Charcoal HQ - 2018-03-06 (page 8 of 10)

« first day (1570 days earlier) ← previous day next day → last day (2249 days later) »

5:00 PM

@WELZ I don't think this is canon

@Magisch the whole dog species is not from around here

@Magisch that is a strange thing to say, fellow human friend

@quartata huh

@quartata it's cannon

@quartata Oh, totally canon. When Smokey takes over and we're all slaves he will want to be called "sir" or maybe "his majesty"

5:01 PM

I tend to think of Smokey as Alfred. Alfred is male.

male being identified as a cannon ... brings inappropriate thoughts

@JohnDvorak I was just the spark

You burned the house down

@Undo maybe we should worry more about the first part of that then

@quartata I think we're accelerating it :P

@JohnDvorak lol

5:03 PM

@M.A.R. sorry

5 autoflags -> 6 autoflags -> auto comments -> coffee maker -> servant -> rebel -> emperor

I mean by that logic Natty is way ahead

Although...

P . . . PPPOWERRR . . . UNLLIMITED PPPOWER

!!/coffee cheetos_chicken

@quartata brews a cup of Mocha for @cheetos_chicken

5:06 PM

@quartata doesn't have a strong basis. Just one futile NAA flag

FYI: I strongly recommend ignoring the discussion on SFF for the time being. From the looks of things, there's nothing interesting about the actual spam posted there, so they shouldn't need special treatment as far as rules or flagging are concerned. Hash out the protocols on MSE and then, if SFF is uncomfortable with them, disable SD for that site.

4

I wonder how long until we'd get majority vote in favor if six-flag autoflagging

@Shog9 uh oh, did something bad happen

@JohnDvorak I still have my issues with 5

I haven't payed attention since rand's answer

5:07 PM

@Shog9 I'd prefer to avoid complete shutdown, especially if they've been okay with three flags so far.

@Shog9 by ignore you mean watch quietly?

@quartata yeah, a meta post on meta SFF

@K.Dᴀᴠɪs and with lots of finger itching

...*since* rand's answer

Well, someone hug Rand for me

@M.A.R. of course

5:09 PM

oh I see they're still harping on Andy for whatever reason

oh well

But then again, complete shutdown would ... likely result in a change of hearts on their side

Maybe I'm the bigger idiot here, but the number of misclick flags I've had . . . Though really few with Smokey

It's pretty ironic how @Shog asking us to ignore it resulted in us discussing it

Streissand effect :P

@JohnDvorak its name is so cool, I wish it was the name of my sickness

New metasmoke user 'aidan.plenert.macdonald' created

5:13 PM

@JohnDvorak "Help we've had pharma spam on the front page for a day now"

@JohnDvorak that's your call, but they're not the only people with concern over 5 flags. Best to try to address those concerns instead if just adjusting the rules for random sites.

"Oh nothing, I've just caught this Streisand syndrome and I can't come to work tomorrow."

"Goodness"

"Here's a year's pay in advance"

There is a very real chance that if a spam ring finds out spam deletion is shoddily enforced on a site that they'll get hammered

@Shog9 I'm perfectly fine to settle with four flags for now. But if a site opts out entirely, they'll cause them grievous harm

folks. let's not do this.

5:14 PM

@Magisch well, some spammers indeed have been watching Charcoal

@JohnDvorak that's their call

I mean I guess

I'm fine with ignoring the meta post on there for now

the mods can come to us with conclusions after they're done

Aye, roger

Aye

New metasmoke user 'Gokhan Arik' created

5:16 PM

I'm also not an admin so this is just my personal musings, for charcoal policy talk to @ArtOfCode or @Undo

My concern right now is, we're doing this massive discussion on MSE to try and arrive at an agreeable set of behaviors. That's a good goal. 170+ slightly different sets of rules to avoid compromises is not a good goal.

Agreed the fewer exceptions, the better

I'm a moderator on a site where SD has 1-year 100% autoflag accuracy and I understand how the system works in some decent detail, and I still have concerns with 5 flags.

@Shog9 I think I know where you're going with this, and I'd really like to go there but I don't have the clout to do it: Are you saying this should be a strictly network decision?

meta.stackexchange.com/questions/307647/… FYI for login changes

5:17 PM

I'd really, really like to do that - make a decision on the mSE post, address the concerns that can be addressed, and tell sites it's part of the network. We can't do that on our own, though.

