17 ms·

YouTube provides its creators LLM-generated replies right in the interface, apparently trained on some of the creator's actual replies: https://www.youtube.com

• mrbungie 2y ago
  Good-faith rationale: None. Rationale: More growth for less effort, even if there is no soul. Fitting for our times.
• janice1999 2y ago
  > It seems like the only use case is to trick people into thinking they're having a real interaction. It's easy to rationalise a time saving measure I guess. I feel I'm unauthentic when I use auto-generated response suggestions in Outlook. But, like OnlyFans, it's 'just business'. Perhaps I'm overthinking it. How genuine and heartfelt can OnlyFans responses be? Probably as much as my response to a budget approval.
  • nickff 2y ago
    The problem is that this seems like DAAS (deception as a service), possibly even fraud.
    • prmoustache 2y ago
      Really? Do people really think OF "performers" really enjoy interaction with random dudes fapping in front of their screen. Shouldn't there be a level of candidness at which it cannot be considered deception? I mean we accept casinos/gambling and religions in our societies for example.
      • lazide 2y ago
        Fraud is lying for financial gain. If the site is saying ‘chat with X for $5/min’, and you never have any interaction with X at all for your $5/min, that seems like a pretty clear cut case doesn’t it?
        • genewitch 2y ago
          "Give god the first portion of your income. Don't spend it and come sunday say i'll just give god a little of what's left. Give god the first portion of your income!" Yeah, ok.
          • dylan604 2y ago
            Okay, I've been in and around both sides of religions, and I do believe this is the first attempt at saying tithing is an act of fraud. I don't necessarily disagree. Just noting that it's the first time. I'd assume the down votes were expected with this one.
            • genewitch 2y ago
              Yes, expected. The explicit order is to give "god" the money. This confers a benefit to the tither, per spec (the spec is kinda old.)
              • lazide 2y ago
                “But why does god need money?” - soon to be murdered guy.
      • hmcq6 2y ago
        I have several friends who are "creators" who answer their own messages. I had an ex who warned me that she wouldn't mind if I watched porn but she wouldn't want me to use onlyfans because the platform encourages 1:1 communication. I wouldn't expect, like, Mia Khalifa (I'm out of the game, who's current?) to answer her own messages but there does seem to be an expectation that smaller creators are actually behind the accounts.
    • TeMPOraL 2y ago
      It's not, not anymore than it already was. I guess there is some truth to the adage that a male body can't push blood both to the brain and to the penis at the same time - it's either one, or the other. I mean, if you step back and take a cold look at the content produced by popular OF performers, it's pretty apparent those are business operations. There are other people involved. Someone does the camera work. Location work. Promotion. And yes, fan management too. Same applies to YouTube personas, too. Brands, not people. Talking to fans is a job that can be outsourced. AI doesn't introduce anything new; AI features on the platform are competing with creators' brand management agencies, which themselves likely already use ChatGPT for this anyway, because why wouldn't they? ChatGPT does better job at this stuff and is cheaper than human labor - but again, the AI angle isn't important here - that human labor was just a generative model in a protein substrate anyway.
      • nickff 2y ago
        >"It's not, not anymore than it already was." Maybe it already was fraud. I expect to see class-action lawsuits on the way soon.
        • TeMPOraL 2y ago
          It wasn't. It perhaps should've been, but instead, it was just "legitimate marketing practice". This is another facet of a larger phenomenon, that I just commented on in another thread[0] - there's a lot of harmful activity that's considered legitimate, thus invisible to our legal and cultural immune systems, yet no less harmful than the slightly less legitimate scams. -- [0] - https://news.ycombinator.com/item?id=42401894 https://news.ycombinator.com/item?id=42401894
          • nickff 2y ago
            Your linked post seems to be a polemic, not a legal analysis. Many advertisements are subject to successful (expensive) class-action lawsuits based on some fraud or truth-in-marketing claim. Also, this isn’t an advertisement, it’s a communication which takes place under an implied contract.
• nonrandomstring 2y ago
  > to trick people "Artificial Intelligence" -> noun artifice: ruse, clever trick, guile, deception, cunning, a skillful or artful contrivance
• deadbabe 2y ago
  On OnlyFans you get so many messages there’s no hope of replying to all of them manually. So it’s either reply to basically no one, or use automated methods.
