The Epik AI Yearbook Trend Taking Over Social Media

0:01Today on the AI Breakdown, we're looking at the wildly viral Epic AI yearbook photos app and what it says about user adoption and artificial intelligence. Before that on the brief, very intriguing rumors that OpenAI is considering fabricating its own AI chips. The AI Breakdown is a podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our Discord, our newsletter, and our YouTube channel. Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. We start off the day with a juicy new exclusive report from Reuters this morning that OpenAI is considering making its own AI chips in order to deal with what has become a huge bottleneck to the development of artificial intelligence, which is access to advanced chips.

0:51So, like I mentioned, this report comes from Reuters. Their sources are people familiar with the company's plans, which tends to mean either investors or employees or partners. Now, what we know is that OpenAI has not made a decision to do this for sure yet, but that since last year, it has been exploring a number of different ways to deal with chip shortages, up to and including actually making them itself. As a way of conveying how serious this possibility is to them, they've actually gone as far as evaluating a potential acquisition target that would help them with these plans. Now, OpenAI's desire to figure out better access to chips is nothing new.

1:25Sal Maltman has been quite clear about how big of a problem it has been this year and how much it has changed their plans. It seems like the GPT Vision tool that we've recently gotten was actually ready a number of months ago, but was held up by access to these advanced chips. On top of just getting access to these processors, the other problem is how expensive they are. Now, of course, it would be a very serious undertaking for OpenAI to actually go out and try to create custom chips. However, it would also reflect the strategy taken by other tech giants in the AI space, including Google and Amazon.

1:54Google has their Tensor Processing Units, and Amazon's chips have actually been in the news recently as they were an integral part of their just-announced deal with Anthropic. While Amazon's investment into Anthropic, which is$1.25 billion so far and can go up to$4 billion, took most of the headlines, one of the big parts of the deal seems to be Anthropic's commitment to use Amazon's custom chips in the development of their future models. From the Anthropic announcement, AWS will become Anthropic's primary cloud provider for mission-critical workloads, providing our team with access to leading compute infrastructure in the form of AWS Tranium and Infersia chips, which will be used in addition to existing solutions for model training and deployment.

2:31Together, we'll combine our respective expertise to collaborate on the development of future Tranium and Infersia technology. So Anthropic gets access to these new chips, you would think at a discounted cost, and Amazon gets plugged into one of the competitors in the major foundation model space. Now, of course, NVIDIA continues to dominate the AI chip space, and there's not really a lot of indication that that's changing anytime soon, but it's clear that NVIDIA alone cannot meet the demand, and that the status quo of 2023 is destined to change. Whether OpenAI ultimately decides to get involved remains to be seen, but it certainly shows how significant an issue this is in the broader AI competition.

3:04Speaking of big companies and AI, one of the trends that we've been watching recently is the shift to AI hardware. Maybe the most notable announcement on that front are reports that Sam Altman from OpenAI and Johnny Ive, formerly of Apple, have been in discussions to build a new type of AI-centric device, and have even talked to Masayoshi Son of SoftBank about a billion dollars in funding for the startup venture. The information called at the beginning of the AI phone wars. Well, if that designation is true, it seems like the existing players are trying to retrofit their current offerings to appear like AI phones, even if that's not exactly what they are.

3:36For example, The Verge's reflection on the Google Pixel 8 launch was that it was a quote parade of AI. The Verge's John Porter writes, I don't know if you've heard, but Google's latest products are filled with AI. There's Magic Editor, a photo editing tool powered by generative AI. There's Conversation Detection, an audio transparency feature powered by AI. There are improved heart rate algorithms, which yes, are also powered by AI. Now Porter ends his review as somewhat skeptical. He talks about the announcement of the BARD update of Google Assistant, and the example that I referenced the other day as well, of how it could help post a picture of Baxter the dog on the user's social media.

4:11Porter writes, As a tech demo for generative AI, it makes sense. AI is increasingly good at recognizing and describing images. And one of generative AI's greatest strengths is writing in a particular style. But ignore the AI part and think about this purely as a smartphone feature, and I think it's utterly baffling. How on earth have we gotten to the point where it makes sense for a smartphone to draft our personal social media posts for us? What's the point? If you're asking a machine to draft an image caption for a photo, then why are you publishing the caption in the first place? What are we doing here?

