Tavern on the Meta

Anna Lear

1:00 PM

@GnomeSlice Thanks!

tchrist

I think I must misunderstand how the ML spam-detector works, or rather, how the ML portion of it works.

When a posting is flagged as spam and deleted, it gets fed into the training engine.

This is to improve that engine.

It is therefore counterproductive to spam-flag things like wrong answers. Doing that is wrong for other reasons, too, of course.

However, when it comes to gibberish, something is rotten in the state of Denmark.

One reason given for not using the spam flag on gibberish like “dfghioik';ojj[phkugiu werkfwkurgiureigorgiuuuuuuuuuuuuuuoiueghoreg” (that’s an actual real-life example) is that because spam-deleted postings are fed to the ML spam auto-detector for training, this would degrade the quality of its decisions. I don’t understand that; gibberish should be easy to detect, easy to train on, and desirable for auto-blocking.

Why would you not want to feed the ML Monster gibberish so that it can block gibberish?

3

If a developer or community manager could please provide some insight here, I would appreciate it.

Vogel612

I think I know the answer..

spam flags are only for promotional content

Unihedron

@Vogel612 Enlighten us!

Vogel612

or rather, blatant "buy this, it helpsz0r" sh*t.

everything else is NAA.

tchrist

That is not an appealing answer; it is rather circular in fact.

Vogel612

1:11 PM

how so?

Unihedron

No, It's VLQ.

Doorknob

No, it's offensive / abusive

Unihedron

NAAs are for links that attempts to answer the question, but content outsourced.

Doorknob

6 abusive flags autonuke a post

Vogel612

@Unihedron that's VLQ...

tchrist

1:12 PM

This does not address my question.

Unihedron

VLQ is for content with severe content problems that can't be salvaged by editing. Like gibberish.

tchrist

I am aware of the current and rather strict definitions of these things.

Anna Lear

I, uh... don't think we have an ML engine going through these at all.

Unihedron

@AnnaLear WHAT?!

Vogel612

the ML monster is supposed to block stuff that actually makes sense when read

tchrist

1:13 PM

@AnnaLear Oh dear!

Unihedron

you kidding

?!??!!???!!!!????????????????

tchrist

I thought spam-flagged stuff (and only spam-flagged stuff) that gets deleted gets fed to an ML engine.

Unihedron

I... I don't know anymore... I'll quietly walk away...

tchrist

Tim posted something about this rather recently.

Nice nom-de-plume he’s got there, too. :)

Let me find it.

Anna Lear

I could be wrong, but as far as I know, spamram is mostly based on activity. There is another system that can block spam based on content, but even that's not automatically trained. I believe ML is the next step if we need it, but we're not google so we aren't writing a complex NLP type thing here. yet.

tchrist

1:16 PM

8

A: Anti-spam: Blocking questions whose bodies end with a raw hyperlink

We've experimented with this sort of thing in the past, and it hasn't been very successful for any appreciable amount of time. This is for two reasons: We have to tell users why something can't be submitted, although we can be kind of vague as to what's wrong, to avoid users 'fixing' only the s...

Is the blue line in Tim’s graph posts that were auto-blocked?

Anna Lear

yes

Braiam

in Unix and Linux on The Stack Exchange Network Chat, 1 hour ago, by Braiam

Each time someone mistype kernel as kernal, I kill a kitten.

now, tell me how bad I am

Unihedron

What's a kernal?

Anna Lear

@Braiam Each time someone misses an "s" at the end of a verb, a kitten cries.

tchrist

@AnnaLear So you are saying that that particular form of auto-blocking is not the result of an ML-trained program?

Anna Lear

1:18 PM

@tchrist Yeah

tchrist

thinks

Unihedron

(shooks head in disbelief)

tchrist

ponders suggesting @Uni should be holpen his head

Vogel612

(considers linking Dr. Edward Richthofen's Grammar N-word)

Braiam

sees heads popping everywhere... scarry

tchrist

1:21 PM

@Vogel612 That wouldn’t help. I happen to have a gold tag-badge for single-word-requests, you know. :)

Unihedron

Thanks for the information @AnnaLear. Sorry for the scream.

Vogel612

@tchrist I happens[sic.] to believe that grammar consists of more than single words.

Unihedron

Someone starred it though.

Anna Lear

np np

tchrist

Anna, could you safely reveal anything about the manual process by which spammage gets autoblocked without risking compromising that mechanism by divulging its behavior?

