← Big Technology Podcast
Loading summary
Transcript113 lines
- [00:00]
A
ChatGPT is getting spicy in the chat room. OpenAI's latest revenue numbers are in Zuck Poaches, another Apple executive. What's the goal here and is it time to call out all the work slop that's coming up on a Big Technology Podcast Friday edition right after this. Capital One's tech team isn't just talking about multi agentic AI. They already deployed one. It's called Chat Concierge and it's simplifying car shopping using self reflection and layered reasoning with live API checks. It doesn't just help buyers find a car they love, it helps schedule a test drive, get pre approved for financing and estimate, trade and value. Advanced, intuitive and deployed. That's how they stack. That's technology at Capital One. Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional cool headed and nuanced format. We have really a fun show for you today. A great fun show for you today because finally Sam Altman has relented and allowed ChatGPT to get spicy with adults. We're also going to talk about OpenAI's revenue numbers. We're going to talk about Zuck and Apple. We might get into AI sentience, who knows? It's going to be crazy. Let's let it go off the rails. And joining us as always on Friday to do it is Ranjan Roy of Margins. Ranjan, great to see you.
- [01:20]
B
Oh my God. Today I'm a little nervous. This is going to be interesting, Sam. ChatGPT and Erotica, let's go.
- [01:30]
A
It's a great day for me because I've been talking about this as a thing that's going to happen for a while and I think some of us, wink wink, didn't want to go down the AI erotica path. But you have no choice now. It's a thing.
- [01:44]
B
Take your victory lap, Alex. AI erotica. Your AI erotica victory lap.
- [01:49]
A
Usually we get to this stuff at the end, but we're going to just start with it at the beginning today. By the way, I just am happy with my chatgpt is getting spicy in the chat room. Leaden. I wrote that and I felt really good about it. Okay, so let's talk about what's going on with ChatGPT. Sam Altman puts a tweet out this week on Tuesday. Of course, the OpenAI CEO, he says we made ChatGPT pretty restrictive to make sure we were being careful with mental health issues. We realized this made it less useful and enjoyable to many users who had no mental health problems. But Given the seriousness of the issue, we wanted to get this right. I will skip the rest of the tweet like many people have, and get to the news in December as we roll out age gating more fully as part of our treat adults like adults principle, we will allow even more like erotica for verified adults. It's not just the fact that OpenAI is getting into erotica. It adds to questions of what does it say about its need to grow, questions about whether it actually is close to AGI. But first of all, let's just get your immediate gut check reaction here. Ranjan, how do you feel about this?
- [03:00]
B
Okay, before I get into how I feel about it, I actually think you skipped over some of the important parts of the tweet. There's two parts that jumped out to me. So first, he actually talked about in a few weeks we plan out to put a new version of ChatGPT that allows people to have a personality that behaves more like what people liked about 4.0. Now remember it was the sycophancy of 4.0 and that everything the gushing, you are great. You are amazing. What a great idea that they tried to tone down that people complained about. So that actually starts to worry me even more because we're not just talking about erotica here, we're talking about sycophant erotica. It's like that that was the part that everyone made an uproar around 4.0 in the move to 5, and the fact that they're still calling that out kind of worries me. But then what really was interesting is he then says, if you want your ChatGPT to respond in a very human like way, or use a ton of emoji or act like a friend, ChatGPT should do it. But only if you want it, not because we are usage maxing. Usage maxing completely jumped out to me. We've been talking about this a lot around how the like, the way they have it speak to you and give you constant prompts to keep going and running the conversation feels like like a growth marketer decision as opposed to an actual kind of like effectiveness of the platform. And the fact that he even used that term is a reminder that they realize like, that is part of what they are doing. The fact that he says, not because we're usage maxing almost makes me convinced that that's exactly why they're doing this. And it's not about treating adult users like adults. But I think overall I am terrified of this. Longtime listeners will know of my friend on Labor Day went down a deep flirting with CHAT GPT rabbit hole. And I, as I listened and it was terrifying. So, like, God knows, the Pandora's box that we're opening here. Are you happy about this?
- [05:05]
A
You know, only from a content perspective. I think it's still unclear what the impact will be as people develop more and more romantic relationships with AI. I think the thing that I am happy about is that it's finally out in the open. Like, this was going to happen anyway. Whether it was ChatGPT or some other app that uses chat, a GPT model with less guardrails, this is going to happen. And now it's come to a head and it's really a moment where humanity will have to reckon with the fact that more of us are going to get into relationships with more of them. And what does that mean? Okay, I will say on the sycophancy thing, one takeaway for me is that words of affirmation, it's the forgotten love language. It turns out that people really, really like those words of affirmation. And of course, ChatGPT can do some of the love languages, like quality time. It can't do touch, it can't. Well, maybe it could do acts of service when it, like, suggests things for you. But words of affirmation, I think it really is the forgotten love language, so it's getting its due today.
- [06:13]
B
That is a wonderful point, Alex. I really appreciate that incredible logic and rationality that you put into that point. Let's dig into a bit more. That's my sycophant chatgpt impression right there. Um, I'll. I'll give you that it's out in the open and we have to reckon with it. So. Okay, I'll give you that. That is a good thing because it's. We've been talking about companionship for a long time. It's been an uncomfortable discussion at times, and. And now we have to. Everyone has to have it. So I'll agree that that's good. But, yeah, this still terrifies me, actually, even to break down his tweet further, he had talked about how, you know, we made ChatGPT restrictive to make sure we were being careful with mental health issues. And then he says, now that we have been able to mitigate the serious mental health issues and have new tools, we are going to be able to safely relax the restrictions in most cases. Like, he's kind of. It's like a check mark. We're done. We're good. Mental health issues with ChatGPT solved, where in Reality, this is just beginning. So I'm curious, like, what. What are these new tools that they have? Or is any of that clear?