  • space_oddity 2y ago
    If fans are paying for a personal connection, doesn’t automating replies risk undermining that value?
    • Almondsetat 2y ago
      This is like plane overbooking. Airline companies could do the right thing and just sell exactly the seats left on the plane (in this case, the OF creators could put a limit to the people being able to DM them) but instead they choose the extra profit by selling more tickets than there are seats (OF creator allowing an ungodly amount of people to flood their DMs) and just take the blame the few times the statistics goes against their favor (someone figures out they have been talking with an AI/employee)
      • space_oddity 2y ago
        That’s an interesting analogy, but I think there’s a key difference here. With plane overbooking, passengers still get the same core service
    • deadbabe 2y ago
      You can still reply personally to a few whales.
• Karellen 2y ago
  Worth noting that - mostly - punters weren't having real interactions with the models anyway. As the article points out, they'd previously reported (and I think HN had linked to the story) that a large number of models had been outsourcing their DM interactions to a rotating cadre of gig-workers already. And as the gig workers wouldn't be able to keep track of the full chat history between each punter and "the model", the conversations could sometimes feel off, or have long-term inconsistencies. I guess LLM hallucinations will just give a slightly different flavour of unreal interactions.
  • xethos 2y ago
    IIRC (from mobile), it was a different article from Wired that corroborated this, with a surprising disagreement (from what you mentioned) being that the gig-workers would keep notes that are attached and shared. The notes are per-... random? I suppose they'd be called? The gig-worker's customer is the streamer (or agency) that pays them, so I don't have a better name for the hoarde of people they're meant to be chatting with Though I don't doubt that less careful (or less well-paid) gig-chatters exist and can have a slightly different feel when handing off a random at shift-change
    • PoignardAzur 2y ago
      It's kind of fascinating to consider the level of automation and UI affordances you can apply to faking authentic human contact.
  • echelon 2y ago
    > real interactions with the models anyway The next step is not having real models at all. It'll be 100% AI all the way up and down the stack. Completely AI porn is going to be booming business. PornHub, OnlyFans, and perhaps even some dating websites will be "innovator's dilemma"'d out.
    • add-sub-mul-div 2y ago
      Maybe I'm a hopeless optimist but I think the novelty of this will wear off too quickly for it to become the future. People will have to realize the emptiness of it. I need to believe that.
• SirMaster 2y ago
  I'm OK with it as long as it would be labeled as such... I think users need to demand that if it's AI generated that it's labeled as such.
  • mewpmewp2 2y ago
    But ultimately there would be external tools that can do those messages on behalf of them, so it really wouldn't matter.
    • account42 2y ago
      Making clear what the expectations are does matter. Abuse of expectations can be dealt with bans and/or prosecutions for fraud.
• immibis 2y ago
  That is the business model of most media.
  • evan_ 2y ago
    I guess so, to a point. Jimmy Kimmel wants you to feel like you're a part of a community because you watch his show and get all of his inside jokes, but he's not sending you fake text messages pretending to be your close personal friend. Seems different to me!
    • lupusreal 2y ago
      The parasocial television friend hustle was developed before that was possible, or else they'd probably all be doing it too. As it is, they stick with their establish form and don't innovate because their audience is people who haven't and probably won't ever adapt to the newer forms of the hustle. If parasocial hustling is food service, then television talk shows are Cracker Barrel.
  • olyjohn 2y ago
    Yeah well most media is trash. So yay for even more trash?
    • wruza 2y ago
      If no one addresses the roots, why bother condemning the consequences? Eight billion people watch bigcorps screw them up in all poses and nobody bats an eye, cause it’s “legal”. We all get what a few democracies count as normal.
