Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Feed the bots

maurycyz.com

Post reply on HN
Print view View original

171–180 of 216 posts

Re: Feed the bots

#171

Post by xyzal »

markus_zhang wrote: post #41

I have always recommended this strategy: flood the AI bots with garbage that looks like authentic information so that they need actual humans to filter the information. Make sure that every site does this so they get more garbage than real stuffs. Hike up the proportion so that even ordinary people eventually figure out that using these AI products has more harm than use because it just produces garbage. I just don't…

Feed them this. https://github.com/emergent-misalignment/emergent-misalignme...

Re: Feed the bots

#172

Post by nl »

63stack wrote: post #170
nl wrote: post #164

Earlier quoted context omitted.

You realise that LLMs are already better at deciphering this than humans?

There are multiple people claiming this in this thread, but with no more than a "it doesn't work stop". Would be great to hear some concrete information.

Here you go:

https://chatgpt.com/share/68ff4a65-ead4-8005-bdf4-62d70b5406...

Re: Feed the bots

#173

Post by 63stack »

nl wrote: post #172
63stack wrote: post #170

Earlier quoted context omitted.

There are multiple people claiming this in this thread, but with no more than a "it doesn't work stop". Would be great to hear some concrete information.

Here you go: https://chatgpt.com/share/68ff4a65-ead4-8005-bdf4-62d70b5406...

I think OP is claiming that if enough people are using these obfuscators, the training data will be poisoned. The LLM being able to translate it right now is not a proof that this won't work, since it has enough "clean" data to compare against.

Re: Feed the bots

#174

Post by liqilin1567 »

wodenokoto wrote: post #68
blackhaj7 wrote: post #66

Can someone explain how this works? Surely the bots are still hitting the pages they were hitting before but now they also hit the garbage pages too?

In authors setup, sending Markova generated garbage is much lighter on resources than sending static pages. Only bots will continue to follow links to the next piece of garbage and thus he traps bots in garbage. No need to detect bots, they reveal themselves. But yes, all bots start out on an actual page.

Seems like these garbage pages can't trap bots. People discussed it in this thread: https://news.ycombinator.com/item?id=45711987

Re: Feed the bots

#175

Post by marginalia_nu »

pavel_lishin wrote: post #98

The blog post ( https://maurycyz.com/misc/the_cost_of_trash/ ) says that gzip bombs don't work particularly well: > Gzip only provides a compression ratio of a little over 1000: If I want a file that expands to 100 GB, I’ve got to serve a 100 MB asset. Worse, when I tried it, the bots just shrugged it off, with some even coming back for more. I thought a gzip bomb was crafted to explicitly be virtually unlimited in t…

You can do that with zip, but not gzip.

The problem with gzip bombs in the web context in general is that they operate on the naive assumption that the client will decompress the payload entirely. This is very rarely the case, and you kinda have to go out of your way to make that happen[1], and it really only makes sense if you're looking at some binary format that can't be truncated like you can with HTML.

Instead most if not all clients will use some form of streaming decompression, with a termination criterion, and to the extent stuff is decompressed in full, very rarely will anything be decompressed in full and held in memory, as that would nuke your crawler the first time you ran into a website mirroring linux ISOs.

[1] This is the zlib api for decompressing a gzip file: https://refspecs.linuxbase.org/LSB_3.0.0/LSB-Core-generic/LS...

Re: Feed the bots

#176

Post by bombcar »

peterlk wrote: post #45
markus_zhang wrote: post #41

I have always recommended this strategy: flood the AI bots with garbage that looks like authentic information so that they need actual humans to filter the information. Make sure that every site does this so they get more garbage than real stuffs. Hike up the proportion so that even ordinary people eventually figure out that using these AI products has more harm than use because it just produces garbage. I just don't…

LLMs can now detect garbage much more cheaply than humans can. This might increase cost slightly for the companies that own the AIs, but it almost certainly will not result in hiring human reviewers

They can’t easily detect garbage; they can easily detect things that are outside the dataset (for some value of such).

Which means that real “new” things and random garbage could look quite similar.

Re: Feed the bots

#177

Post by nl »

63stack wrote: post #173
nl wrote: post #172

Earlier quoted context omitted.

Here you go: https://chatgpt.com/share/68ff4a65-ead4-8005-bdf4-62d70b5406...

I think OP is claiming that if enough people are using these obfuscators, the training data will be poisoned. The LLM being able to translate it right now is not a proof that this won't work, since it has enough "clean" data to compare against.

If enough people are doing that then venacular English has changed to be like that.

And it still isn't a problem for LLMs. There is sufficient history for it to learn on, and in any case low resource language learning shows them better than humans at learning language patterns.

If it follows an approximate grammar then an LLM will learn from it.

Re: Feed the bots

#178

Post by 63stack »

nl wrote: post #177
63stack wrote: post #173

Earlier quoted context omitted.

I think OP is claiming that if enough people are using these obfuscators, the training data will be poisoned. The LLM being able to translate it right now is not a proof that this won't work, since it has enough "clean" data to compare against.

If enough people are doing that then venacular English has changed to be like that. And it still isn't a problem for LLMs. There is sufficient history for it to learn on, and in any case low resource language learning shows them better than humans at learning language patterns. If it follows an approximate grammar then an LLM will learn from it.

I don't mean people actually conversing like this on the internet, but using programs like what is in the article to feed it to the bots only.

Re: Feed the bots

#179

Post by sroelants »

csomar wrote: post #157
mcdeltat wrote: post #123

Maybe a dumb question but what exactly is wrong with banning the IPs? Even if the bots get more IPs over time, surely storing a list of bans is cheaper than serving content? Is the worry that the bots will eventually cycle through so many IP ranges that you end up blocking legit users?

They are usually using residential IPs through SOCK5. I am not sure how they are getting these residential IPs but it is definitively suspicious. So by blocking these IPs, you are blocking your users. (ie: in many coffeshops, I get the "IP Blocked" banner, my guess is that they are running software on unsuspecting users to route this traffic).

They use scammy providers like Bright Data[1] that let app authors embed their malware (for a compensation, I'm sure) which turns users' devices into crawler proxies.

[1]: https://brightdata.com/trustcenter/sourcing

Re: Feed the bots

#180

Post by nl »

63stack wrote: post #178
nl wrote: post #177

Earlier quoted context omitted.

If enough people are doing that then venacular English has changed to be like that. And it still isn't a problem for LLMs. There is sufficient history for it to learn on, and in any case low resource language learning shows them better than humans at learning language patterns. If it follows an approximate grammar then an LLM will learn from it.

I don't mean people actually conversing like this on the internet, but using programs like what is in the article to feed it to the bots only.

This is exactly like those search engine traps people implemented in the late 90s and is roughly as effective.

But sure.

Post reply on HN