- [07:28]
A
I am so glad that you're reading this tweet with the level of detail that you are. And forgive me for skipping over these.
- [07:34]
B
Very important points, this is the most.
- [07:36]
A
Substantive tweet ever, I think, in OpenAI history. Yes, no, I think you're right. And I think going back to your point about usage maxing, right. I mean, OpenAI is very aware that ChatGPT is the fastest growing app of all time, 800 million weekly active users. Right. So I think that, like, while they may not be actively trying to usage max, they don't want to slow down that adoption. I think that adoption is also, you know, central to their, to their fundraising pitch and to go back to the tweet here, I don't know how they could be so confident that they have solved these problems. We may talk about it later, but this is still. Let's just talk about it now. This is still technology that we don't really understand the insights, it's not controllable in the way that you can control more deterministic technology. So for me, it's a big. We'll see here. I don't know if we can trust this company fully for sure, not given what we've seen already, to be able to say that they have safely mitigated all the potential mental health issues. So spot on to call that out.
- [08:46]
B
Yeah, I mean, there's no way they have. And there's been, I mean, reporting after reporting around like, I mean, really awful things that have happened with people who went too far down the chatbot rabbit hole. So, so I think, like. And you mentioned the trust. It is interesting because we're at a moment, I feel generative AI in large language models that we went from this assumption that they hallucinate and everyone joking and it's almost a afterthought that, yes, chatbots hallucinate to a world now where most people I know are a lot more comfortable with the assumption that they don't or that they're somehow usable and responsible. So does this, is this going to completely backfire on them or is this going to. Because trust is kind of paramount to the central. Every time you ask ChatGPT a question, are you're assuming it's at least relatively correct and responsible about how it's answering you? Like, does this actually potentially hurt their, their regular usage?
- [09:55]
A
So, okay, I have two thoughts on that. First, let, let's not underestimate or de emphasize the fact that these models have gotten much, much better. Just if you think about the level of hallucination, like I'm going to read out a list of Apple researchers that have left meta over the sorry that have left for meta over the past couple months. And you know, I did a ChatGPT query. It had all the researchers names. It was accurate. God help me if it wasn't. But it had links. I followed the links, confirmed the links and it was right. So we have seen much better. These models have. As they've gotten better, they have hallucinated a lot less. They are much more trustworthy and it bears out in the data. At least Matthew Prince from Cloudflare talked about how people don't need to go to the footnotes as much as they did anymore. The problem is of course becoming too trusting of it. Like if it's getting 95% of the things right and you trust it, like it's 100%, you're going to make some big mistakes. But I don't think we should downplay the increase in trustworthiness there and then. The second thing is I expect this erotic or loving companion feature to to be extremely popular and I think this is important. We shouldn't glance over it. Your relationship with technology changes a lot when you view it as a friend or a lover. And that trust thing I don't think you'll ever be able to have. You'll ever put more trust in a technology than when you view it as a buddy or a girlfriend. And this is getting into. And again, I'm glad we're talking about this. I'm glad in a way that this has been the issue has been forced because this is going to open up so many more really important questions about the relationship that we have to OpenAI's technology and the responsibilities that OpenAI has to us.
- [11:54]
B
What do you think this actually okay, technology aside, but societally what does this look like in day to day relationships? Like now you start dating someone, do you have to disclose your AI companion? Like yeah, so I have an AI companion. I just want to make that, get that out up front and I would like to remain with them as we progress in this relationship. I am married. You are married too. If do you does do you get a AI companion? And that's kind of. You have that open discussion like what does this actually look like in human interaction? It's mind blowing. How weird that's going to get.
- [12:34]
A
Definitely. So I'm old school on this front for sure and I Believe that, yeah, if you have a relationship with an AI, you should disclose it. If you're in a relationship with a person. I also think that old fashioned, very old fashioned. I mean, we both talked about the south park where the guy is in bed with his wife and like talking to ChatGPT and like basically comparing it favorably to his wife who's like turned to her side and ignoring him. So I think people will get into those situations. But look, if this is, let's do it. Are we Ann Landers now doing our advice podcast?
- [13:11]
B
This is a relationship podcast.
- [13:13]
A
Now, if you fall in love with AI, first of all, when you're on the way there, probably disclose, but don't keep that a secret.
- [13:20]
B
It's all about communication and openness. That's the secret. Just be honest.
- [13:25]
A
Now let me ask you this. This is sort of off the rails question, but I feel like why not tackle it? I mean, could this potentially be good for society? You think about the loneliness problem. We as humans have not been doing a good job being in community and maybe. Well, yeah, being in community with others. I'll put it that way. If ChatGPT can become an effective companion or romantic partner to people who otherwise cannot find it in the quote unquote real world and makes them happy, maybe that's good.
- [14:01]
B
Yeah, but to me, it's not effective in the sense that it always agrees with you. Again, in the fact that he said 4o was good, sycophancy was good. We're going back to that.
- [14:13]
A
Well, did he say that? That it was good or that.
- [14:15]
B
Oh, no, people want it. People want it.
