a quicky logistic regression I slammed together for autoflagging, since the nets weren't working out
This one scores below autoflagging, which is a good sign to me: metasmoke.erwaysoftware.com/post/118971
This classic one also does, albeit just barely (0.91, which as a flagging condition is 99.26% accurate): metasmoke.erwaysoftware.com/post/24413
hmm
 
5:54 PM
@NobodyNada can you see the abuse report now?
 
@ArtOfCode Yep, thanks!
 
sweet
 
fp by Zoe
 
[ SmokeDetector | MS ] Few unique characters in answer: ! You can't use `\\eqno' in math mode by user162756 on tex.SE
tpu- by Zoe
Recovered from BrokenPipeError: [Errno 32] Broken pipe
 
5:58 PM
@SmokeDetector .826, which is only 96% accurate
(below 99.5% = not possible as a flagging condition)
 
!!/location
 
@NobodyNada ArtOfCode/EC2
 
Hmm...could there be network flakiness?
I’d expect an EC2 to have a reliable connection
 
The nice part about logistic regression is that it is really, really fast
So if this works out it'd be great
It's also pretty simple, hence the existence of a Ruby gem for it (most of the complicated ML things don't have Ruby libraries)
 
cheers @tripleee
 
6:01 PM
This gem is pretty good, it uses a C extension that looks like it wasn't made by monkeys
 
mostly avoiding discussing it here :)
 
ok :P
 
@ArtOfCode what is the accuracy of the default flagging condition?
 
100.00%
 
it's 280?
 
6:04 PM
yep
 
how many posts does it catch
 
41698 tp, 1 fp
 
oh so there is an FP
 
yep
 
That's a bummer, the most conservative one here only catches 34471 TPs
The most aggressive one, however, catches 53688 TPs
which I think is much better than the most aggressive all-site ones we have
0.95 (nice round number) is 49193.0/50.0 = 99.89846272566659%
I think the question is more what the FPs are like
that's what I'm trying to find out
 
6:09 PM
What threshold can we get 99.98% at?
 
running into some floating point weirdness, but about 0.988
 
well. time to create a third gist of medium stuff
 
38178.0/4.0 = 99.98952385941018%
The thing about it is
You get poorer performance if you try to go more conservative
 
Recovered from BrokenPipeError: [Errno 32] Broken pipe
 
But it catches a lot more TPs if you give it more leeway
 
6:13 PM
@quartata Aye, I'm just wondering where we'd put the default condition
 
Yeah it's tough
 
Ideally I'd like 99.98 because that's where the default originally was
What about 99.95?
 
BTW: Does anyone have the ability to whip up a script to GasMask Medium? Some of these have images that I'd prefer my parents not see
 
0.98 on the nose gives 41130.0/19.0 = 99.95382633842864%
That'd probably be decent?
Something to keep in mind is that this is not including any information about the number of reasons or rep in the flagging condition
This is in effect just the reason weight
However, it's not particularly trigger happy
that is it does seem to need 3 reasons to get the score high
 
@Mithrandir Does the gas mask do anything other than blocking images? You could also just use a regular browser extension for that, there should be plenty. For example, uMatrix gives great control, but it has a learning curve until you know what to click when to properly (un)block stuff.
 
6:18 PM
What we really should do is give these tools to iBug and let him go nuts
He finds the craziest flagging conditions
 
Have a link to an easy-to-use one for Chrome?
 
:P
 
@Mithrandir heh, yeah
 
Do you want to set up a bot or something that spits out the scores when they come down the ActionCable?
I'm too lazy
the only gem you need is liblinear-ruby
there's two scripts in there you can take a look at
one is evaluate which takes in an MS post ID and spits out the score
the other is accuracy which takes in a threshold and runs over the dataset
should give you the general idea
Also I did this using the bog-standard solver
Changing solvers might improve things, I don't know
This is pretty good though I'm happy it works
 
!!/watch vbucksunlimservice
 
6:25 PM
@ArtOfCode hm?
 
email
 
!!/alive
 
@Mithrandir Added vbucksunlimservice to watchlist
@Mithrandir You doubt me?
 
