It's the End of the Web as We Know It
theatlantic.com
theatlantic.com
The article mentions recipe articles, and how these now invariably have a waffly preamble designed to please the SEO algorithms. It's been obvious to humans how stupid this is for years, just as it's obvious that sites that simply copy from stack overflow are leaches on our attention and should not be highly ranked.
Yet Google hasn't dealt with any of these problems, and with a focus on short term profits, has consistently preferred to take the money from ads hosted on these crappy sites instead of delivering a quality search experience.
Now the chickens are coming home to roost for Google, which can only benefit a range of competitors, from Kagi to OpenAI.
One valid question, though, is why technologists don't think about the consequences of what they do before they do it? Because they focus on short term profits, like you mentioned for Google?
Because they can’t. There’s no way Larry and Sergei could’ve foreseen the multi billion dollar SEO industry back when they were at Stanford. Every decision they’ve made since then has been of the form “stay on the gravy train or jump off?”
I don’t think anyone knows how to solve the problem of adversarial inputs in either the search space or the ML space. It’s a chicken and egg problem: you’re trying to build intelligence but the task of filtering out the adversarial inputs requires intelligence in the first place.
“Prediction is very difficult, especially about the future.” ― Niels Bohr
More pertinently, people have been arguing over the consequences of increasingly performant automation at least since Ada Lovelace was thinking about using something like the Difference Engine for music and Joseph Marie Jacquard was using punched cards for complex weaving.
The 200 years between them and us has given us radical, and for most people incomprehensible, growth. Back then, famine was common even with 90-95% of the population working in agriculture; now, obesity is the bigger problem with 1-5% of the population working in agriculture.
We knew about CO2 being a greenhouse gas a century ago, but we thought it would take a millennium and be a good thing; CFCs were chosen as a refrigerant because they were un-reactive; cars were a solution to all the horse manure on the streets; asbestos the solution to all the fires; radium paint was used on watch dials, nuclear explosives were seriously suggested as a way to widen (and also separately to replace) the Panama Canal.
It's just really hard to see all the consequences of things, even with the best will in the world — not that this completely contradicts what you say, as there's also the other famous quote:
“It is difficult to get a man to understand something, when his salary depends on his not understanding it.” ― Upton Sinclair
Let's presume I dislike my neighbour. He's not loud, just an asshat or some such, to the point that I'd love to punch him in the head. I don't, of course. We live in a modern society. Nor does he, either! The dislike flows mutual, deep and dark, yet we essentially ignore each other.
And even more astonishing? We go to sleep each night with a rival next door, sleep peacefully and in comfort, yet each of us with baser, animialistic instincts to beat on each other, perhaps to death, and yet? Even more astonishing, if the city walls were attacked? Hatred or not, we'd stand back to back and defend each other. Our families. Our friends, our city.
We are genetically predispositioned to live beside danger, and even completely ignore it, put it out of the way. The danger with the neighbour, a rival!! right next door, is ignored. And more so, the danger with those invaders one city over, with the wolf outside our cabin door, all of these things are put out of mind.
We'd go bananas if we had to focus on "the threat" constantly.
So we are designed to ignore even impending threats, let alone threads "decades or more in the future". If fact, I've often wondered if the more dire and dangerous the threat, yet with it not immediately on our doorstep, the more people might simply block it out!
That only mild threats are something that average human can deal with, for there is no need to push it from mind to sleep, to relax, to live without becoming a bundle of nerves. I wonder if there's some metric here, some "perfect level of threat", which ensures it is constantly in mind, yet sleep is possible, this returning the maximum human action on a distant (time or distance) threat.
I guess my frustration is that I can understand how fighting famine brought immediate, crucial benefits (even though it seems likely that the technological improvements that improved life may soon destroy civilization as we know it).
But the very first ideas that got in my mind where I learned about LLMs were black hat oriented. I actually needed to make an effort to see the potential benefits of LLMs (other than translators), but I immediately had many ideas of how that could go wrong (probably all of them being mentioned in the featured article).
In the case of LLMs, it seems a lot more direct than, say, the famous story of the Facebook "likes", that did sound like a good idea at first. LLMs sound like a bad idea (for society) from the beginning on.
I'm pleasantly surprised to learn this.
My experiences here since ChatGPT was launched have generally involved witnessing many of the louder people opining that LLMs "cannot" cause harm and that any example of harm anyone gives is "only" free speech. (The latter is something I find inherently weird as a talking point, given how important speech is).
I hope you're working in some kind of AI safety role; myself, I could only recognise the flaws when others showed them to me, I was not able to figure out these problems in advance.
I would argue that incentives matter, and the system runs like a chemical reaction to conclusion.
The idea that Edenic internet of the pre-AJAX days was scalable, sustainable or can be somehow regained is a fine fable.
> Because they focus on short term profits
It takes the deep pockets of an Elon Musk, or a hardcore community like HN to do otherwise, no? Again, incentives matter, and economic gravity is hard to defy.