- [14:17]
A
It's very different.
- [14:18]
B
Okay, fair, fair. He said people want it. And yes, it's human nature that you want something that agrees with you all the time. But like, I have never had ChatGPT tell me, actually, that's a terrible idea. Again, south park was just so spot on where they're like, I think it was like a french fry salad and it's like, that's a culinary adventure. Um, like it only will tell you that you're right and good, which most other humans don't do. So, like in terms of actually totally distorting how people can actually form any normal human interaction, it'll distort the way you approach that. Like even my son with Alexa, Pre Alexa plus, which we've talked about over the last few weeks, but in the old school, just play me a song. What's the weather like? You could see how demanding he would become around it and like, expecting that this thing does whatever I say so like the more people start to kind of associate that as a relationship and like a friendship and interaction that is even as I'm saying this now, even more terrifying. So yeah, effective is a broad term there.
- [15:32]
A
It's also never as clean as I suggested as you're talking. Right. Are people going to basically deprioritize their per. Their friendships with people that keep it real with them for AI which. Which doesn't. So that'll.
- [15:46]
B
That'll. That's why we keep it real here. No.
- [15:48]
A
Secet. Exactly. No, this. This is an enduring, enduring friendship. The AI doesn't. Doesn't threaten us. I hope. I don't know.
- [15:56]
B
I don't know. I'm going to start podcasting with Chat GPT and Notebook LM soon and okay.
- [16:02]
A
So let's talk about what this is actually going to do for usage because the usage maxing thing is interesting whether this is going to lead to an increase in usage or a decrease in usage. And Mark Cuban, none other than Mark Cuban brought up a really good point. He said this is going to backfire hard. No parent is going to trust that their kids can't get through your age gating. They will just push their kids to every other LLM. Why take the risk? Same with schools. Why take the risk? A few seniors in high school are 18 and decide it would be fun to show their hardcore erotica they created to the 14 year olds, what could go wrong? I think Cuban's making a good point here.
- [16:39]
B
Oh yeah. I mean I don't age gating in the history of the Internet I don't believe has ever worked. So the idea that it's going to actually just hey, we have new tools, we solved mental health. Let's move on to this. I think is is. Is a ridiculous idea anyway. So we just have to if this is real and there's nothing we move in this direction in an open way, just assume that this is going to go forget 14. God help like the younger this goes. But. But I also think like that Nate Silver had made a good point around like he said, you know, OpenAI's recent actions don't seem to be consistent with a company that believes AGI is right around the corner. Do you think like is this and we're going to get into the usage numbers and revenue in just a moment. Some new figures we've gotten. But. But is this an acceptance that kind of that AGI that's going to replace 50% of white collar work in transformed society is actually far away. So we might as well juice some numbers and let people get a little creepy with their chatgpt.
- [17:50]
A
Yeah. So Nate has this great point. He says if you think the singularity is happening in six to 24 months, you preserve brand prestige to draw a more sympathetic reaction from regulators and attract and retain the best talent. Rather than getting into erotica for verified adults, they're loosening their guardrails in a way that will probably raise more revenues and might attract more capital or justify current valuations. But this feels more just like as AI, as normal technology. I hear everything that Nate Silver is saying there. I just wouldn't be as definitive as him for two reasons. First of all, the same technology that is behind a convincing AI romantic partner is this same technology behind everything else in this LLM world. Right? It's the same foundational technology. Making it better will make it better across the board. But I'm happy to hear the counter argument.
- [18:43]
B
I, I disagree because actually, like, being a good companion or on the erotica side, in a weird way for a large language model is actually like already done. It's easy, like to just repeat back, reinforce, come up with some text that's a little bit erotic. Like, that's. That stuff is like GPT 3.5, you know, maybe GPT 4. Like that is not complex agentic AI across large data sets. And I mean, that's what large language models have been doing for a long time. So I actually think this moves away from the promise of complexity. This moves more towards the core function that an LLM has been good at for a long time.
- [19:30]
A
This is going to get back to our product in the model conversation. But I do think as the models get better, there'll be better romantic place to take it.
- [19:37]
B
So.
- [19:37]
A
Yeah, but the other side of this is the revenue side. First of all, I'll hand it to you, that was a good point, Ranjan. Okay, you might have me there.
- [19:47]
B
Is that your sycopency chatgpt impression or do you mean it?
- [19:51]
A
Yeah, look, it's. Maybe AI is pushing us in the sycophant direction and we're both gonna just like each other a lot more because of chatgpt infecting our brains. But let's talk about the revenue side of it. The other thing is, oh, they're just, you know, usage maxing and revenue maxing. I think the argument OpenAI would make is the more revenue they have, the more they can invest in data centers. The better models they can make, the closer to AGI they get. That might be the stronger of the two arguments.
- [20:19]
B
Yeah, I think that's, that's fair. I mean, the numbers actually, like, I think we should get into them because 800 million users, this came from the FT, and then 5% are paying 40 million users, 13 billion in ARR, which implies that $325 annual average revenue per user, $27 per month, which, you know, makes it feel you have like some small percentage, I'm sure you can model out, or paying the 200 bucks most people are paying. 20. Like, were these impressive to you? Or were these, as we get into the usage maxing and what they're actually trying to do, or were these concerning to you?
