Consider https://arstechnica.com/robots.txt or https://www.nytimes.com/robots.txt and how they block all the stupid AI models from being able to scrape for free.
The robots.txt construct is completely voluntary and some bots use it to specially target content.
In my opinion, anyone relying on this to protect their content has no business publishing anything online.
See: https://en.m.wikipedia.org/wiki/Robots.txt
Of course it's voluntary, but if entities like OpenAI say they will respect it then presumably they really will.
Couple of things:
- Do you believe anything coming out of OpenAI when it's abundantly clear that they'll say anything to protect their bottom line.
- OpenAI are not the only people harvesting data and selling it to interested parties.
- There is no legal requirement to adhere to the standard and I'd be shocked if any court in the USA could understand the issue, let alone enforce a voluntary standard.
- The amount of automated data collection online is staggering. On my own services it accounts for 50% of the hits. Good luck with policing that.
I agree with your points 2-4 but I have observed on my own website that the crawlers who don't respect won't, and the crawlers who do respect will.
How did you find this information? I know how to check traffic for my website, but idk how to get from "list of IPs" to "these ones are crawlers"
apologies if this is a silly question
I used to sit and monitor my server access logs. You can tell by the access patterns. Many of the well-behaved bots announce themselves in their user agents, so you can see when they're on. I could see them crawl the main body of my website, but not go to a subdomain, which is clearly linked from the homepage but is disallowed from my robots.txt.
On the other hand, spammy bots that are trying to attack you will often instead have access patterns that try to probe your website for common configurations for common CMSes like WordPress. They don't tend to crawl.
Google also provides a tool to test robots.txt, for example.
Perhaps this will help your understanding of my first point.
https://gizmodo.com/former-openai-board-member-sam-altman-chatgpt-1851506252
but if entities like OpenAI say they will respect it then presumably they really will.
Eh, will they really? It'd be pretty hard to prove they didn't respect it.
It's not about relying on it, it's about changing the behaviour of web crawlers that respect 'em, which, as someone who has adminned a couple scarily popular sites over the years, is a surprisingly high percentage of them.
If someone wants to get around it, they obviously can, but this is true of basically all protective measures ever. Doesn't make them pointless.
I too am a fan of privacy, but in the other hand… I kind of like the idea that we’re training AIs to convince people to become communists and/or change their gender
Such a measure merely punishes entities that respect the rules. If the content can be accessed, it will be scraped and used to train AI.
Lets be honest, AI should be incorporating hexbear. If anything we should have a hundred site mirrors with a free-for-all robots.txt.
After all, this site's only praxis is posting, why not use it to fill the AI with
No, robots.txt doesn’t solve this problem. Scrapers just ignore it. The idea behind robots.txt was to be nice to the poor google web crawlers and direct them away from useless stuff that it was a waste to index.
They could still be fastidious and follow every link, they’d just be ignoring the “nothing to see here” signs.
You beat scrapers with recursive loops of links that start from 4pt black on black divs whose page content isn’t easily told apart from useful human created content.
Traps and poison, not asking nicely.