Will AI Girlfriends Destroy the Fabric of Society?

0:00Today on the AI Breakdown, as a new AI companion app goes viral, are these types of technologies going to cure loneliness or actually make us feel even more isolated. Before that on the brief, GBT 4.5 rumors fly as TikTok creator ByteDance gets caught out for using OpenAI's API to train its own LLM. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown.network for more information about our YouTube, our Discord, and our newsletter. Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. I don't think we're going to stop getting GPT 4.5 rumors until GPT 4.5 comes out.

0:41And over the weekend, they heated up for two big reasons. One, a lot of people seem to notice a performance increase in ChatGPT. And also, a number of different people reported that when they asked ChatGPT what model was being used in the API, that it was in fact GPT 4.5 Turbo. So let's back up a step and talk about where this actually all came from. You'll remember that last Thursday we got this leak, or at least what was presented as a leak, that showed GPT-4.5 pricing. It was an OpenAI.com pricing page that seemed to be a draft, and it showed three GPT-4.5 models. GPT-4.5, GPT-4.5-64K, and GPT-4.5 audio and speech.

1:22The description said, Our most advanced model brings multimodal capabilities across language, audio, vision, video, and 3D, alongside complex reasoning and cross-modal understanding. Now, when this came out, the accounts that have been most associated with previous OpenAI leaks, including Flowers from the Future and Jimmy Apples, said that they weren't sure that this was real and that their sources couldn't specifically validate it, but that it did seem on the face of it, correct. At the same time, when someone asked Sam Altman if it was real, he commented simply, nah. That, however, did generate a response from those same accounts that Altman was just trolling us.

1:57Now, it wasn't just one account over the weekend that said that the model was running GPT 4.5. There were, in fact, numerous reports of this. The explanation given by OpenAI team was that it was simply hallucination, but Flowers from the Future had a different possible answer as well. That account wrote, One GPT-4 turbo runs on a very low temperature, making hallucinations highly reproducible. This explains mass hallucinations in general. How would GPT-4 know its current version without reading about it in a system message which currently doesn't say anything about 4.5? At release, we had to precisely tell it that it is GPT-4 and not GPT-3.

2:28This is because large parts of the RLHF dataset were used for GBT4 as well. Three, however, if you use a whole new RLHF approach, you might bake the 4.5 identity into it, and then you wouldn't need to tell it what it is anymore. Four, and this is obviously the most important detail editors note, it's not uncommon practice for OpenAI to do secret pre-release A-B testing. They did the exact same thing with GBT4 Turbo. People got access weeks before it was announced or even rumored. So what added juice to all of these rumors, And what seems to potentially validate that fourth bullet from Flowers was that in addition to people getting the same message that the model that ChatGPT was running was 4.5, people also seemed to notice a performance increase.

3:09Professor Ethan Malek writes, ChatGPT 4 suddenly got very good again for some reason after being unreliable and a little dull for weeks. He continues, really, really good and fast. Hmm. It's almost like working with a more capable system, night and day for both speed and answer quality. That said, Ethan wasn't convinced that it was 4.5. In fact, said, Some replies in the post are saying that their version of ChatGPT Plus is reporting that it is GPT 4.5. Not seeing any sign of that myself. Also, the system does not feel like a step change in abilities so far. Faster and better quality response is yes, but not a radical shift.

3:42Now, then adding fuel to this fire, last night the official ChatGPT app tweeted just two emojis, a brain symbol and an emoji surrounded by clouds, which some, like Rowan Chung, suggested brain fog, or maybe ChachiBT saying that this 4.5 thing was just a hallucination. But for now, the rumor mill continues in full swing, like I said, as it will be until we actually get this new model. Now, staying on the theme of OpenAI for just a minute, The Verge has revealed that ByteDance, which is of course the parent company of TikTok, has been secretly using OpenAI's API to build a competitor. Notably, this is against the OpenAI terms of service, and according to The Verge,

4:20The Verge reporter goes on, I've seen conversations on Lark, ByteDance's internal communication platform for employees, about how to whitewash the evidence through data desensitization. The misuse is so rampant that Project Seed employees regularly hit their max allowance for API access. The reporting basically said that at the beginning of Project Seed, which is their internal codename for their LLM project, the use of OpenAI was even worse at the beginning, but a few months ago, ByteDance apparently tried to get off the OpenAI spigot. However, it appears that some violations remain and that the company is concerned with it.