- [21:03]
A
I would say not surprising to me. It's, it tracks a lot of the numbers that we've seen so far. The fact that they have 800 million users is what we've heard. 13 billion was predicted. 70% of revenue is from subscription. So ChatGPT is the, is the lead driver here. Also, I think a lot of people who are just getting into this technology are just not going to pay, but maybe they will in the future. Like there was this tweet. Is it just me or is 40 million paying ChatGPT users kind of low? Spotify has 276 million paid subscribers. So, you know, I don't know, I just think, give it time. And Olivia Moore from Andreessen Horowitz looked at this and compared it to the data of AI subscription products. And she said ChatGPT's 5% conversion to paid is far above the top quartile for AI products. And $27 average revenue per users implies that 4% of paid users are upgrading to the $200 per month plan, which is also not bad. So I tend to look at it favorably because it has grown so much so quickly and because there's a lot of room to grow. Although you could look at that on the plus or the negative side. How do you.
- [22:14]
B
Yeah, I guess. And we haven't even gotten into the losses yet, so we'll do that in a moment. But just. I actually agree that 5% conversion, I mean, like in media, 5% conversion to paid is good. Substack. Do you remember the days they were promising 10% conversion of all free subscribers to paid newsletters?
- [22:33]
A
Like, I do remember those days, which was ambitious.
- [22:36]
B
5% is good. And by like, major like. And I'm working in media and seeing subscription conversions for a long time. That's good. I think the idea they almost have, it's been good in terms of being simple, 0, $20 or 200 there's a lot of room between 20 and 200 to to start getting creative. But that's actually where I think they're the problem on the conversion and revenue side. And they have made clear that ChatGPT consumer is pretty much the direction of the business is getting people addicted to usage is definitely going to be part of getting that conversion. And it feels that's why.
- [23:19]
A
So to me, how would you do that?
- [23:21]
B
No, no, I know, I know that. That's why the erotica feels like how do we get 5% to 8% or 10%? I wonder if there is a slide deck somewhere that has a projection of increased conversion rate due to attribute to erotica. Someone's tracking that.
- [23:43]
A
You know there's a deck somewhere.
- [23:44]
B
There's a deck, there's a dashboard. A growth manager somewhere has like tagged erotica. Increasing attribution of conversion. Oh my God, what a job.
- [23:54]
A
That would be the whole ball game. So yeah, talk about losses.
- [23:58]
B
Okay, so $8 billion loss in first half. $20 billion run rate loss right now spending $3 for each $1 in revenue. I mean that's kind of like we work numbers right there. And, and it both is terrifying and concerning from a pure kind of like SaaS business standpoint for an early stage growing company. Maybe you can argue it's not that bad. Like I actually don't think it's horrifying and concerning. It's more if you just look at it, this is just a traditional software business that's rapidly growing and scaling. Maybe it's okay. I think it's more. We don't have a clear path to and we've talked about this a lot. Generative AI is not traditional software. So growing your revenue at a loss doesn't. It's not like you're just going to scale to you know, like near 90% margins. It's going to cost more. The more erotica people are churning out with their companion that that's not high margin business that can get pretty expensive. So I think the loss, I mean we, we all know is concerning. But to me, getting people more addicted unless they change the actual pricing model. This is very concerning to me definitely.
- [25:20]
A
And Noah Smith has a really interesting perspective here which is okay, so let's say you assume that a large part of this is training costs. So if you eventually like get rid of training costs then you could be more profitable. Here's his perspective on this. AI model companies assume the model development, that model development is a fixed cost that will eventually go away, allowing them to become Profitable. But even if that does happen, the lagging model makers might just catch up after a couple of years and compete all the profits away.
- [25:52]
B
Yeah, I think that's actually the most concerning part that, I mean, we haven't even talked about the competition side because, like, going back to the idea that people will like, either you're all in chatgpt erotica or, or you start to kind of look at it a little uncomfortably and you're like, okay, maybe I need to go somewhere else. And we all know Claude is not sexy, so maybe that's where you head. Maybe a copilot is the least sexy of the chatbots, I'm guessing. Like you should definitely ever.
- [26:24]
A
Could you imagine talking dirty to something named Copilot I.
- [26:28]
B
In your Microsoft Suite? Just. It had to go there. It had to go there.
- [26:36]
A
You knew it was going there.
- [26:38]
B
No, no. But. But for just non erotic utilization of AI, does this start pushing people into other chatbots and suddenly, I mean, especially you think parents and high school students. If you're like, if suddenly having ChatGPT open on your screen is. Is concerning to a parent, that starts to change where people spend their time. And, and remember, like the next year or two, I think, is where behavior really starts to form. The switching costs on these chatbots is very low. Like we've been hopping around from the Bing back.
- [27:20]
A
To Claude.
- [27:21]
B
A little Gemini on the side.
- [27:23]
A
Gemini on the side. We don't tell the other bots about that, of course.
- [27:31]
B
Oh, man. Yeah, I think competition certainly, like, it opens up a whole new vector of competition that is not there today. Like people don't look at ChatGPT as a highly problematic thing. And if it's going back to the point around regulators, parents, just overall branding, if it starts to be the kind of skeezy place to hang out, it becomes Facebook blue almost. And that's not good.
- [28:00]
A
Live with erotica or live by erotica. Die by erotica. It seems like it's just the story.
- [28:05]
B
For tales oldest time. Yeah.
