this post was submitted on 04 Jul 2024
127 points (97.7% liked)

Technology

59374 readers
3169 users here now

This is a most excellent place for technology news and articles.


Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related content.
  3. Be excellent to each another!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, to ask if your bot can be added please contact us.
  9. Check for duplicates before posting, duplicates may be removed

Approved Bots


founded 1 year ago
MODERATORS
 

Cloudflare, the publicly traded cloud service provider, has launched a new, free tool to prevent bots from scraping websites hosted on its platform for data to train AI models.

Some AI vendors, including Google, OpenAI and Apple, allow website owners to block the bots they use for data scraping and model training by amending their site’s robots.txt, the text file that tells bots which pages they can access on a website. But, as Cloudflare points out in a post announcing its bot-combating tool, not all AI scrapers respect this.

you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 7 points 4 months ago (2 children)

I’m not much of a programmer and I don’t host any public sites, but how feasible would it be to build an equivalent of Night Shade but for LLMs that site operators could run?

I’m thinking strategies akin to embedding loads of unrendered links to pages full of junk text. Possibly have the junk text generated by LLMs and worsened via creative scripting.

It would certainly cost more bandwidth but might also reveal more bad actors. Are modern scrapers sophisticated enough to not be fooled into pulling in that sort of junk data? Are there any existing projects doing this sort of thing?

[–] [email protected] 3 points 4 months ago

To get more direct to the point you could use those unrendered dummy links to ban whatever IPs click them.

With the vast amounts of training data and how curated they're becoming (Llama and Claude are going that direction) it's infeasible to actually poison a large model to this degree.

load more comments (1 replies)