Lemmy.ca Support / Questions @lemmy.ca m-p{3} @lemmy.ca 11mo ago

Should we block OpenAI from scrapping the server?

Right now, robots.txt on lemmy.ca is configured this way

User-Agent: *
  Disallow: /login
  Disallow: /login_reset
  Disallow: /settings
  Disallow: /create_community
  Disallow: /create_post
  Disallow: /create_private_message
  Disallow: /inbox
  Disallow: /setup
  Disallow: /admin
  Disallow: /password_change
  Disallow: /search/
  Disallow: /modlog

Would it be a good idea privacy-wise to deny GPTBot from scrapping content from the server?

User-agent: GPTBot
Disallow: /

Thanks!

17 comments

Yes, please.

We can't stop LLM developers from scraping our conversations if they're determined to do so, but we can at least make our wishes clear. If they respect our wishes, then great. If they don't, then they'll be unable to plead ignorance, and our signpost in the road (along with those from other instances) might influence legislation as it's drafted in the coming years.
Yes. Ban them.

if ($http_user_agent = "GPTBot") { return 403; }
- Probably want == instead else we will all be forbidden
  
  I would have thought so too, but == failed the syntax check
  
  2023/08/07 15:36:59 [emerg] 2315181#2315181: unexpected "==" in condition in /etc/nginx/sites-enabled/lemmy.ca.conf:50
  
  You actually want ~ though because GPTBot is just in the user agent, it's not the full string.
  
  Strangely, = works the same as == with nginx. It's a very strange config format...
  
  https://nginx.org/en/docs/http/ngx_http_rewrite_module.html#if
  
  Look at me! I'm the GPTBot now!
- Thanks for empowering my lazyness =)
I'm on board for this, but I feel obliged to point out that it's basically symbolic and won't mean anything. Since all the data is federated out, they have a plethora of places to harvest it from - or more likely just run their own activitypub harvester.

I've thrown a block into nginx so I don't need to muck with robots.txt inside the lemmy-ui container.

# curl -H 'User-agent: GPTBot' https://lemmy.ca/ -i HTTP/2 403
- I imagine they rate limit their requests too so I doubt you'll notice any difference in resource usage. OVH is Unmetered* so bandwidth isn't really a concern either.
  
  I don't think it will hurt anything but adding it is kind of pointless for the reasons you said.
1000% yes. Please block them.
Is this even possible without all federated instances also prohibiting them?
- You take action where you can ;)
Yes
Are they even respecting those files?

But yeah, sure, it's worth trying!
- It's from the official documentation.
- Worth trying for what reason?
No, definitely not. Our work posted in the open is done so because we want it to be open.

It is understandable that not all work wants to be open, but access would already be appropriately locked down for all robots (and humans!) who are not a member of the secret club in those cases. There is no need for special treatment here.

17 comments