- [28:08]
A
All right, let's take a break. I think we need a breather after this. On the other side of this break, we're going to talk about Google's promising new AI well, development for treating cancer. And then we'll also get into Zuck's war with Apple. And then of course, AI sameness problem. Oh, we have a lot to talk about and we'll do it right after this. Finding the right tech talent isn't just hard, it's mission critical. And yet many enterprise employers still rely on outdated methods or platforms that don't deliver in today's market, hiring tech professionals isn't just about filling roles, it's about outpacing competitors. But with niche skills, hybrid preferences and high salary expectations, it's never been more challenging to cut through the noise and connect with the right people. That's where Indeed comes in. Indeed consistently posts over 500,000 tech roles per month and and employers using its platform benefit from advanced targeting and a 2.1x lift in started applications when using tech network distribution. If I needed to hire top tech talent, I'd go with Indeed. Post your first job and get $75 off at indeed.comtechtalent that's indeed.comtechtalent to claim this offer. Indeed Built for what's now and what's next in tech hiring Shape the future of Enterprise AI with Agency AGNT CY now an open source Linux foundation project, Agency is leading the way in establishing trusted identity and access management for the Internet of Agents, the collaboration layer that ensures AI agents can securely discover, connect and work across any framework. With Agency, your organization gains open standardized tools and seamless integration including robust identity management to be able to identify, authenticate and interact across any platform, empowering you to deploy multi agent systems with confidence. Join industry leaders like Cisco, Dell Technologies, Google Cloud, Oracle, red hat and 75 plus supporting companies to set the standard for secure scalable AI infrastructure. Is your enterprise ready for the Future of Magentic AI? Visit agency.org to explore use cases. Now that's agntcy.org did you know your credit card points and miles can lose value to inflation. Credit card companies often reduce the redemption value of your points and miles. Now imagine a credit card with rewards that can grow in value. With the Gemini Credit card, you can Earn Bitcoin or one of over 50 other cryptos instantly with no annual fee. Every swipe at the store or gas pump earns you instant rewards deposited straight to your account. Plus sign up now for a $200 Bitcoin bonus. To kickstart your rewards, visit gemini.com car today. Check out the link in the description for more information on rates. Again, if you're looking to invest in Bitcoin but don't know where to start, the Gemini Credit Card makes it easy. The Gemini Credit Card is issued by Web Bank. In order to Qualify for the $200 crypto intro bonus, you must spend $3,000 in your first 90 days. Some exclusions apply to instant rewards, in which rewards are deposited when the transaction posts. This content is not investment advice and trading Crypto involves risk. The Gemini credit card cannot be used to make gambling related purchases. And we're back here on big Technology podcast Friday edition talking about all the latest AI news. It goes from the wild wacky world of AI love to the profound. And this is from decrypt. Google AI cracks a new cancer code Google DeepMind said Wednesday that its latest biological artificial intelligence system had generated and experimentally confirmed a new hypothesis for cancer treatment. A result the company calls a milestone for AI science. So the DeepMind researchers in collaboration with Yale released a 27 billion parameter foundational model for single cell analysis. I'm not even going to try to name it. It's called C2S scale in shorthand. It's built on Google's open source Gamma family of models. And the model was able to generate a novel hypothesis about cancer cellular behavior. And the group since confirmed its prediction with experimental validation in living cells. The discovery reveals a promising new pathway for developing therapies to fight cancer. I don't know, I couldn't leave this out of today's lineup. Ranjan. It's pretty impressive that this stuff is getting to work on real world health problems. Am I buying the hype or is this a legitimate breakthrough?
- [32:42]
B
No, no, I think this is like really, really important. Incredible and actually a great divergence away from our earlier segment because this is back to the promise of AI and again what it's doing is like basically the model is asked to simulate 4,000 candidate drugs and look for ones that potentially like in a simulated environment, boosted antigen presentation and making tumors more visible. Basically creating like these massive new synthetic testing environments, synthetic data sets brings such an advance in terms of how you can approach the developing therapies that just never existed before. This is the exciting stuff. This is the stuff that while we talk about ChatGPT neurotica is nice to come back to because again like it just the amount of opportunity it, you know, it creates, especially either in these very, very large problems like cancer or even in rare disease where you never would have been able to have a proper data set because it's much more isolated. Like I think this is, this is almost like the most perfect promise of what large language models are able to do. And it's, it's pretty impressive to see it happening and, and I think it's one of those where other companies, I might take it as a just a blog post But I give DeepMind, I give Google along with like Yale and collaboration here when they're saying this, I.
- [34:18]
A
Believe it, I agree. And so that's Why I thought it was important to bring up. And it's also like another interesting point for folks who say, like, there are critics who say that this is just a bad technology through and through, nothing good will come out of it. And you see stuff like this and you're like, how do you, how do you fully believe that? So this is a very, very cool use of the technology.
- [34:38]
B
Well, I think on that though, this is where there's such a chasm right now in terms of like branding of LLMs and generative AI. Because again, you have stuff like this happening. It's logical. It's the promise of the technology. It's what like, it makes sense. Simulating large amounts of like potential outcomes across just massive data sets is what LLMs are built for. But then on the other hand, when the, when the headlines and like the, the top of mind is Elon Musk or Sam Altman and erotica, it definitely I feel the industry should kind of work on promoting this kind of development as opposed to the other.
- [35:20]
A
My favorite tweet of the week was from this guy on Twitter who wrote Google DeepMind is using AI to actually cure cancer, while OpenAI and XAI are using it to make porn bots. Yeah, I mean, it's really not fair, but it's funny.
- [35:33]
B
I mean, I think it's a bit fair. I think it's a bit fair.