@ArtOfCode why is it that you're always the one who has to mediate this stuff
 
because I know it's there
 
Restart: API quota is 16299.
 
you might want to use an anonymizing proxy when browsing spam sites anyhow, I guess some will also offer image blocking
 
@ArtOfCode hm? I meant the flags coming out of CRUDE
 
@tripleee I'm not visiting the actual sites at the moment, just the Medium posts
 
6:28 PM
@quartata yeah, I know
 
i only look at spam sites after everyone else has gone to sleep
 
Restart: API quota is 16287.
 
they might have all kinds of embedded cruft, I'd certaily suggest playing it safe
 
at the moment, I'm not even letting the page load fully
 
... All manner of interesting stuff here but dogs insist on walkies now
 
6:31 PM
hi Charcoal folks! I've looked at the findspam.py code, but I can't accurately determine what "Bad pattern in URL body" indicates. I guess my regex-foo is not that strong. Does it simply indicate a url does not match a valid url format?
 
Zoe
IIRC it indicates URLs with a blacklisted keyword, but the URL itself isn't blacklisted
 
@Zoe no it's some specific things
what it means is
 
@quartata yah, looked at that exact line. as I said, I guess my regex-fu is not that strong :-)
 
6:33 PM
er hold up a second
I need to read this because I'm not entirely sure myself
 
in other words it trips if either of the regexes in patterns matches
 
@Art whaddya think - add to the first gist or start a new one?
 
@quartata said everyone who's ever looked at a regex they didn't write.
 
since the first was linked to 'em already
 
OK I got it
 
6:33 PM
@Mithrandir prolly new one
 
So it's like looking for things where "reviews" or "support" or the other patterns is in either the href or inside the <a> (like the link description), and there's a valid URL in there
does that make sensee?
 
yah, and 'Canada' is verbotten too?
 
I think that's like an optional thing
like I guess we got some spam that ended in reviews-canada/
 
ahhh
 
6:35 PM
whereas without that it would only catches URLs with a path of reviews/
the patterns seem to be focused on searching routes
"Bad path" would probably be a better name
 
that's from the -support at the end
of https://bitcoin.org/en/release/v0.10.0#watch-only-wallet-support
 
hmmm seems like that might be generate too many false positives to me.
 
incidentally that gets a score of .42 @ArtOfCode
 
Recovered from websocket._exceptions.WebSocketConnectionClosedException: Connection is already closed.
 
6:37 PM
sometimes reasons actually are more likely to indicate FPs it seems to have figured out hahahaha
it's doing classification for both obviously
@MaxVernon yeah, probably. I'd have to thumb through what it caught that are TPs
 
@quartata "FP"?
 
@MaxVernon false positive
 
ahhh
 
that wasn't really related, we're experimenting with a couple things right now
 
cool
 
6:38 PM
tpu by Nisse Engström on Zoe should kill heself [MS]
tpu by Nisse Engström on Zoe should kill heself [MS]
 
@NisseEngström those won't go away from the review queue because of an MS bug, just ignore them
 
@quartata Ah, thanks.
 
[ SmokeDetector | MS ] Username similar to website in answer: Get token balance with Ethereum RPC? by Furqan Siddiqui on stackoverflow.com (@Zoe)
fp- by Zoe
 
Zoe
@SmokeDetector Has disclosure
 
.48
Username similar to website needs some tweaking one of these days
 
6:42 PM
[ SmokeDetector | MS ] Username similar to website in answer: Can I transfer tokens from an ERC20 contract using PHP? by Furqan Siddiqui on stackoverflow.com (@Zoe)
Recovered from BrokenPipeError: [Errno 32] Broken pipe
fp- by ByteCommander
 
@SmokeDetector affiliation disclosed
@SmokeDetector what's this again?
 
Restart: API quota is 16141.
[ SmokeDetector | MS ] Bad keyword in title: Using `TStopWatch` without `free` by blackcanopus on stackoverflow.com (@Zoe)
fp- by Makyen
 
Zoe
!!/unnotify 11540 askubuntu.com
 
@Zoe I will no longer ping you if I report a post on stackoverflow.com, in room 11540 on chat.stackexchange.com
@Zoe I will no longer ping you if I report a post on askubuntu.com, in room 11540 on chat.stackexchange.com
Recovered from websocket._exceptions.WebSocketConnectionClosedException: Connection is already closed.
 
7:01 PM
wtf
so many exceptions today
 
Zoe
And since it's not normal, does that make today an exception?
 
