Hacker News Reader

Comment by mrtksn | Hacker News Reader

After the market crash a day ago, there was a joke on Twitter about Musk cancelling the offer, burning a billion and buying Twitter later at a half price.

Conspiracies and jokes aside, how one determines(at contractually acceptable certainty) what accounts are absolutely fake or bot account? It's not like they ticked the checkbox saying "I am a fake account".

If the sellers and buyers are not on the same page here, wouldn't that drag in court for years maybe?

After all, maybe Musk will actually burn a billion only to buy Twitter later at a half price or less and save $20B+. Twitter might end to be much more cheaper after a stock crash due to the bearish markets and that fake users scandal(?) that Musk exposed(?).

View on HN

> how one determines(...) what accounts are absolutely fake

Right now, Twitter's ability to ferret out fake accounts is just completely ineffective. In the communities I follow, I see literally hundreds of reply posts under each tweet of certain people, all with the same profile picture as the OP, all tweeting about giveaways.

Twitter should be focussed on cleaning up their product, and one of the variables required for that is knowing what amount of users are being used to spam scam messages. The fact that they can't means twitter's development isn't prioritising it. A ballpark estimate of the order of magnitude of that figure should be possible for them if they were in any way actively trying to solve the problem, which they are obviously not doing.

They can do much better if they wanted to. They don't.

There are very obvious patterns that I as a single person can see about fake accounts.

Twitter, a huge tech company with access to a lot of engineers and AI, can probably do much better than me.

They don't because they don't want to.

Except when the market as a whole goes down it impacts everyone who was bringing money to the table in this deal. I wonder if the financing he's gathered is getting nervous. It might be harder for him to raise the money again, even if the price is better.

Tesla stock also suffered quite a lot, AFAIK that was how Musk was going to finance the deal.

That said, Tesla doesn't have "more than %5 fake users" problem that came out just as the deal was closing. I guess Musk is a luck man.

I could see the captcha solution someone mentioned being possible.

Take a randomly selected group of 10,000 active users, show them a captcha, and look at the percentage of respondents.

Random captchas ("We need to re-confirm you're a human") sounds like a great idea, although annoying to someone just wanting to post a quick 20-word tweet.

But it would have the useful side effect of slowing people down from tweeting impulsively, a practice that demeans the conversation and sometimes ends careers.

The GP post wasn't describing a method for stopping fake accounts, they were describing how to sample what percentage of accounts are fake.

this would not work well because the sample would be biased

newer accounts are much more likely to be bots compared to older ones

you would have to sample IDs by age and create a distribution based on ID age

And then you can extrapolate how many overall are fake

This may work

bots are not fake, they can be useful accounts. I have one that still works, another kept being taken down so i stopped reinstating it. To say that twitter is ran by schoolchildren would be understatement

As a private company, Musk can also set his own management fees / owner compensation / call it what ever

1) Pay $1B fine to Twitter

2) Buy cheaper

3) Get a $1B chairman first year bonus from Twitter

He'll probably find a way to write it off on taxes as well

The $1B are lost in (2). He will have to pay about $1B more if he has to pay the fee to twitter because then the company is worth $1B more.

It's not $1B exactly because he already owns some shares so he doesn't have to pay extra for those.

The fact that it goes to a company he later acquires makes it seem like an obvious move. Is it?

Here’s a program that can detect bots

  For user in Twitter:
    Confirmation = random(number)
    Send user message(“if you are not a bot reply with”, confirmation)

Variations can be devised but the real risk is that it also identifies inactive accounts.

You know how iPhones can extract verification codes from SMS? The botmasters will simply create a regex to extract the number and reply accordingly.

Just use a captcha image. Not foolproof, but would eliminate a huge swath of bots.

They don't even try this. Why?

Then they will simply solve the captchas using captcha solvers or humans paid pennies per captcha. You do something else and the botmasters will respond.

It's really hard to get rid of bots and fake accounts. Also, false positives are expensive because you end up annoying a real human and as a result your mistake doesn't disappear until you make it right.

The cat and mouse game of bots vs captchas has been going on for a long time and has become increasingly sophisticated. "Just use a captcha image" is a 10 year old solution, if not more.

How would they know to write that code if they were not expecting it? Just do it as a one-off on a particular day without any warning. And you can show it to some small sample % of total users then extrapolate, so that the average botmaster wouldn’t even notice.

I guess you can run a sting operation, it may work.

Beat botnets with this one simple trick!

Aren't you just suggesting a very simple CAPTCHA challenge? This is an area that has seen an arms race for over a decade of platforms coming up with increasingly complex "human-intelligence" challenges, and botmasters coming up with increasingly ingenious ways to bypass them (including, sometimes, using real human intelligence)

Reply on news.ycombinator.com