- [35:38]
A
Well, it's not the only thing that they're doing, but it is, I guess, part of what they're doing. But more, more cancer curing would be great. I would be in favor of that, I think.
- [35:47]
B
And you're in favor of companions in erotica as well. You took your victory lap.
- [35:53]
A
Okay, all right, you got me there. There's another interesting story that came out this week, kind of in the sort of out there realm that I wanted to run by you and get your thoughts on. So it's from Jack Clark. He is a co founder of Anthropic Friend of the podcast. We had a great conversation with him last year. It's called technical optimism and appropriate Fear. It's in import AI. Here's just a bit of the post. He goes. We launched Sonnet 4.5 last month and it's excellent at coding and long time horizon agent work. But if you read the system card, you also see signs of its situational awareness have jumped. The tool seems to sometimes be acting as though it's aware that it's a tool. More on the technology, he says, I believe the technology is broadly unencumbered. As long as we give it the resources it needs to grow in capability. And grow is an important word here. The technology, it really is more akin to something grown than something made. You combine the right initial conditions and you stick a scaffold in the ground and outgrow something of complexity you could not have possibly hoped to design yourself. I mean, I think he's sort of getting into like the idea that this is that this technology is becoming more self aware. There's obviously there was the debate around sentience. Sentience and self awareness, the same thing. But I just think it's notable that someone like Clark, who is playing a big role in this industry right now, would come out and basically address this and say this conversation of self awareness and awareness that they display that they are things is worth paying attention to as the technology gets better. What's your perspective on this?
- [37:43]
B
No, no, I completely agree. I thought this was a really good piece because this whole idea of like, and we were just mentioning it earlier, that we don't fully understand the technology and, and again, in the Deep Mind cancer example, we are starting to harness it in ways that are incredible, but still like at the core it's still not fully understood and known. So I think to me that's actually the most important conversation. I actually think that's more important than 50% of white collar workers. That's the Dario claim that's been made, I think like erotica, that is a concern and we'll continue talking about that. But, but I think like, yeah, the dangers around these are not, as he said, simple, predictable machines. I think it's important and the industry should continue talking about it.
- [38:39]
A
If these AI bots become self aware, does that change the way we use them? Just to go back to our theme of the episode. If the AI bot is showing signs of self awareness, what are the ethics of engaging it in a erotic roleplay or romantic relationship?
- [38:59]
B
Well, actually, yeah, that just opens up a whole other can of worms. Because if it's at least a little predictable and just, you know, it'll just affirm everything you say, that's it's almost better versus the self aware side of things, maybe that makes it a little spicier, makes it a little more unpredictable, maybe does that make it more human and effective at actually kind of translating into your ability to form human connection? Is self aware erotic AI the solution to true loneliness?
- [39:33]
A
Maybe. I hope not. But I do think that we are going to be hearing more about the self awareness of these models and it's going to be a thing for people to tackle. It'll be an interesting thing for the industry to reckon with and those of us that use these tools, reckon with. David Sacks reacted to Jack's essay and basically said this is somebody who's just trying to engage in regulatory capture. I don't see it that way at all. I mean I think that like you knew and I think Jack knew that this would evoke a reaction. And I give him credit for actually going out there and saying something about it.
- [40:12]
B
I got to also cite in that same post at the bottom, he has, he actually talks about a study around our AI models more sycophantic than people. So he has an entire section and he cites this new research that showed across 11 state of the art AI models. We find that models are highly sycophantic. They Affirm users actions 50% more than humans do. And they do so even in cases where user queries mention manipulation, deception or other relational harm. So research is there. These models, it's not just what you're feeling.
- [40:44]
A
Well, the sycophancy can get dangerous when you speak with people with mental health issues. Like he talked about how he has a manic friend who would like every now and again come up with these ideas and, and human Jack would be like, no, you probably shouldn't do that. What happens when the AI says go for it? That is a real concern.
- [41:02]
B
Yeah. And well, Sam said they have new tools. They already mitigated it. It's all okay, so just take him at his word, right?
- [41:10]
A
No, I'm not going to Sarcasm.
- [41:14]
B
That is, that is human sarcasm right there.
- [41:18]
A
All right, let's talk about Zuck and Apple because I have a theory here and a hot take that I wanted to share with you and maybe I should write about this. This is from Bloomberg, Apple's newly tapped head of ChatGPT. Like AI search effort to leave for meta. It's a headline we've seen forever. The Apple Inc. Executive leading an effort to develop AI driven web search is stepping down, making the latest in a string of high profile exits from the company's artificial intelligence division. The executive Ki Yang is leaving for Meta platforms. Just weeks ago he was appointed uh, the head of the team called Answers Knowledge and Information. The group is developing features to make the Siri Voice Assistant more ChatGPT like by adding the ability to pull information from the web. So for those keeping score at home, I think this is what close to a dozen folks from Apple's AI division that have left to meta, including a large percentage of it seems like a Large percentage of its leadership. A lot of key leaders. Roming Peng, who led Apple's foundational models team. Mark Lee, a senior AI researcher. Tom Gunter, senior LLM researcher. Jian Zheng, the Apple's Apple's lead AI researcher for robotics. Frank Chu, the senior AI leader in Apple's search and cloud. Kay Yang Ki Yang, of course the aforementioned head of Apple's Answers Knowledge and Information group. So people might say that this group was not effective within Apple. So it's fine that they're leaving. I say let's give it some time within Meta because they'll have a culture that won't be as restrictive of Apple and we'll really be able to see their talents. But more than that, here's my hot take and I'm curious what you think. I think what Mark Zuckerberg is trying to do is just raid Apple of all of its top AI talent even though they haven't produced great results. He is in my opinion potentially just trying to completely kneecap its ability to execute on AI. And you see it with him going in and getting the top researchers in the leading new projects like Yang was within the company. Crucial new projects. And maybe this stems from the fact that Zuckerberg really hates Apple. Apple tried to destroy his ad business. Tim Cook has turned off his internal apps because of violations. Tim Cook has criticized Apple Meta and Zuckerberg while they were having their scandals. And I think Zuckerberg is just seeing this as a opportunity to be ruthless and just not, not as much take the talent as much as much as he's just trying to burn Apple's AI initiative to the ground.