They have bills to pay, and people are willing to pay for that. Technologists with strongly held opinions also exist, but they've been marginalized in every possible area where money gets you leverage. Just look at how open-source became a corporate breeding ground.
When you have a question like "Do N people have no ethics?", the answer is very often: some of N do, some of N don't. And the decisions to promote N1 and sack N2 are rarely made within the N group
Imagine if they had opinions held so strongly they'd be willing to camp out in the CEOs office in protest. I'm sure that the executives would re-examine their positions on the subject and work with the technologists to address their concerns.
So think about this: say your CEO asks you to make a missile specifically to target and kill children. Would you consider that it is not your job to question it, or would you actually think about it? I will assume that you would at least think about it. You would have some choices, then, for example:
- Accept it and keep doing your job
- Raise concerns internally (to your CEO) or externally (to the media), more or less vehemently
- Quit and go do something else
And I assume that for such an extreme case, all of this seems absolutely obvious to you. So let me ask: why does it feel like you can't imagine questioning something less extreme, like working on LLMs or autonomous cars or a connected fridge?
Engineers are paid to think (among others of course, but it is usually a necessary skill). It feels like they could reuse that skill and sometimes think about ethics, too.
People are regularly faced with such situations (with both small and large stakes) and all three of those choices are taken by different people. We know that 1. happens because controversial stuff still gets made. 2. happens because we hear about it in the media. And 3. I assume happens as well. But in the end the company can always find new people that choose 1. "Do it or I'll find someone who will." is both true and powerful. If you quit you lose your livelihood and potentially could be blacklisted from the industry, and yet the bad thing still happens anyway.
That's merely a rationalization to remain a member of group 1.
The truth is that 2 and 3 are not mutually exclusive, and together, they can and do occasionally prevent "bad thing" from happening. In some cases, it's because group 1 is small enough that the company cannot hire a toady to do "bad thing" in time to retain their contract for doing so. It may also be because the combination of external publicity coupled with a high-profile resignation is enough of a PR hit that the company reconsiders.
There's another element to consider: other, unrelated teams within the company who catch wind of "bad thing" can also opt for 2 and/or 3, further disincentivizing the company and perhaps even its competitors from proceeding, as losing a leading team in a highly-contested market segment could mean ceding significant ground to a competitor.
Granted, such moral stands don't affect change as often as one might like, but parroting "bad thing will always happen regardless what you try" everywhere certainly doesn't help things.
and it's why they usually have to have a decent amount of education -- which usually includes ethics -- and in many cases (like civil or mechanical engineers) have to have a formal certification from a governing body. programming doesn't have that, but probably should. regardless, it's a knowledge field, not a fork-and-spoon operator gig.
> "Do it or I'll find someone who will." is both true and powerful.
it's only powerful when there are high salaries attached. most people would swallow their objections for a steady 300k/yr
Well, I wouldn't and I would get fired, and the CEO would find someone with less scruples. And you would say "Damn, these technologists have no mercy" even if 99 engineers have refused before the one that accepted.
You're simply falling for survivorship bias.
What if I told you that I worked in fields that raise such ethical questions, and I have seen with my own eyes how there was never a single instance with 99 engineers who refuse for 1 who accepts?
For what I have seen, there was no survivorship bias. And then regarding LLMs in particular, I struggle to find people around me who actually question the ethical problems they bring. They just don't seem to even imagine that they could think about ethics.
So unless you don't ask them to do something that is extreme (design a weapon to kill children), they just don't question anything at all.
But hey, I guess you could say that's humanity's flaw in general. If all our current social and environmental issues are taken into account, it's pretty clear that most people don't think about the consequences of their actions beyond the next week or so.
Software "engineers" are not part of a professional body at all, let alone one that espouses ethics. That only seems to happen in professions with a high risk of lawsuits due to customer death.
How deeply would you expect an interior designer or a landscaper to think about future consequences?
The article calls that LLMO :-)
Even if Bing is better, that might be simply due to the small player's advantage: Google's market share is so large, they have essentially the whole economy of the web investing in SEO against their search quality, adapting quickly whenever Google tunes the algorithm. Meanwhile Bing can tune their algorithm and nobody cares.
Thank you, kind stranger.
> There’s nothing inherent to the X that guarantees you free access to its content.
Libraries, public transport? Taxes. Love of your parents? Depends on them.
Give me something for which I can say use the phrase and it will be true.
I could do the same with you: this sentence does not make sense:
> Taxes.
I used to pretend to be Googlebot a lot to bypass paywalls but I think these days there is either IP range filtering or more advanced fingerprinting.
And businesses tend to seek more profits, in part by cutting costs. So what's the obvious way to cut the cost of an algorithm that involves a bunch of slow, unreliable meat?
This phrase seems too muddled to be useful to me. An algorithm that aggregates preferences of people isn't "just" a "a finite sequence of mathematically rigorous instructions".
If we run amok with the usage above (which I don't care for), what _isn't_ an algorithm? When everything is an algorithm, the word has no meaning.
