« first day (1185 days earlier)      last day (588 days later) » 

5:02 PM
@Andy hands electromagnet
 
For your amusement:
0
Q: Unsolicited email from a startups user (my email is not public)

DavidPostillHas anyone else received unsolicited email (spam) from a https://startups.stackexchange.com/ user (I can provide the user name if necessary)? Notes: The From and Return path names exactly match a https://startups.stackexchange.com/ username (I realise that these can be forged but this seems un...

 
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, pattern-matching website in body, repeated URL at end of long post: Hydramedix Is powerful Skin care by sdvu john on apple.SE
tpu- by Undo
 
Oooh this sounds fun. Who, @DavidPostill?
lemme guess, this guy
shhhh bot, you're way too late
 
lol?
 
also @DavidPostill, permission to dox you to see if I can find out how they got your email?
(i.e. without cheating and looking it up with mod tools)
 
@SmokeDetector k
 
[ SmokeDetector | MS ] Bad keyword with email in answer, email in answer: How can I help reunite my mother and little brother? by CALISTA SMITH on parenting.SE
tpu- by tripleee
tpu- by Ferrybig
[ SmokeDetector | MS ] Link at end of answer: Selenium PhantomJS webdriver failing to grab ajax content by Rakesh Narwani on stackoverflow.com (@dorukayhan)
 
5:28 PM
@ArtOfCode Feel free :)
@ArtOfCode Correct, it is him. How did you know?
 
@DavidPostill Googled the company, looked up their team page, and compared names to usernames on Startups.
Only one of 'em had an account.
 
@ArtOfCode Yeah, I was wondering whether I should have censored the email ...
 
shrug it ain't exactly sensitive
Dude emailed you, he can't reasonably have any expectation of privacy
 
@ArtOfCode I was thinking about the meta effect ... which is why I didn't disclose his name.
 
Aye. I think you pitched it about right - enough information, but not so much that anyone can email him to shout at him.
though that might be quite effective actually...
 
5:39 PM
@DavidPostill I think I found you
Is your business title "Parts Administrator"?
@DavidPostill or Technical Support Engineer at <censored> supporting Videophones.
Because if so I found out where you work and your email adress and I haven't even strayed from the first google results page
 
I know where he works, haven't got an email yet
 
@Magisch No
@Magisch Yes
 
@DavidPostill Yeah. So I found your email and full workplace info on google
@DavidPostill want the link where I found it?
 
@Magisch I'm unemployed and haven't worked there for 4 years.
@Magisch Yes please
 
Ouch, so it's being stung from where you don't even work
Oh, I found out all I needed
Now I can lure David into my trap; USING PATTERNS
 
5:50 PM
:35433939 Nope. I don't even own that domain any more.
 
Not your email?
 
@Magisch Not the spammed one.
@ArtOfCode No you don't - I'm unemployed for 4 years :)
@M.A.R. Go on then. Spill the beans.
 
@DavidPostill Ah. I came up with the same company as Magisch did (the link)
 
> Your message wasn't delivered to david@postill.org.uk because the domain postill.org.uk couldn't be found.
 
[ SmokeDetector | MS ] Repeating characters in answer: What happens if you don't pray 5 times a day? by your not helping on islam.SE
 
5:53 PM
@DavidPostill were you ever involved in an organisation based in the Netherlands?
 
@DavidPostill Did you compete in a triathlon in 2012?
 
I should update that wiki page :)
@ArtOfCode I worked in NL for 5 years
@Magisch I wish :)
 
in that case I've found you
link?
 
naa- by MAR
 
@SmokeDetector rude
 
5:55 PM
@ArtOfCode Go for it
 
Wut
How rude?
 
I'm sure our spammer didn't go to this much trouble for a random person on SE
 
Nonsense is rude/abusive as per mother meta
 
Restart: API quota is 1723.
 
:35434158 Applause. Well done.
 
5:56 PM
@Mego But that's NAA ''this no anserz mah queschen''
 
@M.A.R. Eh, it can go either way IMO
 
:35434158 kudos
@DavidPostill You'd be surprised
Some people are scarily good at that
 
Shrug
 
Doubly so if its their job
 
@DavidPostill I'd bet you that our spammer ran a program of some sort - probably hit the SE API to find users with more than a threshold of rep on St.SE, then scraped the web for their names and/or websites and/or twitter usernames, and tried to find email addresses
@Magisch Triply so if it's a computer doing it for them and they just have to click Start.
 
5:58 PM
@ArtOfCode Lets see if anyone else says they have received the same email :)
 
@ArtOfCode There are entire companies that sell personal information bundling on "High value targets" (e.G. people companies think its worth targeting individually for spam) via scraping means
 
@ArtOfCode Is it worth mod flagging said spammer?
As it happens I don't really care if my email address is out there. But I do get irritated at such blatant spams.
 
