Hacker News Reader

Comment by _heimdall | Hacker News Reader

The product here isn't necessarily suprising, it's only a matter of time before someone builds this. I continue to be suprised how deep a social media company wants to go into LLM-generated content though.

What is the real value of social media when most of the content posted is created by a bot? Taking humans out of the loop seems to remove all value from showing ads, and the value of content that can be used to train LLMs tanks when the content is itself LLM-generated.

View on HN

> What is the real value of social media when most of the content posted is created by a bot?

Besides the maybe 5% or so of people I interact with online that I actually know IRL, I don't really care if I'm interacting with a bot.

A bot could have secretly written your message and I'd still leave the same reply. Anyone that replies to me could also secretly be a bot and I'd read their comment the same. This whole HN post could just be bots all the way down and it'd still be an interesting read through the comments. Even if the replies are from humans, if they're indistinguishable from other humans I still don't really care about who wrote them.

Engagement is engagement and my Sims-esque social bar refills at the same rate.

Ok, I can see how content can be interesting even if there's not a person behind it

but would you still publish your comment if it was only ever going to be read by computer programs?

Depends on if the computer programs might respond in an interesting way or if my comment might help them respond in an interesting way in the future. Same as when interacting with humans, I guess.

I think there's two likely scenarios here:

1) My comment is only ever going to be read by computer programs and I'll probably get a response back from a computer program (see: ChatGPT/GPTs, character.ai, AI Studio, Replika, the countless "AI friend" services, etc)

2) My comment is posted in a more public setting that's inundated by computer programs, but also visible to other humans (e.g. a comment on a public FB post that computer programs read, but that other humans might also read and/or respond to)

#1 feels like current-gen one-on-one chatbots, but #2 is where I expect FB to end up eventually, using bots to keep conversations flowing and growing while humans interact here and there.

In either case, my answer is yes. It's still a conversation either way, isn't it?

Most people don't post, they only consume the content

maybe you don't care about that, but there are reasons why humans talk to other humans, even strangers:

* to convince someone about something you are passionate about

* to share enthusiasm (or disgust) about a topic

* to learn something new

.... other human reasons and sharing emotions

Bots have no passion or enthusiasm or emotions, so they are useless for the first two. Maybe for the last one, but they are sterilized, don't like controversy or arguments, lie, and will just agree if you press on them. I still suspect you wouldn't like to be surrounded by bots.

I do all three of these with bots (and also humans).

Do you feel you get the same out of it whether talking to humans or bots?

Its interesting that you'd ultimately be engaging in discussions only for what you get out of it. The more important or impactful you find the topic to be, the more useful it seems like it would be to talk with someone you may learn from or teach something to rather than having any object to talk at.

i can do all three with a frozen burrito at a gas station doesn't really mean it's a great alternative

I couldn't meaningfully do any of the three with a frozen burrito, unlike with bots and/or humans. More power to you if you can, though.

I would not take that as a given. Yes the official releases have been tuned to behave that way but even back in the dark ages of AI it has already been demonstrated that chat bots can produce the whole range of expression that humans can. It doesn’t take much to tilt LLMs in other directions including eliciting a variety of emotions. All of that data is in their training set, you just have to bias them in that direction.

This is one reason why I find the focus on making LLMs more efficient so interesting, it’s going to result in highly capable models that can run on cheap consumer hardware or cheap rented GPUs which will lead to a veritable cambrian explosion in bot personalities. Bots like truth_terminal are just the beginning

https://x.com/truth_terminal?lang=en

>This whole HN post could just be bots all the way down and it'd still be an interesting read through the comments

This whole HN post could have been generated by monkeys on a typewriter all the way down and it'd still be an interesting read is also true, but that doesn't mean seeking out monkey generated content is a great strategy for finding interesting reads.

Touché! Likewise, if grandmother had wheels, she'd be a bicycle.

> This whole HN post could just be bots all the way down and it'd still be an interesting read through the comments.