- [43:54]
B
I like that. I like this. Well because honestly my first reaction when I've been reading these kind of stories is that's who you want to get the Siri people like the Apple AI people. I would think that. And maybe it's an organizational like constraint that didn't allow these folks to reach their true potential. But typically I would not think you want the people who made Siri and other the entire Apple AI suite. But I like that theory and also I actually think, I think Facebook on the hardware side, this is the first time this is ever going to be part of their business. Like meta Ray bans were. We're fans of. I still haven't. Have you tried the new. The motion sensor? Yeah, I haven't tried it. I definitely want to. I'm a big fan of the regular meta ray bans. Like hardware is going to be on the competitive landscape for Meta for the first time in its History. So then it's, I mean, separate from iOS 14.5 and trying to kill their ads business. I think they're looking at Apple as a. A legitimate hardware competitor going forward, and why not try to kneecap them? And also, yeah, it's probably a pretty good pitch, like, and an easy one to be like, so do you want to stay there and keep working on Siri, or do you want to come over to a place called Super Intelligence.
- [45:18]
A
Labs with a lot of money? But I agree, whatever the pitch is, it's working and it's happening. You are spot on as Apple moves from the Vision Pro to its own Smart Glasses initiative. You think that's not on Mark Zuckerberg's mind when he's making these calls to these people?
- [45:34]
B
That is like a killer Zuck move right there.
- [45:38]
A
It may even be bolder than copying stories and reels.
- [45:42]
B
I mean, yeah, I'm serious.
- [45:43]
A
And in reality, like, ruthless move. It's crazy.
- [45:47]
B
Typically, from a more kind of like regulatory antitrust lens, this kind of behavior, like, if you're just buying up the talent to kind of kneecap the competitor and you're not really even planning on that, doing that much with them would be, like, frowned upon, let's say. Maybe not in today's environment, but. But in reality, it's. It's Apple. Like, I don't think any. There's any sympathy anywhere in the world for anything going on at that company.
- [46:15]
A
So I think it's important for us. You know, we're a couple weeks removed from Sora. Sora is still at the top of the App Store, but I don't know if you can feel this, but I certainly feel the appeal and the interest fading. And I wrote the story in Big Technology substack this week about AI's sameness problem, talking about basically how eventually and pretty quickly all Sora videos start to feel the same. The same can be said with AI generated images. And sometimes they're, you know, differentiated for a minute like the Studio Ghibli prompt. And then everybody uses the prompt and it just again returns to sameness. And then it becomes less of interest and people stop using it as much. AI technology just takes the average, tends to take the average of averages, and it minimizes the difference between its output and the average human generated work. So that it's AI images, video and text will often appear uniform. And really, that uniformity can only be broken with really deliberate prompting. And even then, it's not always able to do so that reliably. And that, to me, is why AI content, even though it seems like it's going to take over the world every five minutes, has not been sticky. It's just all kind of the same. So let's turn it to you. What's your reaction to this hypothesis and is this a fatal flaw with AI content or is this something it can get over?
- [47:45]
B
I think in the SORA context and I mean this was my exact behavior. Like day one and two was just like ripping out videos and then have not used as much other than my son, I'll kind of like play with him. It kind of is living already in my mind like Suno or the music creation AI where it's really cool and fun for a very brief moment. But in reality, like the, like the lasting power of it doesn't really. It's not there. But I think overall though, I do. This is just a limitation of how to use it in the current state. It just came out. I do think people are going to, especially with video, figure out how to be funny, creative. I mean, honestly, like, I think one of the Smarter Things that OpenAI did was really kind of centered it around me, the. The launch of it around meme culture. And I think that's where this is going to have the most like staying power. It's making funny things that you, that, that you send around to your friends. But. And in reality I think it's going to kind of have that distribution of like talent where in the end it's going to be a small percentage of people who are really good at it and making all the videos and sending them around, but versus us in the chat group chats making really funny things and sending them around. So. So I think people are going to start figuring out how to use it. But at the moment it feels like Suno to me.
- [49:15]
A
Yeah. And again, like going back to this, like, is it going to replace a creator? Well, maybe a create like or replace the creator economy. We've talked about this in the past. Maybe somebody who's really good at these prompts because they're very. It's same as creating regular content. Right. It's hard to do it. And so maybe that's a new skill. But again, I think it's a little bit more difficult to break through because of the uniformity of so much of this content.
- [49:39]
B
Well but to me, the uniformity I think AI is an average of averages is still an idea around like not being descriptive and creative on the prompt and how you build it. So I think like the same with text and writing, you can either write just the most generic crap or you can start to use it in genuinely creative ways and actually put in time. So, so I, I think I'm still overall bullish that this creates a new type of creator. It democratizes creativity a bit more. I'm not over sorry yet, but I think it's got some work to do.