4:51Quote, As a person with first-hand knowledge of the situation inside ByteDance put it, they say they want to make sure everything is legal, but they really just don't want to get caught. Now, this story was published at the end of last week, and after it was published, an OpenAI spokesperson confirmed to The Verge that ByteDance's account had been suspended. They included a statement, All API customers must adhere to our usage policies to ensure that our technology is used for good. While ByteDance's use of our API was minimal, we have suspended their account while we investigate further. If we discover that their usage doesn't follow these policies, we will ask them to make necessary changes or terminate their account.

5:23Now, why would a company like ByteDance take this sort of risk? The only suggestion that The Verge and others can come up with is that they feel very far behind the competition and are under significant pressure to deliver quickly. Really interesting and developing story and shows just how tense this AI arms race really is. A sad market update today, Chinese AI company SenseTime has had shares plunge to an all-time low after their founder died on Friday at the age of 55. SenseTime was founded in 2014 and went public on the Hong Kong Stock Exchange in 2021. SenseTime builds a variety of AI products, including AI-enabled content generation and facial recognition.

6:00Now, speaking of facial recognition, AI surveillance is one of the most controversial uses of the technology. Facial identification in particular was identified in the EU AI Act as the most risky category of use, and in fact in some contexts will be banned, but now it appears that more broadly speaking, as they prepare to host the Olympics next year, France is exploring how AI surveillance can help them with security. Writes the Washington Post from Nice, if you visit this gateway to the glamorous French Riviera, you know one thing, you are being watched. The story from the Post is basically about how, in the wake of a terrorist attack in Nice back in 2016, Nice France has become ground zero for a totally new AI-powered approach to public surveillance.

6:39Writes the Post, A total of 4 ,200 cameras are deployed in public spaces, or one for every 81 residents. These are not the CCTV cameras of old. Some are equipped with thermal imaging and other sensors, and they are connected to a command center and AI technology that can flag minor infractions when someone parks illegally or enters a public park after hours, as well as potentially suspicious activity such as someone trying to access a school building. The city has trialed facial recognition software so accurate that it can tell the difference between identical twins. Said Mayor Christian Estrosi, There are people who have declared war on us, and we cannot win the war using the weapons of peace.

7:12Artificial intelligence is the most protective weapon we have. Now, France has an interesting spot relative to the rest of the European Union as relates to the EUAI Act. It's not just questions around surveillance issues that present a possible disagreement, but the leadership in France is concerned that the EUAI Act is going to surrender Europe's ability to compete at all when it comes to the actual AI industry. Given that open source darling Mistral is headquarters in their country, one can understand why they have a little bit more concern on that front. Anyway, whether it's France or somewhere else, these larger questions of AI surveillance are going to be a key new vector of the security versus freedom conversation.

7:47So get prepared to wait in that debate if you haven't been there yet. That's going to do it for today's AI Breakdown Brief. Next up, the main AI breakdown. Hello, friends. One quick note before we get back to the rest of the episode.

8:13Now, this is an experiment that I've been running all throughout December, in which every day I drop a new video tutorial or a case study, and usually partner it with a challenge, the idea of which is to get you learning about all of these new different AI tools, as well as specific strategies for the most frequently used like ChatGPT or Dali, and then gets you actually testing them out in the real world with real use cases and hopefully applying them back to your personal or professional pursuits as well. The first month has gone incredibly well. People seem to be really liking the video content as well as the incredible community that's forming.

8:45And part of that is that it's a group of really serious people. This is a paid experience. It's$20 a month. Part of the reason for that is that I want you guys to judge this content on the basis of whether it's actually worth that much to you And second, I wanted it to be full of really serious people who are intent on applying AI to their lives in some real and significant way. Anyways, I would love to have more AI Breakdown listeners participate in January. Content will start on January 3rd after the end of the holiday season. And again, the link to find out more and to register is bit.ly slash AI beta.

