IRC spam

Hey folks,

Those of us who are still connected to the IRC channel surely have
noticed spam accounts posting random (machine generated?) garbage.
Moderators ban the nicks, but obviously, random ones come back.

Keeping on banning won't work. At Linaro, the somewhat controversial
move of requiring only registered users to join worked as a deterrent,
but the annoyance of registering may make less savvy users avoid the
channel.

Moving on to another platform (like Discord) will surely alienate a
good part of the community (myself included), but so will letting the
spam accounts run free (I'm almost giving up on the channel).

I don't know automated spam filtering for IRC (banning bots?), but we
usually "speak garbage" ourselves (aka pasting code), so false
positive rates will be high.

Anyone got better ideas?

cheers,
--renato

I also see little evidence that it would fix the problem of someone
having too much time on their hand and wants to be a nuisance.

Joerg

I don't think it’s a spam bot unfortunately, I think its just a mentally ill person. I'm not sure what we can do about the person though, unless we went to a private invite system (which we don't for Discord), I don't see how we could fix this even then.

AFAICS, this is not the problem.

The nicks are clearly randomly generated by smashing words together
and the content seems to be what comes out of a language model after
reading angry Facebook posts.

Registering is quick and easy and would filter 99% of the automated
accounts. Not 100% and not the people with too much time in their
hands. But those, nothing will.

It's like a bicycle lock: the bigger your lock, the more likely the
thief will steal someone else's bike. But someone wanting *your* bike
will take it, no matter the lock.

> I also see little evidence that it would fix the problem of someone
> having too much time on their hand and wants to be a nuisance.

AFAICS, this is not the problem.

The nicks are clearly randomly generated by smashing words together
and the content seems to be what comes out of a language model after
reading angry Facebook posts.

Perhaps, though I've seen them responding to people (not that that
rules out AI), etc. I'm with Joerg and Erich - I believe it's an
actual person. I actually saw them respond to a query yesterday - the
person seemed to get the answer they were looking through (reading
between the lines/etc of the strained grammar of this spammer person),
and their understanding/conclusion wasn't wrong & left before I could
clarify/provide further context - I worry about that happening more
regularly. Either this person scaring people off, and/or giving them
confusing/problematic advice, etc.

Registering is quick and easy and would filter 99% of the automated
accounts. Not 100% and not the people with too much time in their
hands.

Yep - having to create a new email address/verify it/etc does add some
friction (for intended and unintended users of course) - discord and
the like have far more incentive & resources to implement anti-spam
functionality than IRC does, unfortunately.

Yeah, saw it now, too. Sigh...

Let's try more OPs and badword lists, might help :slight_smile:

I also see little evidence that it would fix the problem of someone
having too much time on their hand and wants to be a nuisance.

AFAICS, this is not the problem.

The nicks are clearly randomly generated by smashing words together
and the content seems to be what comes out of a language model after
reading angry Facebook posts.

Registering is quick and easy and would filter 99% of the automated
accounts. Not 100% and not the people with too much time in their
hands. But those, nothing will.

It's like a bicycle lock: the bigger your lock, the more likely the
thief will steal someone else's bike. But someone wanting *your* bike
will take it, no matter the lock.

I'm fairly certain that this is the same person that has been spamming
the #dri-devel and #radeon channels on freenode for years. The pattern
of comments is they same and I've seen them mention people in comments on
#llvm who only join those other channels. #dri-devel requires users to identify
with the server and also uses extbans, but that hasn't really helped.

I'm not sure how to solve this, but I think having more ops
(spread across all timezones) would help.

-Tom

I’ll comment from the perspective of someone that is in the Mesa, #dri-devel, #radeon channels myself and have watched their behaviour over the years. This is a real person that spams a load of information into a channel about their understanding of how hardware works.
I have no idea what their goal is for spamming this information, could be some desire for acceptance from perceived smartness. Or something as simple as wanting to be hired for their “brilliance”. Hard to tell.
A major issue with their personality is that they will retaliate against anyone that tries to stop their ranting, and they become hostile with their phrasing very quickly because of it. Just check the logs for them retaliating against anyone that has kickbanned them.
Another issue is that depending on their mood of the day, they may be entirely lost to any form of reasoning, which makes it difficult for any communication.
So just to reiterate, they are a real person but are difficult to deal with.
On that note, they aren’t completely impossible to work with in some cases, it just might require accepting getting attacked for a few weeks.
I’m a channel operator in one of the Mesa related IRC channels and have had success in communicating with them that their behaviour is not conducive to the environment that we were attempting to create in the channel.
This took a bit of coaxing on their “good” days, and communicating with them while being attacked for around a month on end. At the end of this month-long attack and communication I was able to get them to understand that they aren’t welcome to the channel.
They no longer enter the channel that I moderate; I managed to get through to them on some level at least.
Sadly this sort of baby sitting of a user shouldn’t be required and requiring some thick skin to get through their harsh comments is difficult.
More moderation will “work” but while they are rampaging, you’re going to still have to watch the channel and you’ll get a few lines of harassing text while an op takes a bit of time to see them (and sometimes even perceive them, on “good” days they make comments that make some sense initially).