Because it seems to me that gibberish is something that could be autoblocked without too much trouble.

Anna Lear

1:25 PM

I honestly don't know that much about it. It's fed by spam/offensive flags, user deletions, etc.

Bart

Fed to hamsters?

tchrist

So something is fed to something. This makes sense.

Andy

@Bart No. A monkey

tchrist

But I am surprised that this should include user deletions.

Bart

@Andy I should have known

Anna Lear

1:26 PM

@tchrist You have to keep creating profiles to post, so most spammers cycle through unregistered profiles that can be destroyed and used as signal.

Gibberish is both easy and tricky at the same time - the actual number of gibberish posts compared to the overall amount of crap we block is relatively small (I think). At the same time, we have localization to worry about, sites where the majority of posts aren't in English, etc.

tchrist

@AnnaLear Yes, that would be a very nice bit on the feature vector.

Unihedron

@tchrist A process I've seen happening is that when you post short answers or with many incorrectly backtick'd words, the system asks me to fill in a captcha.

But captchas aren't really that useful.

tchrist

@Unihedron Oh, that’s me. :)

And “incorrectly backtick’d words” verges on the tautological. :)

Vogel612

well it's not a backtick....

because they look ``` ← like that.

aaand yet another markup "bug".. possibly.

tchrist

@AnnaLear I was thinking about both those things. However, gibberish on an English-language site should be pretty easy if you base it on cluster-frequencies and word-length.

@Vogel612 I don’t comprendos you.

Doorknob

1:30 PM

@Vogel612 `; source `\``

tchrist

You know how machine-generated passwords are poorly remembered if they do not look like they could be plausible English-language words.

Vogel612

` ``

(killing baby unicorns)

tchrist

What do think is a problem with ˋbackticksˋ not formatting correctly?

Vogel612

@tchrist my own stupidity (proceeds to hit his head on the table)

tchrist

So the better random password generators use EN CLP data when creating them.

@Vogel612 I fear I was too subtle. I stealthed you.

What do think is a problem with ˋbackticksˋ not formatting correctly?

is

What do think is a problem with \x{2CB}backticks\x{2CB} not formatting correctly?

is

What do think is a problem with \N{MODIFIER LETTER GRAVE ACCENT}backticks\N{MODIFIER LETTER GRAVE ACCENT} not formatting correctly?

Doorknob

1:34 PM

Or just `escape` them

tchrist

2CB or not 2CB, that is the question.

@AnnaLear I think probably the most important thing I’m taking away from all this is that because the proportion of deleted posts that are just gibberish is so low compared with those that are not gibberish, that it is unlikely to be a worthwhile investment in developer time, and possibly even in CPU time.

@Vogel612 Only if your state requires that you annually declare and pay up any syntax due.

Vogel612

@tchrist Thread main threw OutOfMemoryException, see cause:

tchrist

How dear the CPU-time would be I’m less clear on, but it’s still not much ! for the /dev/buck.

skullpatrol

Aha! So this is where all the cool people hangout 😎

tchrist

@Vogel612 Were I seeking awwwwws, I would post pickies of kitties.

Vogel612

1:41 PM

lol @skullpatrol I saw you in about every chat-room I frequent relatively consistently..

skullpatrol

lol

Vogel612

@tchrist Queue does not accept any further input. or should it be a PipeClosedException??

tchrist

Did anybody else just hear a meow?

Vogel612

woof

Anna Lear

@tchrist At least not at this stage, yeah. Maybe later.

tchrist

1:45 PM

Ok.

Andy

@skullpatrol Welcome to the Tavern

skullpatrol

Thank you :-)

tchrist

The Tavern is exempt from stalkers looking at your other-room lists and following you thither — well, or hither in this case.

It’s an alternate universe.

Unihedron

yesterday, by SmokeDetector

[ Wisdom ] Shog is always watching you

skullpatrol

On a higher dimension I hope :)

Vogel612

1:50 PM

@tchrist even across subdomains??

tchrist

      From here, he can see at least three shoggoths. The intersections
      of their watered-milk bodies reflect the light in rainbow bands.
      The colorful fruiting stalks nod some fifteen feet in the air, swaying
      in a freshening breeze.

      From the greatest distance possible, Harding reaches out and prods
      the largest shoggoth with the flat top of his hammer. It does nothing
      in response. Not even a quiver.

