Hacker News Reader

Comment by joken0x | Hacker News Reader

I think there will come a point when the traditional internet will also become diluted in artificiality, content from the procedural web will start to creep into the traditional web, and we won't be able to distinguish. It's interesting to think of it in terms of Baudrillard's Simulacra and Simulation.

View on HN

We're passed that point. Probably by a couple of years at a minimum. No sarcasm. Content farms are definitely using techs like this. GPT-3 is really good at generating text but still has some characteristic failures, and I encounter content farm web pages (despite my best efforts) that have clearly used it or something like it as a tool. Even just in the last couple of weeks I've been seeing some new, innovative content farms managing to pollute my search results that I've not seen before.

Oh boy, we need AI powered blockers that filter out this stuff.

This seems like a terribly difficult problem. I'm sure today there are some give-away signatures, but I put my money on the impersonators in the long run.

One of my current heuristics is the expression of weirdness. I can see it in the way I write, there's like this thread of a human processing information through emotional and idiosyncratic experiences. There's a certain cadence to the way thoughts come out. Maybe it can be emulated eventually, but that might be farther off.

In my fantastical imagination of that implementation, I would simulate "sim agents" as they live a life, collecting unique experiences that are the result of interactions with other sims. They then build up personalities, which influence their expressions in other mediums that real humans would consume. It's psychotic and possibly evil and I thought of it first, so no one better take my idea.

You might be able to see what I mean just in this comment itself. Could this have been written by an AI?

Oh, you mean sentient GANs? https://en.wikipedia.org/wiki/Generative_adversarial_network

Yeah, that seems close. The main difference would be that it would be a co-evolutionary sandbox instead of zero sum.

AI fighting against AI, everything will be AI.

This is how it ends the robots will figure out the real problem is people, better prepare for our new overlords maybe we can get into a human zoo..

Anathem by Neal Stephenson has a great subplot about this. The information age has been a bit stunted because there's too much crap[1] on the internet. Companies sprung up selling filters that would block websites with low-quality or AI generated information. Eventually these companies realized they could drum up business by generating low quality[2] content themselves, especially if they could get it past their competitor's filters. The end result was that the internet became a convoluted morass of bullshit and lies, difficult for non-experts to extract useful knowledge

[1]: Or maybe CRAAP https://en.wikipedia.org/wiki/CRAAP_test

[2]: They quickly realized the trick is make high quality low quality content. 100 pages of gibberish is way less effective than a convincing essay that happens to include a few key falsehoods.

It seems to be a loop of information from which there is no way out, that devours the information and stirs it among piles of garbage that it generates without stopping. Labyrinth and crypt at the same time, growing and churning more all the time. I think we have to start a serious effort to collect and save the human creation, otherwise it will become a hidden treasure buried among layers of artificial garbage. Imagine, if it can be difficult for us to distinguish the human content from the synthetic, how will it be for future generations who will not even have the living context of our time? And who will already be more accustomed to these synthetic contents than to the properly human ones.

How will this change how we evaluate content? Will digital media be socially and culturally devalued? Will print media gain greater status? Live spoken word? Curated content with a reputation layer? Obviously it's pure speculation, but let's indulge for a moment.

His latest book, “Fall, or Dodge In Hell” also uses the same idea as a major plot device. And alike all things written by Neal Stephenson, is amazing.

> I think there will come a point when the traditional internet will also become diluted in artificiality

There is already a concern in corpora creation in ML/AI projects. Researchers would like very much to have human-only generated content when training models on internet-sourced text. People posting GPT-created output has the potential to taint these corpora and create all sorts of strange loopy feedback.

Knowing people it wouldn’t take long for incoherent generative syntax to become a meme and reinforce the human corpora with new syntactic slang.

That's right, the human creation is and will become more and more a very precious treasure.

We are still missing a decentralized identity and reputation layer. I imagine in the future, people will sign their comments with a key that proves they’re a real human.

Robots will of course have keys they can sign with, so perhaps we will rely on government digital ids such as he Estonian public key system. In that case, governments will have a monopoly on sockpuppet accounts.

Has there been any successors to web of trust ?

I would simply ask GPT-n what to write next, copy-paste it into the text box, sign it with my key, and hit reply. Its ability to write better comments than me would serve to gain me reputation rather than reduce it as so often is the present case.

AI might mine cryptocurrencies on their own, and then buy keys from desperate humans, and post as if they were the owner of the keys.

Man, how crazy. Continuing like this, they could create their own identification keys for machines and despise us. It seems that in this whole game only they end up winning.

This. It's the elephant in the room for so many of our problems related to abuse, trolling, influence campaigns, garbage content, attribution etc etc.

OTOH, is proof of personhood really what we want long term, or is that just a proxy? Hypothetically, if an AI is good & trustworthy enough, why not allow a higher "source rating" for that than low quality human content?

Relevant XKCD: https://xkcd.com/810/

Hehe hadn't seen that. For sure. To critize the strip (treading on thin ice, I know), I think the core issue is not the bots generating content, but that they're participating in the voting.

This is exactly what a bot would say.

Hahaha we could all be bots here. Bip bip.

Reply on news.ycombinator.com