Anubis rarely ever annoys me (I still prefer it over cloudflare’s Mitm) but this specific visit feels almost vindictive. It used half my CPU cores at full blast for three straight minutes.
It seems to be effective since I can’t really “bypass” it with extensions, so AI crawlers probably can’t either, but damn there really should be an option to complete a manual captcha or email the administrators to complain.
As I’m typing this the challenge is still running xD
I wish AI scrapes didn’t make the Internet unusable to the point of them using Cloudflare MITM or Anubis
Also this instance; can’t browse it on Desktop with a Librewolf-alike policies.json. Can the sensibility be adjusted server-side? No problems on Archwiki & co.
I’ve had websites like Nix Hydra randomly give me a webpage full of gibberish which is also annoying and a lot moreso than the Anibus ones because you can’t get past it, I thought they got hacked at first also
Honestly better than running into cloudfail every 5mins
A word of advice: download Arch wiki for offline use.
You can use Kiwix to have a full graphical wiki, or use things like arch-wiki-docs.
Difficulty 6 is a little high. The default is 4.
I wonder if this even is effective. If a bot can solve anubis challenges, why can’t it solve all difficulties?
The point is to make scraping too expensive to be profitable. Even the easy level 4 anubis challenge multiplied by a few million pages becomes very expensive.
It’s not about stopping the bots, it’s about slowing them down
But once the challenge is completed, they have the token and can scrape until it’s expired right? So if it’s just taking them a bit more time to get the auth token, what is it solving?
Example: the token lasts for a week and it takes 30 seconds to get the token.
It will still slow them down and cobsume a ton of processing power- for data centers, it gets less profitable for them to scrap, it also completly stop some shitty bots
I would assume the token is only good for a few page access.
E.g. 500 would mean that a person would see this extremely rarely, and still block a bot from crashing the site.
No it actually lasts for a long time. We have it on my Lemmy instance for https://old.lemmy.today/ and it’s generated once and lasts for many days, probably a week. I guess it’s configurable but if you generate it too often, your visitors will also be annoyed.
Think of it like tipping culture. Get mad at AI industry scrapers that make anubis necessary. Sam Altman is the reason you are enduring this. Take a lighter and…
Instructions unclear: tipped my servers, now all offline?
I suspect this is more about all the individual searches than mass scraping. We’ve had google et al scraping the entire web for a long time and nobody gave a fuck. AI makes it easy to launch a ton of web searches without putting much thought into it.
Search engines doing it is good because now you are indexed and will show up in searched and they won’t do it as often as AIs.
AIs will take your stuff, DDOS you in the meantime and do it periodically - by different people too, they don’t give back anything either
Trying to battle AI bots by making using AI even faster than reading actual documentation. Great job.
I wish it were doing protein folding work or something less wasteful.
I wonder if that could be a thing
FoldCoin
put crypto nonsense to work and maybe help fund research
My concern like everything open-source is a single person deciding to profit off of it.
Banano is also one. But this one is more of a memecoin.
That exists already, actually: GridCoin
Would you trust the output of an A.I. swarm?
The idea is to pick some task that is very difficult to calculate, but easy to verify. E.g., confirming a crossword is correct is much faster than filling one in.
Protein folding is one such example, and it’s useful to boot. The results can be used in medical research, for example.
Donating compute via Folding@Home was quite popular for a while for this exact reason.
Folding@Home being built into the PS3 might have been one of the last tech-utopian hurrahs I can think of experiencing. As opposed to everything tech now feeling too dystopian for one of any number of reasons
The problem being the challenge needs to be unique and issued by the server on demand, otherwise a computer can have a bunch of pre-processed challenge answers ready to fire out a huge burst of requests. (Or re-use the solutions on other websites)
I don’t know how protein folding works, but if it’s easy to check it’s correct, then doesn’t that violate that mathematical principle where verifying an answer is as difficult as computing the answer in the first place? I forget the name of it.
I guess what I’m trying to say is, volunteers at home are already self-selecting to be mostly honest and donating correct answers for a good cause. Their output trustworthiness is moderate-high. Spammers, bots, A.I. swarms, ddos networks, random users, etc… not so much. You can’t really trust their output at all, you have to verify it very carefully and treat it as malicious or just incorrect junk by default.
