Hacker News Reader

Comment by akersten | Hacker News Reader

They do this by adding a ton of useless markup and splitting words like "ad" into single-letter spans with random class names and 8-layer deep nests of `<div>` so it's very hard to write selectors against. One has to wonder how that div soup affects accessibility as I find it very unlikely that this content is presented nicely to assistive tools. I hope they get all the ADA suits they deserve for this.

View on HN

I recently found this out the hard way. Instagram’s web app has a super annoying popup. To click the comments button for a post, your cursor has to pass over the username, which just happens to launch a profile preview with the “follow” button exactly where the comments button was, causing you to unintentionally follow the account.

I tried to block this popup with uBlock Origin to no avail. No matter what element I selected, it was still there. Finally fixed the problem by deleting my Instagram account.

Best move you could have possibly done.

“A strange game. The only winning move is not to play.”

A counter example: I got caught out by the deprecation/sunsetting of anonymous "old.reddit.com" yesterday but it turns out uBlock can filter out the ads, boilerplate, sidebars, join/login buttons etc on the main site. It's actually faster and more usable than the old site now.

Small win.

>I got caught out by the deprecation/sunsetting of anonymous "old.reddit.com" yesterday

It appears to still be up for me. ??

Not sure there's been a hard date publicly set but they are sunsetting it

It's A/B tested across IP ranges, geolocated regions, etc. I got it super early and it sucks being locked out for the past 2-3 weeks now.

You now have to be logged in to reddit to use it.

"anonymous" is the key word. old.reddit still works (though it seems to get progressively jankier), but it's now requiring a login.

It was activated for my region yesterday and it did wonders for breaking my reddit addiction.

Then I tried the normal view and ... wow, so many auto-playing video ads disguised as regular posts.

Reading the claim that they're requiring login to old.reddit.com "for safety" made me laugh out loud.

When are they going to admit that the redesign sucks and there's a reason everyone keeps loading the old reddit interface? Who has the ego that can't handle the truth, that their UI/UX design is garbage and loathed?

The random phrases turned into links that are never helpful means I can't trust any link in comments now, there's that stupid AI box that never has useful or helpful info, conversations are a nightmare to try and read through.

I will never understand how it is that a site which exists almost entirely for conversations, made a new UI that severely fucks with "read the conversation."

It's using an algorithm to determine whether or not to allow anonymous access. It looks like it's considering origin IP, browser useragent, among other things. I've had login prompts show up once or twice, but when I'm using my regular browser in incognito mode and not connecting through a third-party VPN, it continues to allow logged-out access.

How do you filter it out? Do you select each element individually?

Yes, the Element Picker in uBlock Origin. Requires Firefox because of Web Manifest stuff in Chromium.

nah, now it loads maybe 3 posts at a time and the latency is really slow to pull up those 3

Reddit tested login walls on the new site since months.[1]

[1] https://github.com/uBlockOrigin/uAssets/issues/32706

> Finally fixed the problem by deleting my Instagram account.

The real fix!

Never delete your account. That account has a bunch of associated signals that scream "real user". Therefore you should sell it on an account marketplace to a spammer, if you really hate the platform.

> Finally fixed the problem by deleting my Instagram account.

The best solution. Highly recommended.

Last I tried writing custom rules for Facebook was years ago, but they did have proper aria attributes, and uBlock origin does have selectors for text content. These are from 2020:

  facebook.com##span:has-text(Suggested for You):xpath(../../../../../../../../../../../../../../../../../..)
  facebook.com##div[aria-label="Sponsored"] span[aria-label="Sponsored"]:xpath(../../../../../../../../../../../../../../../../../../../../..)
  facebook.com##div[aria-label="Sponsored"]:xpath(../../../../../../../../../../../../../../../../../../../..)

What's with the xpath selector? Is that just matching deeply nested elements?

Going from memory, but that's how many elements are in the tree between what I wanted to be hidden and where the matching text/elements actually lived.

Oh jeez, I hadn’t thought about the blocked element being different than the aria-tagged one. What a nightmare

Read up on XPath axes:)

you can use upward. For example

  facebook.com##span:has-text(Suggested for You):upward(8)

That then goes up elements in the dom.

But can't they just use an xpath expression to test the computed text content of a node, like `/path/to/ad/div[contains(., 'sponsored')]`? In that expression, there could be any number of nested elements inside the terminal `<div>` and it wouldn't matter. (And you'd probably have to use a regex test to account for tricky white space.)

The word sponsored in inside an SVG element with a linked `use` tag, so it's not that simple.

> so it's very hard to write selectors against

This sounds perfectly matched for an on-device LLM?

The power cost of doing this widely would be staggering, surely?

Apple Intelligence? I mean, you're only processing for the time your "AI|browser|agent" is acting as a firewall between you and Meta. Cache locally after processing and filtering. Use alongside the accessibility API. LLMs can, in many cases, reliably solve CAPTCHAs. I find it difficult to imagine they cannot defeat Meta ad blocking countermeasures.