4:38Now, frankly, this is kind of what I said around that launch as well the other day, that I have a lot of skepticism around what are being presented as the ways that people will use these AI-powered personal assistants. Porter's theory is that, quote, in the absence of a killer app for generative AI, Google is throwing features at a wall and seeing what sticks. It feels like the search giant has a hammer labeled generative AI and its search for nails is taking the company to weird places. I have two qualms with that analysis. The first is that I'm not sure that it's the right characterization to say that generative AI has no killer app.

5:07ChatGPT is already creating fundamental changes in people's workflows across different professions, across education. it is very hard not to describe that as a killer app based on any definition of a killer app I've ever seen. What's more, I think that image generation in the form of things like MidJourney has been fundamentally transformational as well. Every part of the way that I create this podcast is now in some ways touched and shaped by AI, and I think that just because we all know about MidJourney and ChatGPT already doesn't mean that they're not killer apps. But the second point of disagreement with Porter's analysis is the idea that throwing things at the wall and seeing what sticks is a bad thing.

5:41I actually tweeted about this in a separate context the other day. I said point, every startup in the world is just slapping AI on things. Counterpoint, no way to figure out what stuff AI actually changes without trying it. I think the evidence of services like ChatGPT and MidJourney are intriguing enough that it's only reasonable for startups and big tech companies to be experimenting with AI in just about every context they can imagine. I think it would be a mistake to assume that it's actually going to work and be sticky. I've shared at numerous points what I think isn't or is likely to work.

6:09I have a hard time imagining that meta's characters are going to work, but hey, it might not be for me. But I don't think they shouldn't try. I think the whole point of new products and new features and entrepreneurship is to try new stuff. So I say bring on all the silly AI features. We'll figure out soon enough which ones are actually valuable or not. And I'm almost sure that we won't have been as able to predict what works and what doesn't as we might think we are sitting from where we are today. Moving on and speaking of mid-journey, according to the information, Andreessen Horowitz has recently discussed a big investment, a nine-figure investment into mid-journey rival Ideogram.

6:41Now, Ideogram announced the fundraising round only a couple of months ago, but apparently they've been back at the VC table and are talking about raising between$75 and$100 million. Now, obviously, the image generation space is an extremely crowded field. Indeed, in many ways, the competition that people are really looking at in that space is mid-journey versus the new chat GPT integrated DALI 3. But Ideogram's desire to raise money is a reflection of how much compute costs and how expensive it is to compete in these core AI spaces. Now, as of yet, it doesn't appear that a deal has been done, so we will keep an eye out from here.

7:13Wrapping up today with a look at how CEOs are thinking about generative AI, major professional services firm KPMG has just released the results of a new survey of US company CEOs. The results when it came to artificial intelligence were quite, quite clear. 72 % of US CEOs say that generative AI is a quote top investment priority. And interestingly, it seems like people are taking a medium-to-long-term view of how it's going to benefit their companies. Only 23 % of those surveyed CEOs say that they expect a return in 1-3 years, with 62 % saying they expect a return in 3-5 years. Now, as an interesting substory, some have speculated that the rise of AI could push workers back to the office.

7:52In other words, workers might feel more like they have to prove what they are doing, given how much of their jobs might be able to be automated. And these CEOs seem to be very down with that idea. Last year, only 34 % of the CEOs surveyed said that they envisioned their staff working permanently at the office within three years. And this year, that's all the way back up to 62%. Last year, 20 % said that they envisioned their teams as fully remote. And that's all the way down to 4%. Anyways, it's very clear how powerful and important a trend generative AI is. And so, of course, we will keep covering it for you every day here at the AI Breakdown.

8:25Up next, the main AI breakdown. Welcome back to the AI Breakdown. Today we are talking nominally about a trend that has taken over TikTok and Instagram and Twitter and basically all of social media, which is a new feature from an app called Epic that takes people's photos and turns them into 1990s yearbook photos. Now there is a ton that makes this interesting to me. It brings up questions of privacy and data usage. I think it has things to teach us about how people try and get onboarded to artificial intelligence. It brings up questions about the future of self-perception, and it even relates to deepfake policy.