@DavidPostill Nothing they can do, unfortunately. They won't suspend him, because he's done nothing wrong on the site. SE can't do anything either, because he didn't scrape the email off your profile so it's not against the TOS.
quite literally the only thing you can do is remove your email from the internet (ha!)
well, either that or respond - either send them a blistering complaint and hope you scare them off (likely), or spambait them: pretend you're interested, then start asking dumb questions until they tell you to leave them alone - see: James Veitch
 
@ArtOfCode If I can be bothered I will complain to his ISP :)
 
or that, yes
complain to literally anyone involved who you can find an address for - his company director, his ISP, the company's website host... anything
 
6:05 PM
@ArtOfCode Congressperson? :P
 
this is the UK
 
@ArtOfCode Exactly. If I complain he will get the works.
 
but you could CC your MP's PA and ask them to do something about the growing problem of spam... :)
 
Complain to the Queen
 
Dear Queen Elizabeth
I'd like to talk to the queen sometime
Must be cool being that old and having ruled a country for so long
 
6:08 PM
@Magisch Well if you start it like that, don't expect a reply
She's been dead for a few centuries
 
I'm pretty sure I'm about a billion people down on the list of people the queen would like to talk to
 
6
Q: What features did the Community Team discuss, have implemented, or have denied last month?

JNatOn a weekly basis, the Community Team vets a handful of feature requests to be sent over to the engineering team and estimated, scheduled into next month's plan, and built (or sometimes rejected for technical reasons). We try to select things that we think can make the most people happy, and ...

 
@Mego only the first one
 
@ArtOfCode Well yeah. If you want to address the current Queen, it's Queen Elizabeth II
 
@Mego Just QE would do. Her surname isn't II, funnily enough, and it's her name you're using not her title
 
6:14 PM
@ArtOfCode I know it's not her surname, but wouldn't Queen Elizabeth be ambiguous?
 
@Mego no, precisely because the other one is dead
I doubt you'd write an email to the 300-year-old one
 
Well one could try
 
one would have to be rather more respectful than "Ay whaddup queenie" if one was writing to that Elizabeth
 
@ArtOfCode Well you could be writing to a cruise ship :)
 
API quota rolled over with 1545 requests remaining. Current quota: 9999.
stackoverflow: 1421
askubuntu: 481
superuser: 405
unix: 287
drupal: 198
electronics: 174
math: 173
english: 159
salesforce: 156
gis: 145
security: 140
apple: 133
dba: 133
travel: 130
wordpress: 129
mathoverflow.net: 121
magento: 115
worldbuilding: 111
gaming: 105
sharepoint: 105
ell: 103
webapps: 96
stats: 94
mathematica: 92
scifi: 89
meta: 87
es.stackoverflow: 86
academia: 78
chemistry: 78
money: 77
ru.stackoverflow: 75
health: 75
tex: 74
physics: 73
movies: 62
diy: 60
codegolf: 58
gamedev: 56
 
6:19 PM
@DavidPostill that would also be a touch strange
 
@M.A.R. I'm disappointed in the readability of that screenshot
 
@ArtOfCode Ayy bby u wan sum buckingham palace?
 
@Magisch have this man beheaded at once
 
@ArtOfCode Your british is coming out :p
 
@ArtOfCode not hanged, drawn and quartered?
 
6:27 PM
@Magisch :P
@Loong nah, takes too long
 
6:40 PM
[ SmokeDetector | MS ] Bad keyword in answer, bad keyword in link text in answer, blacklisted website in answer: AVCaptureDevice flashMode not working by Harry M Allen on stackoverflow.com (@dorukayhan)
[ SmokeDetector | MS ] Bad keyword in answer, bad keyword in link text in answer, pattern-matching website in answer: AVCaptureDevice flashMode not working by Ross Alfred on stackoverflow.com (@dorukayhan)
tpu- by ArtOfCode
 
I looked at SE's job postings for giggles
I think I see why they have difficulty hiring new staff
 
[ SmokeDetector | MS ] Pattern-matching website in answer, blacklisted user: How can I read a paragraph in via the terminal as a single string in C? by Ross Alfred on stackoverflow.com (@dorukayhan)
tpu- by Magisch
 
@ArtOfCode Maybe the key is just to turn our inquiry into a Meta.SE feature request thread
At least they're keeping track of those :p
 
heh
 
6:57 PM
@ArtOfCode drop over in the tavern pls
@ArtOfCode Sorry if I'm making you look bad
I was quite a bit too salty
 
No worries, I can handle this :)
 
And I borderline verbally abused tim too with that comment
:/
 
[ SmokeDetector | MS ] Bad keyword with email in answer: How to add custom attributes in shopify customer list by Chelsea Mealey on stackoverflow.com (@dorukayhan)
 
@Magisch it's okay. I'm sure Tim's heard worse. You already know you're on a line, so go take 5 - away from the computer - and come back with a clear head.
 
right, back
I woke and cuddled my disapproving cat for 5 minutes
 
7:13 PM
meow
 
It's a very nice cat
soft with a lot of fur
 
Hey @ArtOfCode, @Magisch.
 