EFF: Adversarial Interoperability - https://www.eff.org/deeplinks/2019/10/adversarial-interopera...

Depending on the model, I’m guessing it’s dwarfed by just about any electron app

Worth a white paper to see which costs less energy, using Apple's built in LLM, or downloading and displaying all the FB ads using radio, playback, and screen animation energy.

Local models are not the same as the giant ones in data centers. They're in about the about the same ballpark as running a AAA game on max settings.

Sure, it's more computationally-expensive than running an HTML selection, but it's also not "staggering" by any reasonable stretch.

I think “staggering” is a pretty accurate way of putting it if blocking ads is in the same ballpark as running a AAA game on max settings.

If the computational requirement is within the realm of another normal day to day task, and if it cannot generally achieved with another cheaper approach, it would be contradictory to say it is staggering.

I would agree if this was an easily solvable problem with classical approaches. Clearly it is not, since even uBlock is giving up.

At any rate, "staggering" is relative. White-labeled web browsers becoming the most common GUI engine would have made an engineer's eyes water 20 years ago.

Blocking ads should not make my fans spin at full blast.

Yeah, it's pretty awful that Facebook has made it such a demanding task.

"Honey, I want to buy a gamer PC. This one is too slow to block Facebook ads."

The giant ones in data centers are models that broadly encompass a variety of data for general purpose applications. You can certainly train models specifically for processing DOM content, making them smaller and more efficient.

Even the data-center-grade models are less intensive than people think. We recently installed a heavy-duty AI server (Gaudi2 cluster with 768GB VRAM) to run larger models up to 400B parameters on, and at full load, its power draw is about equivalent to a consumer stove or clothes dryer.

If you're running thousands of those concentrated into a single data center, your aggregate power draw is going to be huge, but a single server is entirely reasonable for a small business, or even a home user, to operate.

Not really, no. LLMs get more computationally efficient and hardware gets more power efficient with each passing day. We're already at the point where reasonably competent models can be run a laptop GPU off of battery power. NPUs are even more power efficient, (though a bit slower).

It keeps changing all of the time, so it might be easier to have Facebook load up in a headless browser, that takes a screenshot and sends it to an LLM and uses that to convert into a standard JSON document that can be rendered nicely. Or something like that.

Maybe reverse tactics and block/hide anything that isn't something you want to see.

In a few years, I don't think we'll be navigating to Facebook.com. We'll be asking our LLM: "what are my friends up to?" And it will provide a nice UI to browse interesting updates, collated from any open source, including Facebook. No ads, no influencers, no dark patterns, just what the user asks. And it won't require APIs. It will just browse the site itself as an agent and pick out the parts the user cares about.

I really think the new internet is going to be a radically different experience soon.

At least there is no chance of manipulation and subliminal advertising that way. /s

The risk vector shifts from Facebook to the LLM provider. This is why I'm a big fan of the open weights models. They're not immune to attack, but it's much easier to identify censorship, propaganda, and advertising. They're getting really good, really fast.

> No ads, no influencers, no dark patterns, just what the user asks.

For a while now, I've seen blogposts from marketers desperate to do to AI what SEO did to search engines.

I hope they fail to enshitify the LLM experience the way they enshitified search, but I wouldn't count on it in the absence of AI messing with production so hard that nobody even wants to market stuff in the first place.

> very hard to write selectors against.

Or was hard to write selectors against before, you know, AI.

It seems ironic to give up now. You can probably rig up an agent to automatically look for new div labyrinths and write selectors for them.

You can bet Facebook are using AI to generate these labyrinths!!!

Could you not just go after the deeply nested spans? I assume they aren't messing up everything else like that.

Based on inspect element on their homepage, they probably are.

Who would file an ADA suit for a screen reader being unable to correctly read an ad?

Those who want to do it in spite.

The same ambulance-chasing lawyers who are already filing spurious ADA suits today.

> I hope they get all the ADA suits they deserve for this.

So what you're saying is: you want Facebook to pay out something like $200M which is such a small amount it won't even realize, $170M will go to lawyers, everybody else gets $1.30.

That seems like how it might pan out.

Or an unauthorised access / exceeds access to a computer system prosecution in at least one jurisdiction somewhere around the world.

If the owner of the system has a 'no ads' policy and a security mechanism to prevent ad servers from accessing their display without their authorisation, it is one thing if a company accidentally circumvents that by changing their page in a way that breaks detection (and unless there are records it was deliberate) they probably get the benefit of doubt from authorities.

But if they are blatantly going to quite great effort to circumvent the adblock security system to exceed their unauthorised access without the consent of and against the wishes of the owner of the client system, surely that's a clear-cut criminal case?

Of course, prosecutions of these types of things in many countries unfortunately tend to go after individuals who target big companies, but almost never the other way around.

We need to figure out how this is actually DMCA circumvention. That one seems to have teeth.

Reply on news.ycombinator.com