9:01So let's talk about what the actual trend is, and then we're going to look at why it's part of an interesting pattern that goes back to last year, right around the same time that ChatGPT launched. So what is this trend? Well, basically, it is one of these apps where you upload head-on images, give it a little bit of information, in this case things like gender, and then sit back and basically let it do its work. The app turns those photos and makes the user both younger, as well as putting them in a set of clothes and setups that look like a number of different 90s high schooler archetypes. There's the jock, the cheerleader, the nerd.

9:32Now, interestingly, this is not a free experience. People have to pay for these photos. And yet still, it has become the number one app in the US app store across all categories. Investor Olivia Moore points out that this is extremely lucrative. Over the past week, the company has been making around a quarter million dollars per day via in-app purchases, and that's likely going to do nothing but increase as people share this trend more, and more media like this reports on it. Now, Olivia's interpretation is that, quote, Consumers are willing to pay for multiple apps and post the results even if it's cringey.

10:01Allowing people to reimagine themselves triggers something powerful. Now, some think that a better word for this is narcissism. Another venture investor, Justine Moore, says never bet against narcissism. AI selfie apps like Lenza, Remini, and now Epic have proved that you can print millions of dollars in a week by helping people make images of themselves. and they're inherently viral because everyone shares the outputs. Now, this self-imagination is really fascinating. Professor Shelley Palmer writes, The rise of epic has spotlighted a curious trend, students opting for generative AI images in their school yearbooks.

10:32At face value, it's a novel application of technology blending nostalgia with the cutting edge, but it raises questions about authenticity, memory, and our posterity. Imagine future generations flipping through old yearbooks. Look, here's an old-fashioned generative AI image of grandpa in high school. Gee, I wonder what he really looked like at that age. It's a statement that sounds almost surreal. The trend also raises broader questions about the ethics of image manipulation. We already photoshop everything from smoothing out wrinkles to adjusting lighting. We enhance images all the time. Is generating an entirely new image with AI where we draw the line?

11:01If so, why? Where's the boundary between enhancement and fabrication? It's a distinction that might seem subtle, but it will have profound implications. Basically, the point that he's making is that our relationship with representations of ourselves is changing. It's changing because we have access to a technology that we never had access to before. But even if one goes into understanding change without bringing the bias of change being good or bad to it, it's impossible to not recognize that this represents something shifting. Now, as I mentioned, this is part of a pattern of AI-powered apps that suggest to me that there is something more to learn here above and beyond just the fact that people like this stuff.

11:34Back in December of last year, just after ChatGPT launched, ChatGPT was not the only artificial intelligence app that I saw my entire community talking about on their social media. In fact, even more visible to me in some ways was the Lenza app, which was an AI image generation app that once again took photos of people from their Instagram and Facebook profiles and turned them into highly idealized, imaginative AI versions of themselves. Indeed, I still have tons of people on Facebook who have never changed their Lenza AI profile pictures ever since they added them back last year. In a lot of ways, I actually think that for many people, the combination of ChatGPT as this fundamentally new thing, something that we had never seen before, in terms of an AI that could actually talk back at us and write things coherently and interestingly and give us information in a totally new way, that combined with something that was much more visual and visceral in the form of the Lenza AI app, was this potent cocktail, a one-two punch, that announced artificial intelligence in a huge way.

12:30One breakout app, even a really interesting one, is not a trend. But two things blowing up and going viral at the same time, across multiple dimensions of this new thing called generative AI, and that was something that people felt like they had to pay attention to. Now of course, with any image generation tool, and particularly one that's going to take images of real people, it was probably inevitable that there would be some amount of controversy. Lenza had no shortage of it. CNN Style wrote about the fact that for some people, women in particular, the app seemed to produce overly sexualized images.

12:59Another complaint about the app, which is of course part of a much broader trend that we talk about here all the time, was that digital artists said that it was a ripoff of their work. Now, Lenza was powered by Stable Diffusion, and so it was really the way that Stable Diffusion was trained that was the source of artists' ire. Obviously, as we know, this is a very unresolved issue even up to now. Lenza also had quite an interesting history as a company. In January of this year, NPR wrote a piece called, A Rocky Past Haunts the Mysterious Company Behind the Lenza AI Photo App. The article begins, a Belarusian millionaire living in Cyprus, a dinner with the CEO of Snap, a six-figure patent troll case.

