Conversation

Notices

1. Embed this notice
   jwz (jwz@mastodon.social)'s status on Thursday, 16-Jan-2025 13:56:23 JST jwz

   Exterminate all rational AI scrapers.

   Today I added an infinite-nonsense honeypot to my web site just to fuck with LLM scrapers, based on a "spicy autocomplete" program I wrote about 30 years ago. Well-behaved web crawlers will ignore it, but those "AI" people.... well, you know how they are.

   I'm intentionally not linking to it here, but I'll bet you can find it pretty easily. It's kinda funny.
   https://jwz.org/b/ykgX

   • Embed this notice
     jwz (jwz@mastodon.social)'s status on Friday, 17-Jan-2025 06:14:36 JST jwz

     in reply to

     What's the most zipbomb-like JPEG or PNG that can be constructed? Small file size, plausible dimensions, but massive memory footprint?

   • Embed this notice
     mhoye (mhoye@mastodon.social)'s status on Friday, 17-Jan-2025 06:19:06 JST mhoye

     in reply to

     @jwz This guy's been working on it for a long time: https://www.bamsoftware.com/hacks/deflate.html

     The current best looks to be "420 bytes of PNG expand to represent 50 gigapixels and 141 GB in memory".

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Infoseepage #StopGazaGenocide (infoseepage@mastodon.social)'s status on Friday, 17-Jan-2025 06:37:07 JST Infoseepage #StopGazaGenocide

     in reply to

     @jwz Would love to see more of this designed to poison the well. How about big hidden text payloads on pages that randomly pair words from the page in a grammatically correct manner but completely nonsensical manner? Train on this, motherfuckers.

   • Embed this notice
     Infoseepage #StopGazaGenocide (infoseepage@mastodon.social)'s status on Friday, 17-Jan-2025 06:39:23 JST Infoseepage #StopGazaGenocide

     in reply to

     @jwz The sooner LLMs start spouting out stuff that is recognizable to even the most baseline pleb as gibberish, the better. Lets make it happen. Lets poison the well. Right now, far too often, these things spit out gibberish that people earnestly consider to be knowledge. These are NOT knowledge engines.

   • Embed this notice
     marcusb (marcusb@mastodon.sdf.org)'s status on Friday, 17-Jan-2025 07:51:34 JST marcusb

     in reply to

     • Infoseepage #StopGazaGenocide

     @Infoseepage @jwz there are several other tools with similar goals. I wrote one with static sites in mind (no server-side dependencies, all non-sense pre-generated. https://marcusb.org/hacks/quixotic.html). Others I know about are: https://github.com/Fingel/django-llm-poison, https://codeberg.org/MikeCoats/poison-the-wellms, https://codeberg.org/timmc/marko/, https://zadzmo.org/code/nepenthes/, and https://github.com/earthboundkid/heffalump.

   • Embed this notice
     Infoseepage #StopGazaGenocide (infoseepage@mastodon.social)'s status on Friday, 17-Jan-2025 07:51:34 JST Infoseepage #StopGazaGenocide

     in reply to

     • marcusb

     @marcusb @jwz This is great. Love it!

   • Embed this notice
     tom jennings (tomjennings@tldr.nettime.org)'s status on Saturday, 18-Jan-2025 05:26:10 JST tom jennings

     in reply to

     • mhoye

     @mhoye @jwz

     For eg corp social media site avatar image upload, there's usually a maximum file size limit (in addition to pixel). For that case you'd want to bound file size, so the question would be, what's the largest image compressible into eg. 16KB?

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Michele Ortiz (ortiz0852@mastodon.social)'s status on Saturday, 18-Jan-2025 08:03:10 JST Michele Ortiz

     in reply to

     @jwz Hello dear happy new year 2025 will be your best💯🥳 year connect with Mrs William Sarah 👇👇👇👇👇👇👇👇👇👇

     https://www.facebook.com/Williamsarahfx12

     You'll definitely* thank me later .