Seems like a useful machine-learning recognition task… Lots of data… useful to automate… Low cost of false-negatives…

We went with old school solutions, less sexy I know, but seems to be effective.

@Renato, do you still feel there is much spam?

P.S. I’m going to switch to your ML approach once ready :wink:

We went with old school solutions, less sexy I know, but seems to be effective.
@Renato, do you still feel there is much spam?

No, it's good now, thanks to all that volunteered to moderate.

P.S. I'm going to switch to your ML approach once ready :wink:

This is how we start the robot uprising... :slight_smile:

I'll comment from the perspective of someone that is in the Mesa, #dri-devel, #radeon channels myself and have watched their behaviour over the years. This is a real person that spams a load of information into a channel about their understanding of how hardware works.

Hi Ryan,

Yeah, half-way through I realised, as some other people pointed out,
it is a real person.

I identified as a robot for two main reasons:
1. Really confusing sentences, intermixing completely unrelated
subjects and starting a new phrase before the old one was finished.
This is very typical of markov chain or cheap language models trained
with a small subset of unrelated texts. It's also unfortunately common
in people who can't help but work on multiple trains of thought (like
me).
2. The random names and fast rejoin were consistent with either a
mindless bot, or a very persistent individual. I couldn't fathom why a
person would do that, so I assumed bot. That was on me for not seeing
it far enough.

I have no idea what their goal is for spamming this information, could be some desire for acceptance from perceived smartness. Or something as simple as wanting to be hired for their "brilliance". Hard to tell.
A major issue with their personality is that they will retaliate against anyone that tries to stop their ranting, and they become hostile with their phrasing very quickly because of it. Just check the logs for them retaliating against anyone that has kickbanned them.
Another issue is that depending on their mood of the day, they may be entirely lost to any form of reasoning, which makes it difficult for any communication.

There are plenty of life situations that make smart people behave
erratically or crack entirely, most of them mundane to the majority of
people. It's not fun.

On that note, they aren't completely impossible to work with in some cases, it just might require accepting getting attacked for a few weeks.
I'm a channel operator in one of the Mesa related IRC channels and have had success in communicating with them that their behaviour is not conducive to the environment that we were attempting to create in the channel.
This took a bit of coaxing on their "good" days, and communicating with them while being attacked for around a month on end. At the end of this month-long attack and communication I was able to get them to understand that they aren't welcome to the channel.
They no longer enter the channel that I moderate; I managed to get through to them on some level at least.

I'm impressed with your care and stamina. Not many people I know would
have gone that far. Thank you for doing that.

Sadly this sort of baby sitting of a user shouldn't be required and requiring some thick skin to get through their harsh comments is difficult.
More moderation will "work" but while they are rampaging, you're going to still have to watch the channel and you'll get a few lines of harassing text while an op takes a bit of time to see them (and sometimes even perceive them, on "good" days they make comments that make some sense initially).

Any kind of barrier should be enough to get bots and persistent
individuals to stop. Moderators, registration, etc.

I think in the end it worked out well. People quickly realised those
words were meaningless (in the context of the particular channel), and
worked to stop the flood.

cheers,
--renato

Is the channel now purposefully invite-only? If so, how is one supposed to join?

Thanks,
Stefan

It's not. More people were added as moderators to curb the problem
instead. Are you having trouble joining the channel?

I’m aware of the efforts. As of today it looks like it is invite-only. I’m getting the error “473: #llvm Cannot join channel (+i)”, which should mean it is invite only, right?

Thanks,
Stefan

I fixed this a few hours ago I think :slight_smile:

It was made invite only apparently. I was not online at the time.

IMHO, there are way better ways to make this work than invite only,

for example, we can require authenticated users. Invite only w/o a

well documented way to get the invite seems like the opposite of inclusive.