@Pops o/
 
hi
 
7:16 PM
As Sam Clemens once said, the report of my death was an exaggeration.
Hi Nobody.
 
you respawned? nice
 
What brings you to this dark corner of chat.SE
 
@Magisch now let me think... ;)
 
I've been planning to make the trip over here for about two weeks... was hoping I'd have some announcements but really it's just that (despite what it may have seemed like) your pings have not been getting routed to /dev/null.
 
@Pops long walk?
 
7:20 PM
My spam project hiatus took longer than I expected but I'm back on it now. For you, my next step is looking into your quota request. Haven't been able to get hold of anyone who knows about the API, which is my fault, not theirs, but I'll keep at it. Looks like next week at the earliest because people are traveling/on other projects.
 
Belated "hello"
 
@Pops I think the most important thing atm is multiple autoflags
 
@CaffeineAddiction More like I'm pilot testing a new thing where spam-related stuff is hopefully going to be chunked into Tuesdays.
 
And that just requires a yes or no answer from the community team
 
@Pops I know we said this in December's email, but: we don't necessarily need someone who knows the API to press buttons to turn up our quota. If you guys are happy with it, we can just register another key, and handle balancing between them ourselves - that gets us double the quota with no work and little discussion your end.
 
7:24 PM
We can deal with 10k quota per day, but the autoflagging project is on hold for expansion until we get word from a CM
In essence, whats in debate right now is your permission for some ... creative use of the flagging and api mechanics
Like, we already can do all of this stuff and more, we just dont know if we're allowed to do it
 
Executive summary of requests: 1.) Can we gets a higher API quota OR register a second key without consequences? 2.) We are currently issuing one automatic flag on many posts that SmokeDetector reports. Can we raise that limit to between 2 and 5 flags? If yes, what can we raise it to?
 
@ArtOfCode You probably understand this better than I do. Actually there's no "probably" because I don't know the first thing about the API. Which is why I want to ask someone who does; maybe there's some consequence that I don't have the knowledge and you don't have the access to foresee, or maybe there's some additional knob a dev can turn that is superior to you just getting a second key and load balancing.
 
So the API will take time to resolve
what about the autoflags?
 
Hang on a sec, I just spent THAT ENTIRE TIME writing that last message.
 
no pressure
:p
 
7:31 PM
@Pops okay, that makes sense. The quota thing isn't super-urgent, anyway, so we've got plenty of time for you to work on that and talk to whoever you need to
 
And if we didn't have enough requests, we could just turn up batching, right?
 
@NobodyNada yes
at the cost of slower reporting of posts
 
These autoflags, where are they going? Stack Overflow? All sites?
 
Across the network.
 
And right now, they look like they're coming from ArtOfCode?
 
7:33 PM
@Pops All sites.
@Pops 22 different people that entrusted their api write key to metasmoke
 
@Pops, 22 different users.
 
@Pops from anyone who signs up
which, as Andy said, is currently 22 people
 
From all of the Flaggers here
 
Today there are slightly less than 100 such flags issued (metasmoke.erwaysoftware.com/flagging)
 
7:34 PM
In essence, when a post triggers enough catch reasons that we can be statistically at least 99,5% sure it's spam, an autoflag is triggered
 
Now...click all those links and crash Undo's server.
:)
 
Everyone who signs up can modify how "secure" they want to be in getting their autoflags right
 
@Andy please, MS is a little more stable than it was last year :P
 
@Pops It's worth noting that we will never ever (not even if allowed) have the system issue enough autoflags to system-nuke a post
There's always human review involved.
 
What I'm thinking is, everything you've said sounds good, but let's be honest, ultimately, it's not my problem; flags get handled by mods, so if they're happy with the quality of the flags (which I assume they are) and wouldn't mind the extra volume (that one I'm not so sure about) then I would like it too.
 
7:37 PM
Then it doesn't matter if we cast 1 or 5 automatic spam flags.
 
@Pops The vast majority of these flags never go through a mod
If a post is nuked by 6 spam flags community deletes and locks it and no mod ever lays eyes on it unless by accident
 
@Pops this might actually result in lower volume. It means a post accumulates 3 or 4 spam flags quickly, so only the last 2 come in at human speed - the post gets nuked quicker, so mods are less likely to see the flags.
 
Oh, right. Spam flags.
 
We're essentially issuing autoflags to make it take less human flags on posts where we're statistically super-sure they're spam
 
More flags would change nothing from the mods' point of view; it just means its easier for us to nuke the posts
 
7:38 PM
@Pops and I know in the end it's not your problem, but we're asking a CM so that if the mods complain to you guys then we can say "we discussed it with them and it was okay", rather than saying "oh yeah didn't we tell you we've been almost nuking all your posts?"
 