• derefr 2y ago
  > what is the actual good-faith rationale for using this feature So, responding to viewers increases engagement, and thereby a channel's virality. And one human can only do so much of that. So eventually, you hit a marketing scalability bottleneck — you get more comments than you can read, and the people who don't feel engaged with, are more likely to churn from your viewership, so your viewership growth starts to decelerate. Large media companies + MCNs previously solved this bottleneck, by hiring paid human community managers to scale responding-to-comments. But individual creators bootstrapping their growth, had no good solution to this (besides joining an MCN), because the too-many-comments threshold comes long before they make enough revenue to afford to hire their own community managers. This creates a pay-to-win model where companies who already have capital from other ventures can afford to circumvent this bottleneck and so get big on YouTube, in a way that individual creators cannot. And YouTube doesn't like that; those big companies aren't beholden to YouTube in the way that creators that think of themselves fundamentally as "YouTube content creators" are. (Or, to say that in a nicer way: YouTube wants to democratize content creation, ensuring that there's a way for small bootstrapped content-creators to "make it.") AI comment responding is a substitute good for the paid human community managers who already perform this function for large media companies / MCNs / etc. It serves to allow these independent bootstrapped content-creators to overcome the responses-to-comments marketing bottleneck for much lower cost. This doesn't do anything good for the people who post comments, of course; but it does work to ensure a healthy ecosystem of independent bootstrapped content creators, rather than an oligopoly of media companies — which is something that viewers want from YouTube. (An analogy might be to level-1 CSRs in a call center: as a complainant, they just get in your way; but they solve a customer-service scalability bottleneck for the company, which thereby allows the company to grow past the point where it would otherwise stop being able to handle support at all due to the increasing flood of "nonsense" complaints.) > It seems like the only use case is to trick people into thinking they're having a real interaction. Yes, it is, but that's going to happen whether or not there's a built-in feature to do it. That ship has sailed literally centuries ago — ever heard of writing to a famous author/actor/etc, and getting a hand-written response, seemingly from the famous person themselves, but actually from their agent and just signed by the famous person?
• holoduke 2y ago
  Thats a general issue. Lots of comments here could as well be ai generated. You never know for sure at least.
  • johnnyanmac 2y ago
    Could be, but I believe HN can determine such things. And Dang's made it clear multiple times that this comment section is for humans, not bots.
    • TheDong 2y ago
      How could hacker news determine such things? This comment is 100% AI generated, straight from an LLM (admittedly a small and specifically trained one). My first and likely only AI generated comment. Can you tell it's AI generated? Can dang?
      • johnnyanmac 2y ago
        On a micro level there's no clear way to tell, nor is it really valuable. Like most engineering there's a matter of "good enough" and you get surprisingly close to that by simply tracking down the most obvious accounts. It couldn't be completely automated out regardless because intent is important as well. Using an LLM to try and translate yours or a foreign post has more honest intention than a soliciting bot or an otherwise disruptive user that is breaking rules. Now with that context in mind: > How could hacker news determine such things? Good question, I'm curious as well. I'm not well versed in the state of tracking such behavior. But I'm sure any site of size has needed to prepare for this for a while.
        • 8n4vidtmkvmk 2y ago
          It's a cat and mouse game, same as all other spam. There are AI detectors, but then the AI get better and the detectors aren't much better than 50:50 ... I don't know how we'll combat this in the future but I don't think detectors are the answer
• Applejinx 2y ago
  Yes, I'm experiencing this in the form of 'suggestion' buttons. It's so clearly trained on my own replies that it parrots stuff I've said, but it tends to get the sense of the words literally backwards and wrong. If I used that I would be telling my fans the opposite of what I actually meant, or various other catastrophically not-even-wrong assertions. It really, really is not figuring out what it's being fed. Sometimes I let people in on what the AI is suggesting I say. It's not actually wresting the controls from my hands and talking to my fans AS me… yet.
  • genpfault 2y ago
    > It's so clearly trained on my own replies that it parrots stuff I've said, but it tends to get the sense of the words literally backwards and wrong. "Yeah back then I was weirded out by their now standard response prediction thing, it was always just one or two words off exactly what I wanted to reply (usually missing/adding a negation, or substituting a word with an antonym)"[1] [1]: https://qntm.org/perso https://qntm.org/perso
    • kybernetikos 2y ago
      Feels like that ended with an ai escape when I was hoping that the punchline would be that all the apparently happy, well adjusted, coherent users were in fact ai trained in the preapocalypse and the insane sick, repetitive people were the real users living in the post apocalypse.
• everdrive 2y ago
  That's depressing, and I didn't realize. This is truly the death of the the comment section.
  • ClassyJacket 2y ago
    It's like when you find out the band you like probably didn't respond to you, but just some social media manager.