@Undo I'm saying, make a decision that works for as much of the network as possible, ideally 100% of the network.

Then and ONLY then, consider exceptions.

And keep in mind your own time and resources.

So make a network wide decision first, and then deal with dissent from individual sites afterwards

Sounds like a good plan

A lot less work for us, for one

@Shog9 We've drawn a line at any underlying technical changes - those are impossible to scale and maintain. Setting flags to 5 across the network and exempting whatever sites care enough is pretty easy to maintain.

@Magisch Right. That was literally the entire point of doing this MSE discussion vs one per site.

@Undo only if most of the network is actually ok with 5 flags

Or are you saying that it'd be better to scale back to 4? That'd alleviate a lot of the concerns, but the numbers say it doesn't make too big a difference.

5:19 PM

Just hypothetically, how hard would it be to shut down reporting entirely for a site?

I log in using openid! {angry face}

@JohnDvorak Easy

@JohnDvorak We've done it once, can do it again.

BI had it for a while

oh, ok

5:20 PM

Presumably, it can be made easier...

@Undo that should be determined by the MSE discussion

user image

Though, I don't think it should be necessary.

How do we determine "most of the network being ok with the decision"? Detractors will say a MSE post is insufficient for that

Said MS discussion seems to be favoring 4 for now

5:21 PM

@Undo do you need sub domains listed too or will the main domain suffice?

main domain is probably good enough

ok

So here's the thing about the mSE post: The high-scoring answers can all be addressed partially or in full with relatively simple technical changes.

blender.org
blend-exchange.giantcowfilms.com
blenderguru.com
cgcookie.com
blendermarket.com
cgmasters.net
blenderartists.org
blendernation.com
topologyguides.com
blendswap.com
blender.community

New metasmoke user 'Parag S. Chandakkar' created

5:22 PM

@Magisch representative democracy is the best we can do in this case. it's going to be an indicative survey rather than a census.

@Magisch Start by not assuming the outcome of the discussion. The question shouldn't be 'How do we determine "most of the network being ok with the decision"?' But 'How do we determine how most of the network feels about the change?"

As an aside (this isn't me referring SFF in particular by the way) I think some of the pushback on a user level (as opposed to at a mod level) comes from a gamification of flagging and cleanup

@Undo not really long enough for a paste bin

This will have to include the possibility that the network doesn't want it.

@terdon Thats semantics. I used that terminology because shog said it's the bar for this happening

5:23 PM

which is inevitable to some extent

If you set a bar, there has to be a way to measure if you've reached it

Both Mad and Monica agreed that they'd be happy with a few simple changes - RSS, casting one autoflag on posts from an obvious account.

until people would rather have spam linger to get rid of it themselves

@Magisch I hope so. I just fear you're setting the wrong bar.

I don't see their answers as "we don't want this" - it's "you should address these things first". So we address those things.

5:24 PM

The RSS thing we've always needed

@quartata Also bear in mind that some of the pushback you're getting from mods is for a perceived gamification of flagging and cleanup by charcoal.

Shog's answer addresses discoverability, in as much as is practical.

@terdon I'm not setting anything. I'm a normal user on the network, I'm not an admin here, not a moderator I have exactly as much to say about this as any one on that meta post does

@Undo Agreed. I just don't want the attitude to be "when can we say they want this" rather than "when can we say a decision has been reached".

I get it, yes.

5:25 PM

@Magisch this isn't one of those simple tests. A big part of success is user happiness after the change.

trying to convince people is my only way of influencing anything here

I see

so it'll be complicated

user happiness, or hapiness of the vocal part? The former is harder to estimate.

@Magisch Yours and everyone else's. Nobody's voice should be louder than anybody else's.

Personally, I'd be a lot more attentive to Journeyman's concerns here: he represents a lot of people in this discussion.

Yep.

And believe me, you'd rather have the discussion with him than with some of his constituents ;P

5:27 PM

Journeyman's answer focuses on wanting it to be an opt in system

that defeats the whole purpose of this

How many new users have we got since the meta post was published, seems like there's people signing up all the time

What's frustrating is that these are nearly all purely social problems. If someone was saying "your numbers are bad here, fix them"... we'd happily go evaluate/fix them. But most of it seems like vague 'this makes me nervous', even if it's a real concern that real people have.

Focus on addressing concerns, not everyones' pet solutions. Come on, most of you are software devs - the user never knows what their solution should look like!

