Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Dead Internet Theory

kudmitry.com

Post reply on HN
Print view View original

301–310 of 735 posts

Re: Dead Internet Theory

#301

Post by danielbln »

meander_water wrote: post #294

Not foolproof, but a couple of easy ways to verify if images were AI generated: - OpenAI uses the C2PA standard [0] to add provenance metadata to images, which you can check [1] - Gemini uses SynthId [2] and adds a watermark to the image. The watermark can be removed, but SynthId cannot as it is part of the image. SynthId is used to watermark text as well, and code is open-source [3] [0] https://help.openai.com/en/ar…

Synth id can be removed, run it through an image 2 image model with a reasonably high denoising value or add artificial noise and use another model to denoise and voila. It's effort that probably most aren't doing, but it's certainly possible.

Re: Dead Internet Theory

#302

Post by sheept »

eru wrote: post #192
sheept wrote: post #173

Earlier quoted context omitted.

a reliable giveaway for AI generated videos is just a quick glance at the account's post history—the videos will look frequent, repetitive, and lack a consistent subject/background—and that's not something that'll go away when AI videos get better

> [...] and lack a consistent subject/background—and that's not something that'll go away when AI videos get better Why not? Surely you can ask your friendly neighbourhood AI to run a consistent channel for you?

AI is capable of consistent characters now, yes, but the platforms themselves provide little incentive to. TikTok/Instagram Reels are designed to serve recommendations, not a user-curated feed of people you follow, so consistency is not needed for virality

Re: Dead Internet Theory

#303

Post by krawcu »

bandrami wrote: post #292
viccis wrote: post #73

>which is not a social network, but I’m tired of arguing with people online about it I know this was a throwaway parenthetical, but I agree 100%. I don't know when the meaning of "social media" went from "internet based medium for socializing with people you know IRL" to a catchall for any online forum like reddit, but one result of this semantic shift is that it takes attention away from the fact that the former typ…

I'll come clean and say I've still never tried Discord and I feel like I must not be understanding the concept. It really looks like it's IRC but hosted by some commercial company and requiring their client to use and with extremely tenuous privacy guarantees. I figure I must be missing something because I can't understand why that's so popular when IRC is still there.

it's very easy to make a friend server that has all you basically need: sending messages, images/files and being able to talk with voice channels.

you can also invite a music bot or host your own that will join the voice channel with a song that you requested

Re: Dead Internet Theory

#304

Post by raincole »

pants2 wrote: post #106

Are there any social media sites where AI is effectively banned? I know it's not an easy problem but I haven't seen a site even try yet. There's a ton of things you can do to make it harder for bots, ie analyze image metadata, users' keyboard and mouse actions, etc.

The said hypothetical social media, if gaining any traction, will be the heaven for adversarial training.

Re: Dead Internet Theory

#305

Post by agoodusername63 »

eks391 wrote: post #283
lucumo wrote: post #269

Earlier quoted context omitted.

Not long ago, a statistical study found that AI almost always has an 'e' in its output. It is a firm indicator of AI slop. If you catch a post with an 'e', pay it no mind: it's probably AI. Uh-oh. Caught you. Bang to rights! That post is firmly AI. Bad. Nobody should mind your robot posts.

I'm incredibly impressed that you managed to make that whole message without a single usage of the most frequently used letter, except in your quotations.

Bet they asked an AI to make the bit work /s

Re: Dead Internet Theory

#306

Post by sznio »

8organicbits wrote: post #203
GMoromisato wrote: post #153

Most of this is caused by incentives: YouTube and others pay for clicks/views, so obviously you can maximize this by producing lots of mediocre content. LinkedIn is a place to sell, either a service/product to companies or yourself to a future employer. Again, the incentive is to produce more content for less effort. Even HN has the incentive of promoting people's startups. Is it possible to create a social network (…

I think incentives is the right way to think about it. Authentic interactions are not monetized. So where are people writing online without expecting payment? Blogs can have ads, but blogs with RSS feeds are a safer bet as it's hard to monetize an RSS feed. Blogs are a great place to find people who are writing just because they want to write. As I see more AI slop on social media, I spend more time in my feed reader…

I've been thinking recently about a search engine that filters away any sites that contain advertising. Just that would filter away most of the crap.

Kagi's small web lens seems to have a similar goal but doesn't really get there. It still includes results that have advertising, and omits stuff that isn't small but is ad free, like Wikipedia or HN.

Re: Dead Internet Theory

#307

Post by raincole »

ryanjshaw wrote: post #298
seiferteric wrote: post #116

My parents were tricked the other day by a fake youtube video of "racist cop" doing something bad and getting outraged by it. I watch part of the video and even though it felt off I couldn't immediately tell for sure if it was fake or not. Nevertheless I googled the names and details and found nothing but repostings of the video. Then I looked at the youtube channel info and there it said it uses AI for "some" of the…

I’m spending way too much time on the RealOrAI subreddits these days. I think it scares me because I get so many wrong, so I keep watching more, hoping to improve my detection skills. I may have to accept that this is just the new reality - never quite knowing the truth.

Those subreddits label content wrong all the time. Some of top commentors are trolling (I've seen one cooking video where the most voted comment is "AI, the sauce stops when it hits the plate"... as thick sauce should do.)

You're training yourself with a very unreliable source of truth.

Re: Dead Internet Theory

#308

Post by ycombinator_acc »

theteapot wrote: post #255
Refreeze5224 wrote: post #248

Earlier quoted context omitted.

[flagged]

I'm noticing more of these race baiting comments on YC too lately. AI?

No, that’s a common cope.

Not AI. Not bots. Not Indians or Pakistanis. Not Kremlin or Hasbara agents. All the above might comprise a small percentage of it, but the vast majority of the rage bait and rage bait support we’ve seen over the past year+ on the Internet (including here) is just westerners being (allowed and encouraged by each other to be) racist toward non-whites in various ways.

Re: Dead Internet Theory

#309

Post by bandrami »

krawcu wrote: post #303
bandrami wrote: post #292

Earlier quoted context omitted.

I'll come clean and say I've still never tried Discord and I feel like I must not be understanding the concept. It really looks like it's IRC but hosted by some commercial company and requiring their client to use and with extremely tenuous privacy guarantees. I figure I must be missing something because I can't understand why that's so popular when IRC is still there.

it's very easy to make a friend server that has all you basically need: sending messages, images/files and being able to talk with voice channels. you can also invite a music bot or host your own that will join the voice channel with a song that you requested

Right.... how is that different from IRC other than being controlled by a big company with no exit ability and (again) extremely tenuous privacy promises?

Re: Dead Internet Theory

#310

Post by adrian17 »

meander_water wrote: post #294

Not foolproof, but a couple of easy ways to verify if images were AI generated: - OpenAI uses the C2PA standard [0] to add provenance metadata to images, which you can check [1] - Gemini uses SynthId [2] and adds a watermark to the image. The watermark can be removed, but SynthId cannot as it is part of the image. SynthId is used to watermark text as well, and code is open-source [3] [0] https://help.openai.com/en/ar…

I just went to a random OpenAI blog post ("The new ChatGPT Images is here"), right-click saved one of the images (the one from "Text rendering" section), and pasted it to your [1] link - no metadata.

I know the metadata is probably easy to strip, maybe even accidentally, but their own promotional content not having it doesn't inspire confidence.

Post reply on HN