The assumption is that the comments are a function of the post and the present public info. Real world comments can disclose private information (My Google account banned), make real impact (S*e/C*e support site), connect to celebrities (I'm Karpathy ask me anything). And even if the assumption holds and the site is a sample from a probability distribution, the particular sample can be referenced in other sites so it makes sense to check what everyone is viewing right now.

I struggle to imagine that your perspective is representative. Most human beings, I surmise, value human interaction more than we value object interaction. Even the output of these new language models—while flexible, reliable, and available—pales noticeably in comparison to what a living creativity can produce.

You might be able, sometime in the future, to fool me. That was the hideously disgusting premise of The Matrix, wasn't it? Ah, no; even in the Matrix, we had one another.

> Most human beings, I surmise, value human interaction more than we value object interaction.

I think the key here is that you never see a human (or object) when interacting with others on the internet; your brain fills in the gaps to say there's a person on the other end, but you never know that for sure [1].

For interactions on the internet, I don't think there's any meaningful differentiation between human interaction or object interaction; it's all just interaction that acts like interaction with any human would. You judge that conversation based on its content and quality and continue it if you're enjoying it (or find whatever value you're interacting for), or just move on if you're not -- regardless of who's on the other side of the screen.

[1] https://en.wikipedia.org/wiki/On_the_Internet,_nobody_knows_...

I think the ultimate goal here is to build the Metaverse. This was a big strategic initiative for Meta, so much so that they changed the company's name. But the product wasn't prime time ready when Zuck first introduced it. With LLMs, it's going to be a core enabling technology to make the Metaverse more immersive. Hundreds of millions, maybe billions of simultaneous conversations running inference locally with various digital avatars.

Do people want to hang out in a sexed up habbo hotel though?

Meta (well, SV) is too prude for that. It will be a controlled, sterile and ad-laden habbo hotel.

They absolutely do. Check out the subreddit for Character.ai users. It's a million teenagers too nervous to talk to their IRL crush, but they can live out fantasies with their dark brooding cyber boyfriend no problem.

Cool but that’s a minority of people. Like I’m sure on whatever unpleasant Chan sites there are a bunch of people. But they are weirdos.

I am not sure of this considering the TAM of pornography. With each generation being more insular, having less opportunity to meet and interact in third spaces, etc it seems like people interacting at scale with bots continually improving is the natural trajectory we’re already on.

AI boyfriend, girlfriend, companion, whatever is obviously the next step. Way less effort than navigating the in person social environment, expectations, etc.

The key here is engagement. I have been increasingly hearing people complain that they don't see content from their friends as much or at all anymore on social media. LLMs are a way for Facebook to present information that it wants to present to keep you engaged, as if it came from someone you care about. That seems like it would keep people even more hooked than presenting information from people you don't know.

That's interesting. I don't use Facebook, but I would have expected anyone annoyed with not seeing content from their friend to be even more frustrated if the content they do see is a bot posting as their friend.

It'll be really interesting if this approach were to actually work at keeping people engaged as if they were interacting with friends.

I think it's more interesting when you consider that the bot may genuinely post things that reflect the opinions and experiences of the person that it is posting on behalf of. The real person may seamlessly jump in mid conversation, and that does a lot to validate the experience of talking to the bot.

I think it's an extremely unhealthy path to walk down, but I can see why they're doing it.

Embedding ads deep into language models that then seamlessly and unobtrusively roll them into their responses is definitely the endgame for commercialization. Nobody wants to be the first to do it of course, since they'll get dropped by everyone like a billion degree potato.

Do they assume people are stupid and won’t see these as ai bots?

In my experience, many people can't even identify blatantly AI generated images as such on social media. Determining some text came from an LLM will probably be even harder for that crowd.

> they assume people are stupid

i think that was the fundamental assumption from day 1 (cue the "they trust me" quote)

They could instead assume that people won't care. Based on the thread here there are people who said they wouldn't care, and I could see HN being a population that would would be more biased against obvious LLM content than the rest of the crowd.