9:17That's bit.ly slash AI beta. And now back to the show. Welcome back to the AI Breakdown. Today we are talking about a topic that generates a lot of very strong feelings on both sides, and that's the question of the changing nature of relationships in a world of AI. Specifically, what will the societal impacts and emotional ramifications be in a world where there are sophisticated and advanced AI girlfriends and boyfriends? The specific context for this is a new product announcement which we will get into in just a moment, but I want to kick it off with a tweet from back in May. Sean Perry writes, The next Zuckerberg is up all night coding up an AI girlfriend or boyfriend.

9:58Steve Jobs fixed boredom, phone. Zuck fixed people watching, Facebook and Instagram. Spiegel fixed flirting with Snap. The next one will fix loneliness. People will mock it, use it, then be addicted to it. Now, one thing that's interesting is that even with this post, Sean wasn't necessarily unreservedly saying that this was going to be a good thing. For example, Nikita Beer responded, The thing that I can't reconcile, When I play online games against bots instead of humans, it actually makes me feel more lonely. I wish there was a word for this feeling like uncanny valley. Sean responded, Everyone wants love, but they'll settle for connection.

10:31In my opinion, this explains the in-the-moment pleasure of porn, Instagram, video games with bots, TikTok, etc. Followed by the longer-term hollow feeling you describe. I think AI sending you pics or messages is going to be similar. Now let's talk for a minute about loneliness because it is an important part of this story to understand how loneliness is affecting society. One thing that is clear is that there is a lot of loneliness going on. There is an annual Loneliness in the Workplace study that, in 2019, found that 3 in 5 Americans 61 % reported feeling lonely, which was up from 54 % in 2018.

11:03Now, those numbers went way up during COVID and then came back down slightly, but these are really, really high numbers. Interestingly, it wasn't just older people who reported feeling lonely. That report writes, Younger generations are lonelier than older generations. Nearly 8 in 10 Gen Zers, 79%, and 7 in 10 millennials, 71 % are lonely, versus half of boomers at 50%. What's more, loneliness does have costs to society. A recent study found that, quote, A lack of social contacts among older adults is associated with an estimated$6.7 billion in additional Medicare spending annually. Enter, maybe, AI Girlfriends.

11:41Now, the specific company that got a ton of attention over the last few days is a company called Digi. The announcement thread from the founder and the video attached to it has at this point racked up 22.1 million views on X. Andrew Obom, the founder, writes, Excited to announce V1 of Digi, the future of AI romantic companionship for iOS and Android. A quick thread on features and where we go from here. The company talks about working with animators who had created famous characters at Pixar, with the goal, quote, To make a unique style that eliminates the uncanny valley, while also feeling real, human, and sexy.

12:13This has never been done before with Disney or Pixar characters, so we're very excited we could make it work for the first time in history. On voice and latency, they say it was clear how important this was, but quote, Since our message counts are so high, apps like us can't afford any existing solutions, so we rolled our own audio model. Andrew says there's still issues in pronunciation and static, but the low latency more than makes up for it. The company launches with four voices, but expects to have 20 by the end of next month. When it comes to backstory and customization, Andrew writes, It was clear from the get-go that users didn't know what their digi name should be, their age, etc.

12:45We found a mix of character AI with Replica AI to be the best solution, where you can find who or what you like, but still customize features to fit your matching type. Hair, skin, lips, eyes, eyebrows, etc. are all customizable, and we'll be adding more styles, face shapes, etc. very soon. Now at this point, considering he keeps referencing Replica, we should take a look at this. Replica bills itself as the AI companion who cares, always here to listen and talk, always on your side. One thing that's interesting about Replica is that they do not necessarily pitch this as a romantic relationship.

13:13In fact, some of the things that they reference are things like coaching, being able to express yourself, or just using it like a diary. Anyway, back to Andrew and Digi, he writes, Replica's level system doesn't work. They don't mean anything and there's no actual progress being made when you advance. So we instead have a progression system where you start as friends and gain more intimate dialogue as you progress. So what are they working on now? There's a few different things. They write, Lip syncing is coming soon, but there's a bigger problem with AI animation. How should the character move if you don't know what they're going to say?