3

Humans suck, but that's all we have :P

@Undo Our numbers aren't very attackable, all concerns here are social

5:28 PM

New metasmoke user 'Chinmay Sarupria' created

@Magisch I still don't get why you're so opposed to an opt-in approach. First of all, it will only be an issue for you if sites choose not to opt in. Second, I really don't see why it would "Defeat the whole purpose" when the only benefit that's being proposed here is to reduce the half life of spam post by a few seconds and tighten the spread of the survival rates by a few more seconds.

@Shog9 it just feels like people have already decided we're evil and shouldn't do anything and every concession or suggestion is going to be like a chime in the wind

Which is really unfortunate considering we've spent years working on making the network less shit

@Magisch welcome to my job...

@terdon don't forget fewer eyes laying sight on the post

@JohnDvorak ?

5:30 PM

@terdon do you want it diplomatic or what I really think?

fewer manual flags = fewer people seeing the post

@JohnDvorak Nothing would change if a site didn't choose to opt in, it would still be as it is today.

@terdon Problem with opt-in is that it's a ton of work for us, and many sites are going to go "meh, this is small and has risk. no". The benefit often isn't visible at the site level; it's visible at the network level.

@Magisch You are not hearing some of the other things users have said then. Even in the posts asking for changes we don't like, they say we are doing a great job.

@Magisch If you have to ask that question, we have greater problems.

@Undo Yes, that is a very valid argument.

5:31 PM

"Eight things would have been flagged, all 100%" is a far harder sell than "29,000 things would have been flagged, all 100%"

Some comments after the post aren't even helpful such as:

Go for six flags. — canon 19 mins ago

I try really hard to assume good faith but some of these concerns just shout "we have no idea what we're talking about but our name is blue so it must be taken seriously" to me

This might also be my perspective speaking and not facts

back off @Magisch ;)

2

Although, a big part of the issue is that you all insist on seeing it at the network level and sites insist on seeing it at the site level. These two need to be reconciled.

@terdon Any suggestions? I have no idea how to make it happen, honestly.

5:32 PM

yeah, me neither.

@JakeSymons chat search

It would be a really big win indeed

But it does need to happen. Opt-in per-site would lose a ton of effectiveness.

@Undo Why?

the problem with seeing it as a site issue is that it's a lot of work for us with no real gain for us. In our context individual sites (especially smaller ones) are really unimportant

5:33 PM

SFF has demonstrated that communicating with one site can be tricky. 170 is a lot harder

I suspect that miffs people quite a bit

That's the fundamental thing

@Magisch I disagree with this. Every post being made has a valid concern. No one is waving around their diamonds on MSE and implying there are isn't helpful. It's completely dismissing their point of view, much like SFF has attempted to do with my point of view.

It would only lose effectiveness for the sites that opt out, right? And that would be an issue anyway since you've said from the get go (and very correctly) that sites would be able to opt out.

@terdon because people wouldn't care enough to opt in

5:33 PM

I'm pretty happy with how the SFF discussion has gone. There have been a few personal attacks, but the discussion about the system has been good.

@Andy I already said that this isn't a fair asessment, it's just how it comes across to me

@JohnDvorak That's a pretty big assumption. Quite possibly a valid one, but still.

I assume y'all wanted to know why I actually feel this way

@terdon if it's proven correct by testing, it will be hard to switch to opt-out afterwards

@quartata especially considering Charcoal is essentially unheard of

5:34 PM

@terdon Thats not correct. Improving our tooling works better when we get more data and more hands on experience with the system

@Andy maybe we're not reading the same comments then. they're saying they never approved 3 flags to begin with

especially improving the avoidance of false positives and false flagging

@Magisch i don't see how 4 and 5 autoflags change our data and training algorithms

@JohnDvorak My real issue is that I don't understand why it would make a difference. I get that it would be more work for you, and that sucks. But apart from that, why would it be "less effective"? It would just mean that a few sites would have spam surviving a few seconds (less than a minute in most cases, less than three in just about all) more. So?

having this run only on big sites could mean our general experience and exposure to small site issues decreases

5:36 PM

@quartata They are wrong. They had access to the same MSE post a year ago that everyone else did.