Just because Open AI and co have taken the strategy of intentionally training the style of output to be stilted and robotic with RL doesn't mean it's a fundamental limitation of LLMs. LLMs can sound way way more natural and indistinguishable than GPT-4, Claude etc do by default.

Even as things are, people can't actually easily tell. Has nothing to do with being stupid.

The wave of AI friend apps also seems like a semi-persistent trend. If the goal of a social media company is to fulfill a social need. Then we should expect them to try their hand at AI friends.

If a technology is an existential threat to your business I think it makes sense to try and control and understand that technology. Someone is going to build this stuff and ruin social media anyway, might as well be Meta so they can put it to good use or better understand how to prevent those problems.

It's sort of (but not entirely) similar to what happened with Kodak and digital cameras.

It's also more interesting when Meta (and Google) are so dependent on advertising income, yet their moves in AI directly oppose that income stream. Views based ads are worthless on a platform with AI. Click based ads will also become worthless when bots can click through. Nothing gets purchased and brand value doesn't mean anything if few humans even see it.

I suppose none of that matters when you operate as they do, both as the advertising network and also the agent that keeps the ad prices artificially high. Guess we'll have to wait for companies to make this realization a decade later after wasting billions advertising to bots.

I don't think they directly oppose it. They're not creating views. They're creating content for views (and thus ad impressions).

What’s the value of looking at fake news all day *shrug* I wouldn’t be surprised if people were into this.

There’s nothing surprising here. Facebook isn’t appealing to young people and user engagement is falling. Right now they’re just throwing ideas on a wall to see what sticks. Threads didn’t quite took off either.

whats also puzzling is giving away the foundation model

the llamas will be used to pollute the social media reducing their value to advertizers

I wonder if they have a watermark built in that allows them to detect text generated by Llama easily.

https://www.youtube.com/watch?v=XZJc1p6RE78

https://arxiv.org/pdf/2301.10226

I'd be suprised if they weren't making sure that Facebook could recognize content made with their tooling. Watermarks will always be on the honors system though, people could always remove them before posting.

For all the talk about safety it's surprising to me how little watermarking is going on, these AI producers are content with generating unmarked garbage.

I figure there's a lot you could do with interwoven zerowidth space characters and other unicode tricks (using other code blocks for otherwise normal characters like you would when spoofing a url) - sure it would be easy to write software to normalize it back to ascii but at least that requires intent to deceive - we could even write laws against stripping encoding schemes meant to identify automated content

Prior art is that time Genius encoded "red handed" into morse code by way of alternating apostrophes to catch Google copying their lyrics: https://gizmodo.com/genius-claims-it-busted-google-stealing-...

If I'm not mistaken, didn't the major LLM companies try to band together for watermarking type features only to walk it back later and admit it wouldn't work?

I am surprised to see this: https://www.404media.co/where-facebooks-ai-slop-comes-from/

dont know how valid this is, but encouraging ai-slop as a business strategy to keep sticky customers !

It’s the perfect echo chamber to get woke to go back meta, you can hardly have any disagreement if the bot is leveled on yourself

I think this is the real killer app here for users: A custom echo chamber, specifically tailored to what the user wants to read, brought to you by virtual friends. I hate to say it, but that would be eaten up by the growing cohort of lonely (sometimes estranged) elderly people.

I stole a sneaky peek at an elderly family-friend's Facebook feed once, and it was just a constant drip of political memes from a handful of people in a large-ish Facebook Group. Now imagine if, instead of relying on a handful of people keeping that group alive, he could populate it with bots that feed him a deluge of AI-generated text and memes that all agreed with and validated his opinions...

There is more value in manipulating people to fit an echo chamber you control than in tailoring one for each person, losing the advantages of controlling thousands with a unified creed.

Reply on news.ycombinator.com