You know, the 40k+ true positives to 40 false positives kind of sure
 
@ArtOfCode Sorry, what I meant by that was "I'm not the one doing the work of physically handling the flag, I have to put myself in that person's shoes" but I somehow momentarily forgot that we were talking about spam flags.
 
@Pops aye. I mean, we're confident in this thing, but not so confident that we want to risk CM rage :)
 
We're confident but we're still technically doing something that CMs could construe as sockpuppeting
So ask first is imperative
You know, to avoid getting all of us suspended for a year
 
7:42 PM
@Andy we might not survive an onslaught like that :)
@Magisch this, especially if we use Smokey's account to cast some flags (though given how many people have signed up, we probably don't need to)
 
@Andy ...are those fish?
 
@NobodyNada Yes
 
ARE YOU CALLING MY USERS FISH, SIR?
 
Yup.
 
@Andy Why does that just make me think of
 
7:44 PM
hahaha
 
well then
In essence, we'd still be taking personal responsibility for the flags
We just want to get on the same page that we're not doing something illegal by SE rules
 
aye, it's just the whole casting multiple flags from multiple accounts based on one source of data
 
We have some ridiculous data sets and quality control in play, ridiculous enough that the predicted accuracy of autoflagging is far above that what we have as humans on the site - and we're all very high accuracy flaggers
 
Okay. So a post comes in. You are as close to certain as you can be that it is spam without actually being truly sure. Currently, your system is already casting one spam flag on that post automatically. You are asking permission to cast an additional n spam flags in that situation (but not enough to get it nuked without a human looking at it). Is that about the shape of it?
 
We're talking about accuracy ratings like 99,74% where the .25% are posts that are borderline or NAA anyways
Yes
Essentially
We all trust the algo enough to take personal responsibility for flags it casts in our name
And we're all super picky with flags, most of us have 99%+ personal accuracy
 
7:48 PM
You've probably considered this already, but what about waiting until a k before casting more autoflags? Just to minimize the damage from the .1%
 
We even have automatic self-correction in place where if a reason becomes unreliable due to recent events the system will self-adjust what it considers worth flagging
 
Could Metasmoke also retract an autoflag if it later receives an F feedback on that post?
 
nope
no api for that
 
IIRC there's no way to retract flags via the API
 
@NobodyNada because at 99.98% accuracy do we really need to?
 
7:51 PM
F's are sufficiently rare that this won't be a problem
 
How about trying it out at three autoflags instead of the current one, seeing what happens?
 
Also, the .02% we're talking about aren't clear Fs
 
it's balancing quick nuking time against likelihood of false positives, the latter of which is stupidly low
@Pops we can do that, yeah - just want to make sure you guys are okay with it before we do any more than 1
 
@Pops If it interests you, here is a simulation: metasmoke.erwaysoftware.com/flagging/conditions/sandbox
180 weight is about the lowest people will go atm
 
(That just now was me saying your request is approved. At least, until someone else on the team comes by and says "WAIT, POPS DID WHAT?!?!?!")
3
 
7:53 PM
It yields 99,74% accuracy
 
@Pops heh, okay. So we'll start it off at 3 flags per post. If that goes okay, are you happy for us to increase it (with discussion among ourselves) to 4 - or potentially later on 5 - or would you like us to have another discussion with you guys before we do that?
 
Might not be a bad idea to ask the people who have volunteered their accounts to re-sign-up, just to be extra-clear about the "responsible for the flags cast in your name" thing, now that the autoflagging is going up. Really just on principle, not because I'm worried or anything.
 
> Warning: for all the joking about, this is serious. By ticking this box, you are explicitly consenting to your Stack Exchange account being used to cast flags automatically based on data collected by SmokeDetector.
> While we believe this has a high rate of accuracy and have designed the system to reflect this, flags cast using your account are your responsibility, and many invalid flags can in some cases result in your SE account being suspended. Only tick the box if you are happy with this responsibility.
 
I don't imagine it'll be a problem, but check in with us first.
 
Thats the text above the big red "turn on" button
 
7:55 PM
@Pops cool, no worries
 
Thank you, @Pops!
 
Sure, sorry it took so long.
 
Long but thought-out is better than quick but hastily done.
 
I'm curious if you have examples of the false positives. I thought about asking before but I didn't think it was too relevant because I think 0.02% is better than my own spam detection rate.
 
I haven't seen an actual FP yet, tbh
Only borderline cases
But art probably has some
 
7:58 PM
@Pops we've had a whole one FP so far, in 3020 flags cast. Lemme see if I can find it.
 
heh
genuine FP
 

« first day (1185 days earlier)      last day (588 days later) »