Comment on Perplexity AI is complaining their plagiarism bot machine cannot bypass Cloudflare's firewall
Ekybio@lemmy.world 7 months ago
Can someone with more knowledge shine a bit more light on this while situation? Im out of the loop on the technical details
AI crawlers tend to overwhelm websites by doing the least efficient scraping of data possible, basically DDOSing a huge portion of the internet. Perplexity already scraped the net for training data and is now hammering it inefficiently for searches.
Cloudflare is just trying to keep the bots from overwhelming everything.
Cloudflare runs as a CDN/cache/gateway service in front of a ton of websites. Their service is to help protect against DDOS and malicious traffic.
A few weeks ago cloudflare announced they were going to block AI crawling (good, in my opinion). However they also added a paid service that these AI crawlers can do, so it actually becomes a revenue source for them.
This is a response to that from Perplexity who run an AI search company. I don’t actually know how their service works, but they were specifically called out in the announcement and Cloudflare accused them of “stealth scraping” and ignoring robots.txt and other things.
It should be pointed out that Cloudflare didn’t say they were going to block AI traffic, they give you the option to. The service is a free opt-in for people who want it.
they don’t outright block ai crawlers. but they added some new tools and options for managing or blocking ai bot traffic which the cloudflare customer can choose to use or to not use.
im running a free educational resource and i let the crawlers hit my site all they want because its useful knowledge and it’s served to them from cloudflare’s free tier cache.
But the website owner can still choose to continue blocking them right? Without using additional stuff like Anubis that is.
A few weeks ago cloudflare announced they were going to block AI crawling (good, in my opinion). However they also added a paid service that these AI crawlers can use, so it actually becomes a revenue source for them.
I think it’s also worth pointing out that all of the big AI companies are burning through cash at an absolutely astonishing rate, and none of them are anywhere close to being profitable. So pay-walling the data they use is probably gonna be painful for their already-tortured bottom line (good).
they already said they wernt profitable, they are trying to keep on life support til the VC funds run out.
It’s more than simply astonishing, it’s mind-blowingly bonkers how much money they have to burn to see ANY amount of return. You think a normal company is bad, blowing a few thousand bucks on materials, equipment, and labor per day in order to make a few bucks revenue (not profit)? AI companies have to blow HUNDREDS OF BILLIONS on massive data center complexes in order to train their bots, and then the energy cost and water cost of running them adds a couple more million a day. ALL so they can make negative hundreds of dollars on every prompt you can dream of.
The ONLY reason AI firms are still a thing in the current tech tree is because Techbros everywhere have convinced the uberwealthy VC firms that AGI is RIGHT AROUND THE CORNER, and will save them SO much money on labor and efficiency that it’ll all be worth it in permanent, pure, infinite profit. If that sounds like too much of a pipe dream to be realistic, congratulations, you’re a sane and rational human being.
It’s more than simply astonishing, it’s mind-blowingly bonkers how much money they have to burn to see ANY amount of return
See, that’s the trick, and it’s used by LOADS of startups:
You don’t actually have to see a return… You just have to have a good story showing there MAY be a GIANT return. The founders collect enormous salaries (Funded by VC dollars, not their own), they burn through the money to create more illusion, then ask for more, then burn through that, foretelling of the coming days when the money is just coming!
Meanwhile, just before it’s “projected” to become insanely profitable, they sell out to someone, walk away with a giant check, and the product evaporates.
BetaDoggo_@lemmy.world 7 months ago
Perplexity (an “AI search engine” company with 500 million in funding) can’t bypass cloudflare’s anti-bot checks. For each search Perplexity scrapes the top results and summarizes them for the user. Cloudflare intentionally blocks perplexity’s scrapers because they consider them to be malicious traffic. Perplexity argues that their scraping is acceptable because it’s user initiated.
Personally I think cloudflare is in the right here. The scraped sites get 0 revenue from Perplexity searches (unless the user decides to go through the sources section and click the links) and Perplexity’s scraping is unnecessarily traffic intensive since they don’t cache the scraped data.
lividweasel@lemmy.world 7 months ago
That seems almost maliciously stupid. We need to train a new model. Hey, where’d the data go? Oh well, let’s just go scrape it all again. Wait, did we already scrape this site? No idea, let’s scrape it again just to be sure.
rdri@lemmy.world 7 months ago
First we complain that AI steals and trains on our data. Then we complain when it doesn’t train. Cool.
ubergeek@lemmy.today 7 months ago
I think it boils down to “consent” and “remuneration”.
I run a website, that I do not consent to being accessed for LLMs. However, should LLMs use my content, I should be compensated for such use.
So, these LLM startups ignore both consent, and the idea of remuneration.
Most of these concepts have already been figured out for the purpose of law, if we consider websites much akin to real estate: Then, the typical trespass laws, compensatory usage, and hell, even eminent domain if needed ie, a city government can “take over” the boosted post feature to make sure alerts get pushed as widely and quickly as possible.
spankmonkey@lemmy.world 7 months ago
They do it this way in case the data changed, similar to how a person would be viewing the current site. The training was for the basic understanding, the real time scraping is to account for changes.
It is also horribly inefficient and works like a small scale DDOS attack.
jballs@lemmy.world 7 months ago
It’s worth giving the article a read. It seems that they’re not using the data for training, but for real-time results.