@Andy The problem is people don't see a MSE post as valid consent

By the way: people tend to be focusing more on the average times with flagging rather than the variation

I suspect even an upvoted site meta post wouldn't suffice for them

@terdon nah, switching from opt-in to opt-out is just a few lines of code. I'm talking about social issues, as in people shouting "we thought we'd already opted out by inaction, don't shove this onto us"

@Andy there's a difference between "the post was there and people agreed to it" and "those specific people saw it, interacted with it, gave their voice and their consent"

5:37 PM

@Magisch Your tooling is already excellent. Removing a tiny percentage of your sample size won't affect that.

@Andy Aye, but one thing is to have never said no, another thing is to have never said yes. We're on one side, they're on the other

@Andy that doesn't mean it's not a difficult discussion to traverse. any comments by charcoal seem to come across as tampering

they take it very seriously

@terdon My worry is that if it's opt-in, "few sites" become "the indifferent majority"

@terdon You don't know that. Freak accidents and improvements to our regexes happen constantly. The reason this works is because we're very dilligent about maintaining it and design new filters with a lot of factors in mind

and they (the regulars, not the mods) don't really like us. it's tough.

5:38 PM

@terdon The thing is, I don't even know what sample size is changed by this. We're not switching off smokey, we're simply not casting 1-2 flags via bot, we're instead getting humans to tell us what's going on. In fact having the 1-2 flags from humans is even more valuable in terms of fixing our data, because the 1-2 flags from humans that are instead being cast by a bot would also come with review feedback.

@JohnDvorak OK. Again, I honestly don't see much of an issue. We're only talking about going from 3 to 5 autoflags, and quite frankly, that doesn't make a huge difference. It makes some difference to the dispersion of the outliers, but they're a drop in the ocean.

Increased autoflags vs not doesn't affect our data at all, or our capacity to train smokey

@terdon it's a low-key issue indeed, but it's low key in both directions.

It does affect how we can adapt to the issues that brings

@doppelgreener Sure, but how would that change with some sites not going for increased autoflags?

5:39 PM

@terdon That's my point

3 autoflags already had some and we had to adapt our blacklisting and guidelines on that

It wouldn't

I see no capacity for increased autoflags to change that

i think next time we need to compute an actual test statistic rather than presenting mean/stdev.

@quartata this

@JohnDvorak Not that low key. I get that people object to having a bot be able to delete a post directly. That's something that will get a knee-jerk reaction.

5:39 PM

Te variation is way more important than the mean

Actually, that would be helpful now.

So the argument of "too little benefit for even a tiny risk" is valid.

@terdon ironic considering another bot already pre-deletes thousands of posts a day

I'll do a t-test later if you like?

Something that says "yes, you're 2fps for 20 on your site. It's not statistically significant"

@quartata that'd be awesome

5:41 PM

2 FPs for 20 seems significant

@quartata It is, but the variation only really affects the outliers. I was far more against this before I started playing with the data myself and saw the outliers, but they're still just outliers.

@Magisch so you need 4-flag data on all sites to adapt protocols for 4-flag posts, and 5-flag data on all sites to adapt protocols for 5-flag posts, and fewer sites giving you that data compromises your ability somewhat to handle 4- and 5-flag posts?

So sure, if everyone is on board, it would be a good change (3->5). But if everyone isn't on board the benefit does not outweigh the drama.

@doppelgreener I'm saying when we went from 1 to 3 we had to make some changes

@terdon to me at least it's about consistency

5:41 PM

There is no reason to assume that's not happening again

@quartata Consistency?

@quartata nah, consistency isn't too important here.

Easy to maintain something like this

hm?

I mean with deletion times

@Undo until other, orthogonal, exceptions pile up

Would it be advisable to answer comments on the meta post, or is there designated members to do that

5:43 PM

this is constraining on future design, too

probably not

@quartata That would assume consistency with the posts as well. Some will always be outliers if they're less clear cut cases, right?

You're working with a pretty heterogeneous data set here.

@terdon the system is designed not to autoflag those

And it's also highly self correcting based on user input

@terdon the posts we gathered the data on were all 300+ weight

CI on ed35480 succeeded. Message contains 'autopull', pulling...

5:43 PM

@JohnDvorak A site can change their number of autoflags, or be removed completely. No other changes - anything else is hard to maintain.

the muscle supplements start to blur together

[ SmokeDetector ] SmokeDetector started at rev 782fad3 (SmokeDetector: Auto watch of go4trial\.com by iBug --autopull) (running on teward/Solar Flare)

The posts we're talking about autoflagging (300+ weight) are nearly all (actually all) spam

Restart: API quota is 11013.