Recovered from BrokenPipeError: [Errno 32] Broken pipe
 
At least the Zoe troll isn't creating troll accounts in metasmoke...
metasmoke account creation is back on, wouldn't want that troll taking advantage of it
 
7:18 PM
:facepalm:
 
[ SmokeDetector | MS ] Toxic answer detected: Disk on key is read only on every computer by aaron83 on unix.SE
naa- by ByteCommander
 
Now we're playing
 
Zoe
:44533642 del
 
@Undo keybase?
 
@quartata Hmmm... interesting that it considers hahaha a negative indicator. When I looked at it a several hours ago (& it hasn't changed since then) the hahaha regex was exactly 50/50 since it was added as a watch and >83% TP over all MS posts. Getting more complete matches data, including those posts not in MS, is one of the reasons to have a watch.
 
7:27 PM
[ SmokeDetector | MS ] Potentially bad keyword in title: Google Smart Home for Speaker or TV type devices by Robert Bartle on stackoverflow.com
Recovered from BrokenPipeError: [Errno 32] Broken pipe
fp- by Makyen
 
@SmokeDetector ignore-
 
@SmokeDetector What's this?
 
Restart: API quota is 15924.
 
Zoe
Yeah, that's me
 
Anyone done a large amount of video editing? I've been looking at Lightworks, but the community version doesn't support video exports higher than 720p (and the license prices are more than I want to pay for a hobby). Suggestions?
Ooooh...wait...SOFTWARE RECS!
Always forget about that
 
7:35 PM
@Andy Resolve
 
lol
 
wanders off
 
it's only the best site ever
 
@Andy what OS?
 
needs a bunch of RAM and a decent video card, but it does good stuff
 
J F
7:36 PM
I’m having some issues with file names being too long on my PC. Any suggestions for tools I could use to fix the issue?
 
yeah, hold on.
 
danke
 
@JF LONG PATH TOOL
 
LMAO
[flags as spam]
 
Zoe
lmfao
 
7:38 PM
@JF I actually ran into this a while back, in real life.
 
flags Art
 
J F
@ArtOfCode I’ve heard bad things about that though. Maybe I should ask a question about it on SE.
 
fp- by doppelgreener
 
prolly, yeah
 
@WELZ Mostly Windows - that machine is more powerful than my Ubuntu one
 
7:39 PM
@Andy Do you have any Adobe software?
 
[ charcoal-se.github.io ] continuous-integration/travis-ci/pr failure on 6b3ed85: The Travis CI build failed
 
@WELZ nods head None that I've bought
 
8
A: Windows Error : Source Path Too Long

Andre FigueiredoNewest versions of npm fix this issue flattening the path: https://github.com/npm/npm/issues/3697. Try D:\vms\fe> robocopy d:\path\to\temp\dir node_modules /purge to remove the nested dirs. Update As pointed by Rohit Jindal, another option is use rimraf util (which basically calls rm -rf): >...

or use robocopy to copy deep paths then purge them
 
...on second thought, probably shouldn't leave that email sitting there
 
was thinking same thing
 
7:41 PM
...internet, please stop being bad. please and thank you
nuuuuu
 
[ charcoal-se.github.io ] continuous-integration/travis-ci/push success on a5f3477: The Travis CI build passed
 
@ArtOfCode Can it do 1080p? It looks like only the paid version can do 4K, but I can't find a decent comparison between free/studio that mentions 1080
 
@Andy yup
 
:44533761 i removed this message because reasons
 
(psst. @zoe thanks for that help w/ setting up the email)
 
7:45 PM
Recovered from websocket._exceptions.WebSocketConnectionClosedException: Connection is already closed.
 
Zoe
@Mithrandir No problem. Just email me if there's a problem (see my profile for the email)
 
don't worry, i had it already ;)
 
Which site? Or chat profile?
 
fp- by doppelgreener
 
@SmokeDetector fp-
Someone who speaks Portuguese should take a look at these 3 MS reports of the same post and determine if these are really TP or FP. From Google translate, there doesn't appear to be anything actually TP in these, but Google may be softening something. I'm assuming the watched word is a group which is the target of hate-speech. This is the post for which the user in the last SD report was blacklisted.
 
7:55 PM
Recovered from websocket._exceptions.WebSocketConnectionClosedException: Connection is already closed.
 
@SmokeDetector ignore
 

« first day (1635 days earlier)      last day (2185 days later) »