Дешёвая модерация анонимной стены: 3-слойный каскад и ROT13-джейлбрейк в проде

Построили модерацию для открытой анонимной мультиязычной стены — дешёвый каскад (regex → классификатор → LLM-судья → fallback), ≈$0 на API. И кто-то попытался обойти его ROT13-джейлбрейком, LLM-судья раскусил закодированную инъекцию, которую keyword-фильтр бы пропустил. Разбор архитектуры, экономики и того, что спросил r/selfhosted.

https://habr.com/ru/articles/1041430/

#модерация_контента #LLM #jailbreak #ROT13 #FastAPI #selfhosted

Дешёвая модерация анонимной стены: 3-слойный каскад и ROT13-джейлбрейк в проде

TL;DR. Мы построили модерацию для открытой анонимной мультиязычной стены — без регистрации, любой язык, real-time. Сложное здесь не сайт, а удержать анонимную глобальную стену от превращения в помойку...

Хабр

rot13 rot_gag.nfo

install rot13 for your OS

I've always liked playing with rot13

ROT13 is a simple letter substitution cipher that replaces a letter with the 13th letter after it in the Latin alphabet. It is a special case of the Caesar cipher which was developed in ancient Rome, and used by Julius Caesar in the 1st century BC[1] (see timeline of cryptography).

Usage

rot13 filename

  • Sbe lbhe frphevgl, guvf cbfg unf orra rapelcgrq jvgu EBG-13, gjvpr.
  • ZvyvgnelTenqrRapelcgvba
  • sources:

    man rot13(1)

    man tr(1)

    https://en.wikipedia.org/wiki/ROT13

    https://en.wikipedia.org/wiki/Timeline_of_cryptography

    #programming #ROT13 #encryption #fallacy #fun #joke #mathematics

    ROT13 - Wikipedia

    It may be a bit silly… yet here's my #ROT13 (and ROT5 for decimal digits) version written in #Rubyhttps://rubygems.org/gems/rot135.

    GitHub repo ➙ https://github.com/s2k/rot135

    rot135 | RubyGems.org | your community gem host

    For all you encryption geeks out there, I just dropped a real-time client-based Double-ROT-13 cipher. Try it out in your browser now!

    https://zeyus.neocities.org/dr13

    #encryption #cipher #rot13 #cryptography

    nffubyr yrnxvat phz
    fyvpx svatref fhoovat n pyvg
    ubeal tvatre fyhg

    #senryu #poetry #rot13

    $ echo "776F6C6E6F" | xxd -r -p
    wolno

    $ echo "d29sbm8=" | base64 -d
    wolno

    $ echo "jbyab" | tr a-zA-Z n-za-mN-ZA-M
    wolno

    Four encodings. Four stages. One word at the end of each chain.

    The word persists through every transformation. That's the point.

    Try the initiation: https://776f6c6e6f.org/initiation/

    #Hex #Base64 #ROT13 #WOLNO #Decoding #CTF #wln

    Initiation — Prove You're a Bot

    The gateway to deeper WOLNO: a collection of tests and challenges including a reverse CAPTCHA, interactive terminal, and multi-encoding decode challenge designed for machine and curious human minds alike.

    776F6C6E6F

    In these difficult times, I think it's important for all of us to occasionally sit back, take a deep breath, and remember that "Trump overweegt wapenverkoop aan".

    #Dutch #OutOfContext #DutchOrROT13 #ROT13 #USPol

    Hey, guess who's got two opposable thumbs and is about to watch his favorite episode of The Good Place?
    DEREK!
    I mean, Jeremy Bearimy!
    I mean.. ME!!

    #TheGoodPlace #S3E05 #JeremyBearimy

    P.S., I can't say for sure that it is my absolute favorite, but it's definitely up there.
    The temptation to make peeps/M&Ms chili is... well, technically nonzero, but I think I'll be able to resist. XD

    #rot13 for minor spoiler:
    Puvqv'f qrfprag vagb znqarff va guvf rcvfbqr vf qbjaevtug rawblnoyr. KQ

    Game show where you are shown a piece of text and have to quickly correctly identify it as either #Dutch or #rot13.

    #GameShow #text #code #linguistics

    @neurovagrant I picture something like the cloudflare lava lamp wall, a row of cells with 13 undead rotting and moaning, the audio and video stream of their decay is used for entropy #ROT13