      He calls out to the fisherman. “Do they ever do anything when they’re

Excerpted from “Shoggoths in Bloom”, by Elizabeth Bear

I hate that it doesn’t do the whole thing.

Vogel612

Do I want to know what's a shoggoth?

Sam

@Unihedron Finally submitted the tag edit btw.

Unihedron

@Sam I saw, as you posted on the tavern :)

tchrist

@Vogel612 “What kind of a fool would come poke one to find out?”

Sam

1:52 PM

Yeah, I didn't know if you were still active there.

Vogel612

@Sam Ding! ftw

Unihedron

Haha, I take dual monitors to my advantage. dw. :)

tchrist

@Unihedron I’m not even sure if a duel is lawful without a monitor.

Unihedron

Screw dueling, let's go parse xhtml with regex.

tchrist

Be careful: remember whom you’re addressing.

Unihedron

1:54 PM

I know, you're the person who posted about parsing HTML with regex, now try xHTML. :D

Vogel612

and he calleth upon the unspeakable horror from outside any dimension

tchrist

@Unihedron All things are lawful for me, but all things are not expedient: all things are lawful for me, but all things edify not.

Sam

@Unihedron D̘̊͒͊oͥͩ̿͊̋̃nͭ'̌̂t̓̄ ̻̇́̎͐͌ͮ̍e͛̏͆̌ͩͮv̖̫̏̆e͌͐̅̏n ͗t̥͔̹r̂y

10

tchrist

@Sam I can do whatsoever I should put my mind to, reges gratia.

Braiam

the start wall has a different font?

tchrist

2:00 PM

@Sam There is no try { }; there is only do { }.

@Braiam The starboard has the same face, but the points differ.

Undo

@try
{
    [self thingThatDoesntExist];
}
@catch
{
    NSLog("crap");
}

@tchrist ^ valid Obj-C :P

Sam

@tchrist There is no Exception. There is only EffingExcuse.

Unihedron

do or do not, there is no try.

tchrist

@Undo I lament to inform you that like St Dennis, I do not personally find C so holy objectionable as all that would entail.

Sam

@Unihedron Do not try to not try to do, only do to try to not try to do.

2

Unihedron

2:06 PM

@Sam (only )?do (?:not|to) (try to (?:not|do) ?)+

Sam

@Unihedron Naughty regex ninja, thy knows I not speaketh thy's secret art of Regex.

(removed)

Anna Lear

@Sam I see what you did there.

tchrist

@Sam flags for unantiagrammaticalisticality

skullpatrol

🙈🙉🙊

3

tchrist

@Sam Quod licet Iovi, non licet bovi.

Vogel612

2:17 PM

@tchrist um. you definitely are a German. only we can combine words into arbitrarily long constructs.

Sam

@tchrist Googles what on earth that is...

tchrist

in English Language & Usage on The Stack Exchange Network Chat, Jul 25 at 13:30, by RegDwigнt

I reject your premise that "English grammar doesn't allow for concatenating". The English mathematics professor is exactly the same as the Swedish matematikprofessor, and control flow graph visualization software is exactly the same as the German Kontrollflußgraphvisualisierungssoftware. Some languages opt for writing compounds as single words, others use whitespaces. But they are still compounds. Spoken language is primary, and orthography is a random approximation. We could start putting whitespaces before all suffixes, but that would not affect their function in the least. — RegDwigнt ♦ Nov 8 '12 at 10:25

Vogel612

@skullpatrol flag for offensive, as it's needlessly obfuscating against users of google chrome. (can't read it. Font missing) </joke>

@tchrist "unantiagrammaticalisticality".matches("*\w*") -> false

tchrist

@Vogel612 Non sequitur.

Vogel612

not that this would be valid anywhere..

tchrist

2:22 PM

@Vogel612 Illegal nonregex passed to regex compiler, core dumped.

Vogel612

@tchrist don't be such a killjoy ;)

tchrist

macbook# perl -le 'print "Match unantiagrammaticalisticality =~ $_\t=> ", "unantiagrammaticalisticality" =~ $_ ? "Yippee" : "Nope" for qr/ \w /x, qr/^ \w $/x, qr/^ \w+ $/x'
Match unantiagrammaticalisticality =~ (?^x: \w )	=> Yippee
Match unantiagrammaticalisticality =~ (?^x:^ \w $)	=> Nope
Match unantiagrammaticalisticality =~ (?^x:^ \w+ $)	=> Yippee

YANNETUT

looks for the Julie Andrews version