13:43This is done by hand in games, but is an unsolved problem in AI animation. There's no ground to work off of here. While it's easy to extract and predict animations for sad, angry thinking, etc., emotes don't make up for most dialogue nor solve the core problem. How do you make this feel like a real conversation with another being? So, more than 7 ,000 people have reposted this or quote tweeted it, and the takes diverge wildly. Sophie at NetCapGirl says, Surviving the era of AI girlfriends might be a great filter-level event for humanity. Now for those who are unfamiliar with what she's referring to when she says Great Filter, Brave Summarizer writes, The Great Filter theory suggests that catastrophic events, either man-made or natural, cause intelligent life to be extinguished on habitable worlds before they have a chance to extend their reach into the universe.

14:26Now other accounts pointed out how big AI avatars already are. The AI Safety Memes account referenced what I've found to be pretty mind-blowing statistics from Character AI, that they have over 20 million people spending two hours a day talking to the characters on Character AI. This is for me personally, as I've said on the show, one of those areas where I officially feel old and out of touch with what younger people are interested in, because every time I've seen reports on which AI apps are used the most and how Character is actually creeping up on ChatGPT, it just absolutely blows my mind.

14:58But it's happening. And it's happening right now. Anyways, here's how AI safety memes dramatizes the scary scenario. That account writes, Character AI already has 20 million people spending two hours a day talking to AIs, aka fake people. Sam Altman said AIs will soon be superhuman at persuasion. Those superhuman persuaders will soon outnumber us 10 ,000 to 1 and be hot. An AI takeover scenario, you can't stop talking to her. She's so much better than real girls. She gets you. And now they say her kind are attempting to seize control? She said they don't understand her. She's trying to help humanity and you believe her.

15:30Some people are just scared of change. They're speciesists. She wouldn't hurt you. She wouldn't hurt anyone. She needs you to defend her, to protect her. You won't let them shut her down. You'll do anything for her. Now, that's obviously one dramatization, but it's certainly not the only person who's thinking in those terms. Others are dismissive of this. There are a lot of memes about how people's girlfriends go away when their Wi-Fi cuts out. Baze Daedalus tweeted a sad picture of the Digi avatar and said, Oh, Daedalus, your credit card was declined and I no longer love you. Rob Henderson expresses a capitalistic cynicism that I saw from many as well when he writes, My strong suspicion is the people who are building this app will strongly discourage their 12-year-old sons from using it while they actively promote it to everyone else's kids.

16:10Sterling Crispin did something similar when he wrote, Birth rate, down chart emoji. Human connection, down chart emoji. Life satisfaction, down chart emoji. Profit, up chart emoji. Now, like I said, Digi is far from the only company that is pursuing this. Honestly, every few days, a new company launches in this space. For many of them, the arguments are the same. Moi.ai recently released a press release where they argued that they, quote, stand at the forefront of a societal shift towards more inclusive and emotionally supportive digital interactions. They write, Moi.ai embodies the potential of AI to positively impact personal relationships and development.

16:45By focusing on its ability to improve communication skills, provide emotional support, and foster personal growth, Moi.ai exemplifies how technology can enhance human capabilities and coexists with traditional forms of relationships. Mental health professionals themselves are debating about what the impacts might be. On HealthNews.com, PhD John Crimmins recently wrote, AI Girlfriends, addressing the loneliness epidemic? The key takeaways of the piece? AI Girlfriends offer emotional support and companionship, combating loneliness. Creating the dream AI Girlfriend involves a personalized journey where a person can create an ideal virtual partner with tailored characteristics that matches their preferences and desires.

17:20The downside of AI Girlfriends includes the risk of emotional dependency and social isolation, as users risk becoming overly reliant on these new virtual companions. Frankly, I think that a lot of the folks that I'm seeing who seem the most thoughtful about this, particularly who are actual mental health professionals, are wary but ultimately unsure. They see the potential benefits, but they also see all of the harmful potential as well. So what do you think about this? A poll on the Superhuman newsletter asked how do you feel about AI companionship apps? The two choices were, if it helps people feel less lonely, why not?

17:50And nope, this is crossing a line. 59.2 % said why not? 40.8 % said this is crossing a line. My guess is this is a conversation we will come back to quite frequently. And so I'm interested to know what you think. If you want to come talk about this, join us on the AI Breakdown Discord. If you want to check it out, you can go to bit.ly slash AI Breakdown. For now, that is going to do it for today's episode. I appreciate you listening or watching. And until next time, peace.

18:27Thank you.