/tech/ - Tech


Mode: Reply

Max message length: 8192


Max file size: 20.00 MB

Max files: 3


(used to delete files and postings)


Remember to follow the rules

(60.78 KB 909x555 post_reply.jpg)
Comrade 05/04/2020 (Mon) 19:31:07 No. 1577
The enemy uses bots. What do we know about how they work? Is there any literature that would help in writing our own? >>478855
>>1577 >What do we know about how they work? They do what users usually do, but they're automated. To write a bot, all you have to do is determine what exactly the browser sends when you make a post, and send the same information in your code. The rest is trivial, just provide some text for the bot to send. Just make sure you don't send like 10 posts a second because it will almost immediately trigger the anti-ddos filter, possibly banning your ip.
waste of time, the people at nsa, cia, jidf and correct the record get paid to write spam bots with their unlimited funds, don't fantasize symmetric warfare
>>1580 >The rest is trivial, just provide some text for the bot to send. This is the only part I'm concerned about automating actually.
I used to make tons of scripts to game polls, etc. before captchas became ubiquitous.
>>1587 This part if you want it to be of any quality you may need some form of machine learning for. https://www.youtube.com/watch?v=aircAruvnKk Watch this and come back if you got any questions. Alternatively you could just have a list that randomises phrases, for example it opens /pol/ and goes "[Hey/Morning] [cucks/simps/niggas], why is [porky/jew/irish] making so much money?" etc etc
Just going to leave this here https://arxiv.org/abs/1506.05869
Some /pol/tard has tracked shills on his board. Here is his findings: https://archive.4plebs.org/pol/thread/256579275/
>>1641 That thread scared me a bit. Maybe I should go back to uni to learn about AI, but I suck at maths so not sure how much good that would do me.
>>1646 All you need is a (quite good) understand of linear algebra with AI, statistics would probably help too.
>>1641 That is very interesting thank you. It would be interesting to see if shillbots on reddit can be detected (I've heard of using AI to figure out if AI has been used) and what percentage of the comments are from them. It's probably much, much more than I realized. Fuck.
>>1656 This is sort of like the “do we live in a simulation” argument. If AI are possible to this level, then there’s pretty much no reason there aren’t thousands upon thousands of them. Reddit themselves could probably easily find it out by measuring the time taken to “write a post”, but it’s not like they’d ever tell us that. I’ll look into this from a technical view soon, if anyone has any other good resources on recognising AI let me know
>>1641 holy shit this is some spooky stuff, how do I know any of you are actually real?
>>1657 An interesting thought I had is that the next generation of AI that will be trained might be trained on a dataset that contains AI generated text itself, if it becomes hard to filter that out. Presumably that would exaggerate the nonsense and artifacts that come with using those tools, but I have no idea what actually will happen.
>>1658 How do I know YOU'RE actually real? Hm? HMMM?!
>>1641 In the thread >Have I been talking to actual AI possibly >designed to keep me in a haze of rage and bullshit the entire time? yes. >>1674 You can also pullute the data. Type random sentences without meaning For example to be a good take from a person is a day and a new normal circumstances and then perhaps a day or two levels are very important for a long term partner or whatever is in Mexico you can bet your head in Mexico city or whatever is your best friend for example and I think a good alternative would give them are good to you and the way it works for a promotion or the way it does it is counter. Eat that, bot.
>>1698 fug
>>1698 I'm an bot. :) :)
>>1699 They can filter on coherency too you know. They probably have an ever-expanding set of filters that censor different datasets to create different kinds of bots each time. They'll measure the response amount and type too to see if and how the post worked on the subjects targeted. Maybe we could compile some of the resources the /pol/ thread found (like the banned original research reddit OP image) for a thread on /leftypol/ for this stuff. It's important for all anons to become aware of this, and if it discourages people from engaging in meaningless twitter or reddit drama all the better.
>>1712 do it
>>1712 Yes, do it. This [info itt] was eye opening even though I already knew about it. You see we're aware of the government "spying" and "tracking" you but bots have hijacked internet discussion. One thing that's odd is that sometimes you read about this shit in the news and it's played off as something that's been going on for a long time so it's less of a big deal. The reality is that this is relatively recent and the full effects are unknown to us.
(679.38 KB 1074x1419 3bbko1axdmn41.png)
>>1587 There is definitely no shortage of relevant training bodies of data out there, as well as neural net frameworks you could probably get running with just a bit of Python or C++ knowledge. https://en.wikipedia.org/wiki/Comparison_of_deep-learning_software >>1596 To add to this, you could even just take already-existing posts from a board/subreddit and change a word or two to pass the more basic of spam filters, then repost it. >>1657 >but it’s not like they’d ever tell us that. There is over $100m at stake within that site. If you really think about it, added to their tiny moderation teams, they're really blurring the lines between their project and regular corporate MSM at this point. Mods have been bribed before. What makes you think that certain bot networks aren't *desirable* to Reddit's staff? >>1698 Synonyms. Bots suck with synonyms. Or so it has been vocalised to me.
>>1585 what's funny is that by solving those "i'm not a bot" captchas people are literally training google's bots to be able to solve those captchas.
>>1596 You don't need neural networks for text generation Non-meme ML exists.
(1.43 KB 404x250 Flag-Poland.jpg)
I have experience with writing bots, i can make you imageboard bot for cheap
>>2945 can you write a bot that can fall in love with me?
>>1958 >neural networks >meme literally they are the oldest type of ML invented dude
>>2950 I can write a bot that tells me you to stuff a dildo up your ass. 10 PRINT "Stuff a dildo up your ass" 20 GOTO 10


no cookies?