    • genewitch 2y ago
      I got a friend that goes to live shows in Los Angeles and then interviews the bands after the shows for a decent amount of time and with decent quality. mostly punk rock bands. I've personally met Les Claypool and Trent Reznor and a bunch of smaller artists after shows. However i've never tried to email or otherwise contact any musicians or whatever, except one time in the late 90s there was hosted AOL chatrooms or something similar on yahoo or something where many fans would be in a room and get selected to ask a question that the person of the hour could see, and they'd reply. I also got to ask Reznor a question like that, 26 years ago or something. I may be an outlier, i've actually met a couple dozen famous people, because a lot of my friends were (strong were there) industry adjacent. Like, i've been to NAMM multiple times, and iirc the tickets were quite pricey and limited.
      • account42 2y ago
        Yeah, there are some not-nobody bands that do actually talk to fans after shows - as in face to face without special appointment, just hanging around at the exit merch store. I would rather have those "celebrities" that don't want to directly interact with plebs be honest about it. Nothing wrong with setting boundaries for your job, a lot wrong with deceiving people even if it might seem benign at first.
    • johnnyanmac 2y ago
      gets a bit sadder when you paid $1000 to send that letter out. But I don't think bands commonly promise that they hand-respond to every letter.
• TeMPOraL 2y ago
  > what is the actual good-faith rationale for using this feature? There is none, because the whole thing was never good faith. It's always been about tricking people into having a real interaction, to make it easier to exploit them for ad clicks or to sell them junk. That's the whole YouTube content creator business, and it's what it always has been.
  • xvector 2y ago
    This is a bunch of hand-wringing. If the AI reply is what I would have said anyways; awesome! Finding thumbnails is hard, but if AI comes up with a good one? Fantastic! None of your interactions except the ones with your friends and family are authentic, and they never were. It's a bit silly to get annoyed over OnlyFans models or some mega YouTube celebrity using AI because you're losing out on "authentic human interaction" - you only mattered to them insofar as you provided them with money to begin with.
    • EGreg 2y ago
      This is a silly take, I think. The point is that people pay to have a genuine interaction. If everything is fake, why interact at all? Imagine you were typing into the void and no one saw what you said. Would you continue doing it if you knew? How about on this forum?
      • JSteph22 2y ago
        If I use AI to generate the text of a happy birthday greeting to grandma, if I review and approve of the message, is it genuine?
        • nosbo 2y ago
          I would say no.
          • taberiand 2y ago
            What if you use it to give you ideas and you come up with your own greeting. And then you generate one more, and it's the greeting you came up with.
• cyanydeez 2y ago
  They're already tricked into ceding power to billionaires. We are way passed the Rubicon on the ethics of capitalism and propaganda as a service.
• bdd8f1df777b 2y ago
  China's Weibo also has a similar chatbot, but the chatbot is a separate account with clear indication that it is a robot.
• insane_dreamer 2y ago
  IG has this feature as well - or at least was trialing it. I got an AI-generated response to a comment on a post -- I was not amused (or impressed), but at least it was clearly labelled as such. Still warranted an immediate unfollow. It's hard for me to see why anyone would want to receive AI-generated replies, like what's the point?
• manquer 2y ago
  They are selling an experience, that includes the style of content, conversation, personality or persona of the creator. Given that premise the functionally the same whether the creator or the app using an LLM responds, as long as you can keep the immersion in the role play does it matter if its the Youtube or OnlyFans actually responding? It was already some underpaid offshore worker responding for the large creators, and this is nothing new, ATN and other phone sex line providers were doing this in 1990s and there were probably similar things in times past.
  • eloisant 2y ago
    As soon as they're clear about that, sure. The problem is that they lie to their fans and selling it as a conversation with the creator.
    • manquer 2y ago
      We weren't talking to a sexy young attractive lady on those phone sex lines either. Some didn't care and still used them, many were scammed, but that is the nature of the industry. Sadly it has always been on the consumer of unregulated sexual services to be aware of how the industry of their era works. Should we legalize, regulate and make it safer for both the buyers and also sex workers? Yes, but that is an another conversation altogether.
• space_oddity 2y ago
  The line between helpful automation and deceptive practices is thin
• _yb2s 2y ago
  Google has also outsourced the rationale for their actions to LLMs.
  • speedgoose 2y ago
    Hopefully it's not only trained on their own data.
• robertlagrant 2y ago
  People already weren't quite a lot. For any channel that can afford it, other people write messages, not creators. This just automated that. These aren't real interactions.
