Hacker News Reader

Comment by keiferski | Hacker News Reader

Projects like this remind me of a plot point in the Cyberpunk 2077 game universe. The "first internet" got too infected with dangerous AIs, so much so that a massive firewall needed to be built, and a "new" internet was built that specifically kept out the harmful AIs.

(Or something like that: it's been awhile since I played the game, and I don't remember the specific details of the story.)

It makes me wonder if a new human-only internet will need to be made at some point. It's mostly sci-fi speculation at this point, and you'd really need to hash out the details, but I am thinking of something like a meatspace-first network that continually verifies your humanity in order for you to retain access. That doesn't solve the copy-paste problem, or a thousand other ones, but I'm just thinking out loud here.

View on HN

The problem really is that it is impossible to verify that the content someone uploads came from their mind and not a computer program. And at some point probably all content is at least influenced by AI. The real issue is also not that I used chatgpt to look up a synonym or asked a question before writing an article, the problem is when I copy paste the content and claim I wrote it.

> The problem really is that it is impossible to verify that the content someone uploads came from their mind and not a computer program.

Er...digital id.

Ignoring the privacy and security issues for a moment, how would having a digital ID prove that the blog post I put on my site came only out of my own mind and I didn't use an LLM for it?

> the problem is when I copy paste the content and claim I wrote it

Why is this the problem and not the reverse - using AI without adding anything original into the soup? I could paraphrase an AI response in my own words and it will be no better. But even if I used AI, if it writes my ideas, then it would not be AI slop.

There doesn't need to be any difference in treatment between AI slop and human slop. The point isn't to keep AI out - it's to keep spam and slop out. It doesn't matter whether it's produced by a being made of carbon or silicon.

If someone can consistently produce high-quality content with AI assistance, so be it. Let them. Most don't, though.

I think the main issue is that when content is hand written you can be certain someone put at least the effort it takes to write into it. And while some people write fast, I would assume that at least means they have read their own writing once.

AIslop you can produce faster than you're able to read it. This makes it incredibly costly to filter out in comparison. It just messes so much with the signal to noise ratio on the web.

Bring AI-written is a proxy for being spam. Almost all AI-written content is spam and that's why it's bad.

> And at some point probably all content is at least influenced by AI.

[citation needed]

(I see absolutely no reason why that should be the case)

The issue is most things being derivative along with AI now representing an increasing share of "most things" from which to derive from.

The solution is not to be able to upload content. Extremely dumb services, basic trusted information sharing. Just like a newspaper.

I share an opinion with Nick Bostrom, once a civilization disrupting idea (like LLMs) is pulled out of the bag, there is no putting it back. People in isolation will recreate it simply because it's now possible. All we can do is adapt.

That being said, the idea of a new freer internet is reality.. Mastodon is a great example. I think private havens like discord/matrix/telegram are an important step on the way.

how does one keep ai out of private havens? thorough verification? is that the future? private havens on platforms?

In person web of trust in order to join any private community. It'll suck and be hard in the beginning, but once you reach a threshold, it'll be OK. Ban entire trees of users when you discover bots/puppets, to set an example.

So we expect either 1. people using AI and copy pasting into the human-only network, or 2. other people claiming your text sounds like AI and ostracizing you for no good reason. It won't be a happy place - I know from anti-generative AI forums.

Yep and then you deperson them

If you play it again, make sure not to miss the Blackwall gateway quickhack:

https://cyberpunk.fandom.com/wiki/Blackwall_Gateway

Absolutely brutal: https://www.youtube.com/watch?v=LD5z3GmQRXQ

---

I also noticed how simple the "new web" is when interacting with it. Of course, that's a game mechanic, but also kinda makes sense.

It's a shame that the Militech Canto only has 4 quickhack slots

Arguably this is already happening with much human-to-human interactions moving to private groups on Signal, WhatsApp, Telegram, etc.

There were also similar plot points mentioned in Peter Watts' Starfish trilogy, and Neal Stephenson's Anathem.

> a new human-only internet

Only if those humans don't take their leads from AI. If they read AI and write, not much benefit.

Reply on news.ycombinator.com