--
I agree with the idea. But I can't come up with any feasible model of motivations on how to implement it.
This all existed not so long ago.
See: Wirecutter, a billion other 'review' sites, blogspam, etc.
We can also name it altavista
If we want to look for solutions to apply to the internet, we need to look at what has developed in the real-world domain, where different interests have been trying to alternately solve and exploit trust-issues for millennia. It might not have final definitive answers, but we can at least avoid reinventing too many wheels.
I don't understand what you're trying to say here. Sure, people are using more computers everywhere, but that doesn't necessarily change fundamental guidelines we already used, like "each time someone could betray you and doesn't, that suggests they are trustworthy in the future."
That's the hard part, isn't it? The emergence of LLMs makes it increasingly difficult to automatically distinguish AI-generated content from human made content. Heck, even humans may find this difficult!
This kind of spam is an annoyance, but you learn to filter it out soon enough.
Stack Overflow is a great example : its value is not so much in giving you a quick answer, but even more because you will see several answers, and the discussion around them. And yes, sure, convincing malicious actors there are always possible, but not probable.
Same thing with Wikipedia's Talk pages, blog's comments (with the censorship issue), or third party comment sites like HN (with the platform issues).
I'm much more concerned about artists that make works for their aesthetic value losing their already precarious livelihood to AIs. And even considering that it's sometimes feels like it's worth it : https://youtube.com/@D4BadRadio (who would have paid enough for this to be made by artists ? nobody)
The problem is that LLMs are a whole new level of spam, orders of magnitude more powerful than anything we know. Before LLMs, you could not generate millions of comments for e.g. TripAdvisor without the users realizing that those were generated. With LLMs, users cannot say anymore if a review was written by a human or not. You just cannot trust any anonymous review you find online anymore.
If Google cannot automatically discriminate between spam and genuine content, then Google is broken. And one fundamental promise of LLMs is that it is (or will be) powerful enough to reach that point, and therefore to break Google.
Also, if even pseudonymous accounts can be (meaningfully) penetrated by bots, then what prevents article's deep web from being penetrated and destroyed too ? (In fact we've already seen it happen on the likes of Facebook years ago, its faster, pseudo-ephemeral nature making things worse I guess ?)
> the web … will cease to exist in any useful form
What has tended to happen in the past is, drum roll please, competition appears making the interwebs more useful again.
Example: DuckDuckGo’s popularity came out of the desire for a more privacy friendly search engine.
The only constant in life is change. Just because a thing becomes less useful short term, does not mean it will always be less useful long term.
On a more meta note —
Title —> Subtitle —> Advert —> Four sentences —> Advert -> Four sentences —> Paywall
I find it mildly ironic that it will be the push for GenAI’s fault, when the publisher is doing everything they can to stop me from reading an article about how it’s going to be the push for GenAI’s fault.
I am for fair pay for creators, and I’m definitely not a big fan of GenAI. I just found it to be an interesting irony, given the setup of the article wrt the origins of the interwebs.
DDG has a market share of 0.66% [1], that doesn't feel like popularity. That feels like an 'etc.'
That also proves most people lack taste and knowledge.
And also socially through authenticity and originality pledges from content producers and their platforms identifying what was created entirely by humans and what predictive text completion tools were used, if any.
I gleaned these words before TFA fades and prompts me for a subscription; how about we stop linking to paywalls and excessive data collectors, or at least flag posts as such?
Actually, writers haven't been able to reach audiences for a long time now, much less make a living (see also the "nobody reads books" story featuring on HN right now) as illegal ad providers/search engine monopolies and others have established themselves as new men in the middle, something the web was designed to break to begin with. The other route to monopolies is lack of payment options/microtransactions, the subversion of regulated stacks with fair access (such as 3G networks used to be) by de facto "platforms", and the utter and complete failure of politicians to uphold currency sovereignty or (pretend to) even grasp the problem.
Perhaps the web dying as it is can benefit actual authors and other content producers? At least we don't have to endure BS freeriding articles like these pretending to "save the web".
But because you could not read the article does not mean that it is "BS freeriding articles like these pretending to "save the web"".
Did you check who the authors are (you can see that for free). Bruce Schneier is a very respected cryptographer. And usually after a few days he publishes the article for free on his blog. So, there's that.
A quote that's unironically behind the paywall.
I'm expecting that the essay will be published on his blog in a few days, with free access.
is such a terrible and deeply dishonest trope in my view, I really dislike it.
(I know I will probably get downvoted for this, but as an old R.E.M. fan, I just couldn't help it)
After that, the powers that are in charge have engaged in social manipulation to try and steer the internet to provide only the ”approved” concepts.
Gemini’s results on blacks proliferating in the confederate army is an example of how far it has gone.
Wikipedia is not trustworthy on articles on notable people (watch when an election kicks off and how manipulated they are) and Google tries to openly manipulate search results because of the SEO decisions made years ago.
Our only hope is Web3 and an ability to host stuff on there.