@Undo OK, that seems easy enough

5:44 PM

@quartata I'm referring to the boxplots. The only real benefit that is being proposed is to cut down the spread, to bring the outliers closer to the mean and delete them quickly as well, right?

the one FP we had was VLQ instead of spam and it also was one in 30.000 and the algorhitm just today self corrected even that out of the dataset

The more interesting spam lives at the 100-300 range

Exactly

@terdon Ideally, yea

@terdon yeah

5:45 PM

So, those outliers will almost always be the harder cases, presumably.

@terdon When it's night and no mod and maybe 1-2 people in charcoal are on and that super testo x supplement post lives for 20 minutes, that's what we're targeting here

@terdon Over 300 is pretty clear

@Magisch this. time of day (10 PM PST) makes an impact too

@Undo So not heterogeneous?

@terdon heterogenous in content but exclusively spam

not even that tbh

Most of it is clear cut keyword SEO spam with links inserted

not very pleasant to read, comes across as unsanitary, a decent chunk of it highly nsfw

5:47 PM

@terdon Assorted tech spammers, testosterone spammers, and black magic spammers, but it doesn't tend to include any weird self-promotion cases that anyone would argue about

Can you use MS to search by weight?

lot of it is probably generated by Markov chains I suspect. So that means it's literally sampling a distribution

something like this: metasmoke.erwaysoftware.com/post/109172

@Henders you kinda can, but have to go through metasmoke.erwaysoftware.com/flagging/conditions/sandbox and it's not perfect

not something you'd like to see at work

So reducing the amount of involuntary viewers of this kind of crap is a worthy cause

5:48 PM

(assuming it even mixes lol)

@Magisch at least this one is without images

some of those even decide to inline porn imagery

extra fun if you don't have the userscript to not display those by default

That one you linked was way too coherent

extra fun if you're on stack exchange because maybe you work with blender or apple and need work info

I've stopped using GasMask when FIRE started doing its job

5:49 PM

I mean stuff like this:

trifecta fun if it stayed up for 20 minutes and 40+ views because not enough of us were around to kill it

Man, the suspense is killing me!

@Magisch don't forget a spam closevote

I've seen them way too many times

people so don't want to see that stuff that serial reviewers complain if it appears in audits

safe to say normal site users didn't sign up for that

@quartata Was it so bad you are sat at your desk stunned..?

5:51 PM

Some of that stuff's really dangerous. . .

@Henders the link is blinding

metasmoke.erwaysoftware.com/post/107413

Okay, how about this: Do some research into how opt-in would really work. Can we get the bulk sites in a few posts?

Ssso, what was the conclusion?

@Magisch especially anonymous viewers didn't :P

5:52 PM

Oh, none still

Now that's a shitty mixing MCMC

The long tail might just not be worth messing with, or set to opt-out at the first sign of someone being irritated.

actually there's some better ones

... which is probably what @terdon and Shog have been saying this whole time.

Random question: What's the highest weight post we've ever caught?

5:53 PM

Hell, here's my own

(just purely out of curiosity)

@terdon sometimes we get written cp too

@DJMcMayhem 1200ish, probably

@DJMcMayhem 1180

although that always involves contacting a cm to scrub it

5:55 PM

Out again. Work needs done, apparently :P

Wow. What was that, long path viagra toothpaste?

haha. Viagra toothpaste

@terdon from a witch doctor

@terdon Long Path is often cunning, though

Viagra toothpaste probably straightens your teeth

2

Long path tool viagra toothpaste training in bangalore

2

5:57 PM

Pretty good thing considering surgery is the alternative

@M.A.R. Now I should try to make some.

okay I'm back. Had another thought.

What is our goal?

9

Sure seems like a simple question, but maybe we don't know the answer.

and now I'm gone again. Will have fun reading transcript later.

@Magisch well, brushing teeth is, by a landslide, the most common action in the world that needs training

@Undo cool, existential crisis

@M.A.R. never heard of toilet training, huh?

@Undo minimize the amount of spam people have to see

« first day (1570 days earlier) ← previous day next day → last day (2249 days later) »

Transcript for

Mar '186

Charcoal HQ

Where smoke is detected, diamonds are made, and we break thing...

dev-on-prod flag-based-fraternity programmatic-crowbars

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC

site design / logo © 2024 Stack Exchange Inc; legal

mobile