13:32These are all a part of the history of Prisma Labs, a largely obscure artificial intelligence startup that spent years under the radar until November when the company introduced magic avatars. Now, one really interesting piece of data from that, outside of any of the interesting intrigue around the company's history, apparently in November, when the app was at peak hype, the company netted more than$70 million from the app. Now, of course, when entrepreneurs see apps going viral, they think about how they could do a different version of that same thing. And a couple months later, in the middle of this year, we saw another viral AI avatar trend in the form of Remini.

14:03Now, at first, what made Remini go viral was people particularly on TikTok using it to turn their selfies into professional headshots. Users wondered why if they had access to this technology, they would ever pay for professional photos ever again. Like Epic has in the last couple days, the Remini app stored all the way up to the number one spot on Apple's App Store charts. Now once again, this was a multiple years to overnight success sort of story. Romini had launched initially in 2019, and hadn't added generative AI to the app until 2022. And of course, it wasn't until this specific use case of these professional headshots that it actually shot up to top the charts.

14:36And once again, there was controversy. In July, Insider published a piece called TikTokers are flocking to the viral photo app Romini, but some say it's editing their bodies beyond recognition. It was not just that Romini was occasionally adding extra fingers, but that in some cases, it seemed to be making people unrecognizably skinnier. One user captioned a TikTok video, when you can't use your AI headshots because it took off 80 pounds and everyone would notice. Now this particular person just viewed it with humor and even said it gave them some fitness inspiration, but it's obviously brought up much bigger questions of larger societal norms, of questions of body ideals that are now intersecting directly with technology.

15:11Said one user, for women especially, body dysmorphia and poor self-esteem is at an all-time high due to social media. I hope that women and men who have used these apps to generate these headshots can see past the over-edited fakeness and look back in the mirror and see a really rad and unique person. Now, zooming back up to the yearbook trend, there are a couple different dimensions of its particular brand of controversy. One actually has to do with the recent Hollywood strikes. Newsweek writes, With artificial intelligence and usage rights flagged as concerns during the recent WGA and SAG-AFTRA strikes, why are some celebrities taking part in the trend?

15:42Continuing, they write, Some actors pointed out on social media that by using the app to create the yearbook photos, Celebrities are signing away their likenesses and images for future use. This is called into question whether the simple-to-use app undermines the months of strikes. Now, it's getting a little bit beyond the scope of this show, but lawyers for SAG-AFTRA basically said that whatever the terms of service of this app say, they actually don't outweigh whatever negotiations are going on. That this random app's terms of service doesn't undermine what celebrities have negotiated with the studios.

16:09Still, it's a fascinating reflection on how these seemingly small things can be implicated in a part of much bigger conversations. Another big question is, of course, privacy. Can someone explain how this makes sense? Even though the app claims to delete the photos immediately, isn't a data stored for future recognition? And then there's the political dimension of this. Yesterday, in an unrelated story, two Democrat members of Congress sent a letter to the CEOs of Meta and X, basically saying that they needed to get on board with new policies around deepfakes, particularly in the wake of the upcoming elections.

16:37Said Senator Amy Klobuchar, They are two of the largest platforms, and voters deserve to know what guardrails are being put in place. We are simply asking them, can't you do this? Why aren't you doing this? It's clearly technologically possible. Now of course, the deepfakes here that the politicians are worried about are political deepfakes and misrepresentations of politicians who might be running for election, but the whole deepfake policy gets a lot more complicated when people are basically putting up fakes of themselves, when this has just become a normal part of the new social experience.

17:03The last dimension of this that I want to point out has to do with questions of AI adoption. Ultimately, all of these different apps that have gone viral over the last year haven't gone viral because of some novel technology. None of them are more powerful than Midjourney or Dali or anything like that. What they've done is rip out the complexity and design an entire experience around a single social-minded use case. What that has meant is that a huge number of people's first interactions with artificial intelligence are on the basis of an extremely simple social trend. Perhaps it is not surprising that it's a trend that puts them and their likeness at the very center of the experience.

17:36My strong guess is that in about a week, the yearbook trend will be almost entirely played out. But I think that the lessons that it has to teach us about the state of AI and the state of society remain. Anyways, interesting little conversation for the end of the week. I appreciate you guys listening or watching as always. And until next time, peace.