You’re thinking, as someone else said, of P = NP. This is technically an unsolved problem. We don’t know if P = NP is true, there is no proof that guarantees it. However, it’s looking pretty unlikely.
This problem would have huge consequences for a wide range of fields, and it would break most cryptography and encryption algorithms we have (assuming it is proven true).
Anyway, this is what Anubis is already doing. That’s why OP complained about being stuck for 3 minutes with the CPU at full blast. Anubis gave the PC some problem that was difficult to compute. The problem itself was not important, it’s just to keep the computer busy for a bit. (The idea is that this significantly slows down spam bots and other automated access, I believe).
Of course Anubis has to verify the answer is correct, and that the client didn’t just send garbage. So the problem Anubis already uses is one that is easier to verify than to compute.
Btw, it is possible to verify solutions where checking is as expensive as solving. You provide each client with at least 2 problems. One where you already know the answer, and one where you don’t. You have the client solve both. If the known one is good, you let them through, and add the unknown one to the known set. You can even verify it by sending it to two other clients, see if they all get the same answer. But I believe that this solution wouldn’t be required, because protein folding ‘violates’ the P=NP problem.
that mathematical principle where verifying an answer is as difficult as computing the answer in the first place
You’re thinking of the P vs NP problem, but you got the details wrong. There are known cases where the answer is easy to verify.
verifying an answer is as difficult as computing the answer in the first place
What the distributed computing stuff does is check many many potential answers. You’re only interested in the right answers, and they’re easy to verify. If there’s a concern about junk answers, the answers could probably be compared across multiple clients to verify.
Unfortunately yeah, people with exceptionally shit computers get caught in the cross-fire of these things
The idea is that at individual scales the additional load is ignorable
So uh anyone anyone gonna tell them the idea was wrong? Three minutes is not ignorable.
The difficulty is typically dynamic based on current request load. At this difficulty typically means their server is getting hammered really hard, better this than the server just respond with a bad gateway error.
It would be better to serve a static page saying “the server is getting hammered really hard”, and, if possible, context about the duration of the current event and historical frequency.
Disagree, that just stops me from using it completely instead of just having to wait a bit
Waiting 3 minutes stops me from using a site. If it’s something I really need I’ll check back in an hour or two, when hopefully it’s not getting hammered; the same for both approaches.
Less sites are doing that to intentionally avoid tipping off the AIs about thresholds and trigger behaviours. The two main solutions are browser crypto challenges like this, or silently diverting flagged connections to a tar-pit of fake pages. The former is easier to implement but the latter actively sabotages the scraped data.
I’ve had Anubis get stuck for me before. Usually clears up if you refresh the page
Maybe that’s part of the challenge, lol.
Wait, I’m not mining crypto trash for strangers whenever I see that annoying fox girl, am I?
crypto* as in cryptography, not *currency
Proof-of-work in approach, not towards the same ends. The goal is to prevent scraping, not monetization.
Vain*
Vein*
Reworded it instead, but thanks for the correction.
Same process, different product.
Every time I watch Youtuße through Invidious and I get this crap, it gets stuck every time and goes nowhere.
This is every scraper for YouTube.
The only applications that I’ve ever seen reliably access YouTube are the ones that work by patching the YouTube app and then using what the app would do.
Applications like invidious or Grayjay almost always have a 15 to 30 second delay before any video.
My guess is that they, unlike the yt client, will wait for all the video to be streamed by the source instead of playing the first chunk
Public invidious instances are more a proof of concept, its much more performant to run it yourself.
True.
Will set one up, if more need arises.
…but people keep sharing invidious links.
The last part of the url is still the same Video id from youtube. You can copy that in whichever you use.
The internet actually really needs an overhaul to how links are presented. So you can press a video link and opening in the compatible player you prefer.
Same issue exists on lemmy. A link will send you to the instance of the poster, we have special ! Links to bridge the gap but i can never remember how to use em.
Technically, the internet already has what you’re looking for. It’s just not really standardized for video media.
The beginning of the URL or the schema is already able to on most mainstream OS be controlled by a specific application.
There’s nothing really stopping anyone from making it so there’s a video:// schema, and then configuring a schema handler to handle it. on Debian it would be a single desktop file with a MimeType=x-scheme-handler/video line indicating that type of url should be opened with that desktop file. I had to configure it for vortex to get it to open mod links, I never knew how easy that was
