Dead Internet Theory

What secret is hidden in the phrase “you are absolutely right”? Using Google's web browser translation yields the mixed Hindi and Korean sentence: “당신 말이 बिल्कुल 맞아요.”

You know, one thing we could do is to get the costs for energy usage sorted out. Like, people who use a lot data-center electricity, pay accordingly.

If AI would cost you what it actually costs, then you would use it more carefully and for better purposes.

> What if people DO USE em-dashes in real life?

I do and so do a number of others, and I like Oxford commas too.

bots are everywhere and Ai bots making this theory very true.

Given the climate, I've been thinking about this issue a lot. I'd say that broadly there are two groups of inauthentic actors online:

1. People who live in poorer countries who simply know how to rage bait and are trying to earn an income. In many such countries $200 in ad revenue from Twitter, for example, is significant; and

2. Organized bot farms who are pushing a given message or scam. These too tend to be operated out of poorer countries because it's cheaper.

Last month, Twitter kind of exposed this accidentally with an interesting feature where it showed account location with no warning whatsoever. Interestingly, showing the country in the profile got disabled from government accounts after it raised some serious questions [1].

So I started thinking about the technical feasibility of showing location (country or state for large countries) on all public social media ccounts. The obvious defense is to use a VPN in the country you want to appear to be from but I think that's a solvable problem.

Another thing I read was about NVidia's efforts to combat "smuggling" of GPUs to China with location verification [2]. The idea is fairly simple. You send a challenge and measure the latency. VPNs can't hide latency.

So every now and again the Twitter or IG or Tiktok server would answer an API request with a challenge, which couldn't be antiticpated and would also be secure, being part of the HTTPS traffic. The client would respond to the challenge and if the latency was 100-150ms consistently despite showing a location of Virginia then you can deem them inauthentic and basically just downrank all their content.

There's more to it of course. A lot is in the details. Like you'd have to handle verified accounts and people traveling and high-latency networks (eg Starlink).

You might say "well the phone farms will move to the US". That might be true but it makes it more expensive and easier to police.

It feels like a solvable problem.

[1]: https://www.nbcnews.com/news/us-news/x-new-location-transpar...

[2]: https://aihola.com/article/nvidia-gpu-location-verification-...

I’m a bit scared of this theory, i think it will be true, ai will eat the internet, then they’ll paywall it.

Innovation outside of rich coorps will end. No one will visit forums, innovation will die in a vacuum, only the richest will have access to what the internet was, raw innovation will be mined through EULAs, people striving to make things will just have ideas stolen as a matter of course.

What is the next safe haven for smart people?

It used to be Internet back when the name was still written in the capital first letter. The barrier to utilize the Internet was high enough that mostly only the genuinely curious and thoughtful people a) got past it and b) did have the persistence to find interesting stuff to read and write about on it.

I remember when TV and magazines were full of slop of the day at the time. Human-generated, empty, meaningless, "entertainment" slop. The internet was a thousand times more interesting. I thought why would anyone watch a crappy movie or show on TV or cable, created by mediocre people for mere commercial purposes, when you could connect to a lone soul on the other side of the globe and have intelligent conversations with this person, or people, or read pages/articles/news they had published and participate in this digital society. It was ethereal and wonderful, something unlike anything else before.

Then the masses got online. Gradually, the interesting stuff got washed in the cracks of commercial internet, still existing but mostly just being overshadowed by everything else. Commercial agenda, advertisements, entertainment, company PR campaigns disguised as articles: all the slop you could get without even touching AI. With subcultures moving from Usenet to web forums, or from writing web articles to posting on Facebook, the barrier got lowered until there was no barrier and all the good stuff got mixed with the demands and supplies of everything average. Earlier, there always were a handful of people in the digital avenues of communication who didn't belong but they could be managed; nowadays the digital avenues of communication are open for everyone and consequently you get every kind of people in, without any barriers.

And where there are masses there are huge incentives to profit from them. This is why internet is no longer an infrastructure for the information superhighway but for distributing entertainment and profiting from it. First, transferring data got automated and was dirt cheap, now creating content is being automated and becomes dirt cheap. The new slop oozes out of AI. The common denominator of internet is so low the smart people get lost in all the easily accessed action. Further, smart people themselves are now succumbing in it because to shield yourself from all the crap that is the commercial slop internet you basically have to revert to being a semi-offline hermit, and that goes against all the curiosity and stimuli deeply associated with smart people.

What could be the next differentiator? It used to be knowledge and skill: you had to be a smart person to know enough and learn enough to get access. But now all that gets automated so fast that it proves to be no barrier.

Attention span might be a good metric to filter people into a new service, realm, or society eventhough, admittedly, it is shortening for everyone but smart people would still win.

Earlier solutions such as Usenet and IRC haven't died but they're only used by the old-timers. It's a shame because then the gathering would miss all the smart people grown in the current social media culture: world changes and what worked in the 90's is no longer relevant except for people who were there in the 90's.

Reverting to in-real-life societies could work but doesn't scale world-wide and the world is global now. Maybe some kind of "nerdbook": an open, p2p, non-commercial, not centrally controlled, feedless facebook clone could implement a digital club of smart people.

The best part of setting up a service for smart people is that it does not need to prioritize scaling.

Such posts are identifiable and rare, disproving Dead Internet Theory (for now).

Are em dashes in language models particularly close to a start token or something? Somehow letting the model continue to keep outputting.

The problem is not the Internet but the author and those like them, acting like social network participants in following the herd - embracing despair and hopelessness, and victimhood - they don't realize they're the problem, not the victims. Another problem is their ignorance and their post-truth attitude, not caring whether their words are actually accurate:

> What if people DO USE em-dashes in real life?

They do and have, for a long time. I know someone who for many years (much longer than LLMs have been available) has complained about their overuse.

> hence, you often see -- in HackerNews comments, where the author is probably used to Markdown renderer

Using two dashes for an em-dash goes back to typewriter keyboards, which had only what we now call printable ASCII and where it was much harder add to add non-ASCII characters than it is on your computer - no special key combos. (Which also means that em-dashes existed in the typewriter era.)

>The other day I was browsing my one-and-only social network — which is not a social network, but I’m tired of arguing with people online about it — HackerNews

dude, hate to break it to you but the fact that it's your "one and only" makes it more convincing it's your social network. if you used facebook, instagram, and tiktok for socializing, but HN for information, you would have another leg to stand on.

yes, HN is "the land of misfit toys", but if you come here regularly and participate in discussions with other other people on a variety of topics and you care about the interactions, that's socializing. The only reason you think it's not is that you find actual social interaction awkward, so you assume that if you like this it must not be social.

The irony is that I submitted one of my open source projects because it was vibe-coded and people accused me of not vibe coding it!

What is now certain is Dead StackOverflow Theory.

But what about the children improving their productivity 10x? What about their workflows?

Think of the children!!!

lol Hacker News is ground zero for outrage porn. When that guy made that obviously pretend story about delivery companies adding a desperation score the guys here lapped it up.

Just absolutely loved it. Everyone was wondering how deepfakes are going to fool people but on HN you just have to lie somewhere on the Internet and the great minds of this site will believe it.

[dead]

[dead]

[dead]

[dead]

[flagged]