• llm_trw 2y ago
  How far we have fallen that a few bytes sent to a server are a 'real interaction'.
  • freehorse 2y ago
    To the extent that we think that even real-life interactions are just byte exchanges themselves, and human mind is just a processor of information.
    • timeon 2y ago
      These nihilist arguments often ignore physical, nonverbal interaction. You can interpret it as byte procession in the end but it is still missing from digital interaction.
• vachina 2y ago
  YouTube’s best interest is ad, and having more content to retain you, the consumer, on the platform as long as possible would be their main goal. Every single online content platform operates in the same way. Their KPI is number of hours spent on their platform, because the longer you spend there the more ads they can shove down your throat.
• vasco 2y ago
  I mean they started calling people content creators and the people liked it and thought that was a compliment, that's all you need to know. A content creator is not an artist or a videographer or a grandpa sharing some craft. A content creator is not even human, it's whatever can fill the <div id='content'></div>. You only need programmers for everything else, but that damn div was hard to fill until now. Soon enough they'll have what they wished and then they'll realize the div that managed to sell ads is probably becoming useless.
• itsoktocry 2y ago
  >It seems like the only use case is to trick people into thinking they're having a real interaction. As opposed to the genuine interaction the porn star you follow offers to you and the 10,000 other subscribers. The whole thing is theatre.
• highwaylights 2y ago
  I sort of get why this feels dishonest, but in the same way I’m not sure social media is something that users expect to be a reciprocal experience in 2024, save speaking to ones own immediate circles. Kids might believe that MrBeast is replying to them personally, but I think some Internet education goes along way to help here, and I don’t know of anyone of legal age to be engaging with these platforms that would expect genuine “celebrity” responses to their comments.
• jstummbillig 2y ago
  Assistants/employees are being democratized. That's a good thing, because it levels the playing field. In the end, you are vouching with your name. If you feel the suggestion is missing the mark, don't hit send. If the suggestion is missing the mark most of the time, the product is bad and will be shut down.
• hmmm-i-wonder 2y ago
  >trick people into thinking they're having a real interaction. I'm curious what real interaction means here. At best the person is pretending to be a character with the intention of monetizing it. Having an AI play the character vs a person doesn't seem to make it "less real" considering the entire premise is, in this case, selling a fantasy and character?
  • account42 2y ago
    If it doesn't matter then why is it hidden?
• snarf21 2y ago
  Things seem directly headed to the movie Her where your monthly subscription is mostly used to pay for hosting your personal interaction model in the cloud.
• heresie-dabord 2y ago
  > It seems like the only use case is to trick people into thinking they're having a real interaction. This is a broadly successful economic use-case.
• RobKohr 2y ago
  I would so much rather the creator not participate in the comments than have a fake AI that doesn't know the creator's mind in the comments. I don't care if someone with a million viewers replies just to me. Almost any video's comment stream, you can scroll through, get a general sense of what people's reactions/questions/etc are to the video, and then good creators respond in one or two comments to the masses. The second I see someone with a huge following obviously AI responding to each of the comments, I have lost all respect for them, and people will call them out on it. I see this a lot on reddit now where people are responding to AI generated posts pointing this out, and it is causing some major issues in communities.
• pengaru 2y ago
  The use case for Alphabet is to control more of the interactive bandwidth so they can steer the influence on behalf of the highest bidder / inject increasingly subtle conversational advertising. You can bet your ass that in the fullness of time the generated content is going to put Alphabet's interests first.
• AlienRobot 2y ago
  The other day I was helping an elderly family member purchase something from the Internet. They were talking to a chat bot. One of the automated replies was a voice recording. There was not much "AI" in it, just a bunch of spam in Whatsapp all the time telling them to purchase all sorts of products. They were added to multiple "group chats" where only admins could send messages (i.e. a spam fest). But what worried me was that because the automated messages are never marked as such, I wasn't sure if they were perfectly aware that they were talking to a bot, since it seemed they were tried to have a conversation, saying hello and explaining things. Maybe that's just they think that's how bots work, and, loosely, the bot responds appropriately, by ignoring almost everything all the time and just spamming more. I found it repulsive on a visceral level. Add AI to this and I'll need to start praying to God to give me the restraint of not breaking other people's smartphones.
• anigbrowl 2y ago
  Ugh. Another reason I stay signed out nowadays.