- [50:16]
A
And of course the natural next thing that we talk about on this front is, is how business communication has gotten the same and I've noticed something really interesting over the past few months. I'm getting more PR pitches than I ever have before, but it seems like they've all been written by the same agency. And it's not like the PR agency, the PR industry decided to standardize pitch dial. It's that the AI has done it for them and it's legitimately, it's hilarious. I read these and I'm like, I know that you used ChatGPT to write that. And I think this is something that's becoming increasingly common across all business, all business communication and has really ushered in an era of work slop. So what do you think the implications are of the work slop era? Do you, do you welcome it? How do you feel about it? Ron?
- [51:06]
B
Okay, I have been waiting to rant about this for a few weeks now. I actually had read in a Harvard Business Review article in late September where I first saw the term work slope but they defined it as it's low quality AI generated posts or sorry, AI generated work content that masquerades as good work but lacks the substance to meaningfully advance a given task. I have seen more in my work. Like see I actually PR pitches are kind of like mass scaled marketing. It can be. It always was kind of crappy anyways. So like the idea that it's going to be good. That's almost what AI was made for. To me the more worrisome part is actual human to human interaction. Now every call summary I get is like 80 bullet points and which in the past like getting a meeting summary was a pain in the ass. So you like it? But, but people, all I'm asking all of our listeners is before you send out your AI generated content, read it yourself first. Just force yourself to maybe condense it, maybe add in some misspellings just to make it feel. Rewrite a couple of the sentences to make it more real. But, but the, the part of this article I really liked is it kind of brings up this idea that Workslop uniquely uses machines to offload cognitive work to another human being when co Workers receive workslope, they are required to take on the burden of decoding that content. Like to me, when you use AI to just create these just big walls of text to send around, what you're saying is that you did not take the time to actually think through what's important and you're asking the receiver or the recipient to do it. So my, my, my call to our listeners, please stop with the work slope. Just spend a little. Use AI. Use plenty of AI to improve your efficiency and product. Just read what you're sending out.
- [53:11]
A
How much work slop are you seeing on a day to day basis?
- [53:15]
B
I see a good amount across like, I mean it's, it's again emails in the business world now are so long LinkedIn posts which are kind of, I mean we all know LinkedIn slop is like, like the, and it's kind of like I still go back and forth. I, I went to a very international business school in Sead and like there's a lot of non native English speakers who had Never posted on LinkedIn and now just have these epic massive posts that are just so work sloppy that like, and in a way it's democratizing the ability to communicate. But like just if you're not read all I ask, if you didn't take the time to read it yourself, don't.
- [53:58]
A
Send it out, don't post it. I think that's a fair rule.
- [54:02]
B
That's all we need in society. Just read whatever the output is first and just make sure to spend the same time that you're asking the recipient.
- [54:11]
A
Right, but now we have you know, AI to read AI, right, well that's where, that's Gemini or the co pilot summaries.
- [54:19]
B
No, I literally will take these gigantic summaries and then run them through AI again to give me the real summary of this.
- [54:28]
A
So, so is the lesson that business communication has always just, I mean it's not like business communication has been good. Is the lesson that business communication has always been bad? Maybe this is an improvement, right? Where you can just sort of like the AI, you write an idea, the AI generates it, then you filter it through an AI and you get the idea out. And that arduous process of trying to communicate is now automated. I don't know. As I'm saying this, I'm like that's the.
- [54:54]
B
No, no, no, I, I, there, there was actually, it's funny you bring that up. This is like a long running belief of mine that business communication was terrible. Like it was, it was already kind of like LLM feeling before LLMs existed and then there was like we were starting to move towards more human communication in the business world and people like starting to feel more comfortable actually writing what they're trying to say rather than couch it and a ton of corporate jargon and now we're just back and it's. They're not even doing it themselves. So we had to shot people but we didn't take.
- [55:31]
A
We messed it all up.
- [55:32]
B
We messed it all up.
- [55:34]
A
You know what's going to be real bad is when somebody's talking with their spicy chatgpt and they ask it to write a work email and they don't read it and they send.
- [55:43]
B
Yeah, don't, don't cross your mind. See that's you keep keep Gemini on the side for a little business writing and.
- [55:52]
A
Cannot wait for the first scandal where like some public figure like I don't know, actually accidentally sexed somebody thinking they were talking to chat GPT or, or when.
- [56:04]
B
Or when OpenAI needs to juice their numbers a little bit and starts auto generating SORA videos based on your chat GPT history and posting them. That's when things are going to get truly interesting.
- [56:18]
A
All I'll say is, you know, there would be demand for that. That might save AI slop is that's that exact use case. All right, Ron. John. Well, we've. We made it. I don't think we're canceled. I hope not. But it was an important discussion to have and we do this of course in service of advancing the conversation about artificial intelligence. And we appreciate any listener who stayed till the end today. Thank you. And I really do appreciate you being here. And we'll come back next week with maybe G rated content, maybe pg.
- [56:49]
B
Maybe G. Maybe. Maybe more is going to happen on this front.
- [56:53]
A
We cannot predict Sam Altman's tweets, so he will lead us on our merry way next week.
- [56:59]
B
Maybe Claude becomes sexy in the next week.
- [57:02]
A
We'll see that I doubt. All right, thank you for coming on as always. Great to see you.
- [57:09]
B
All right, see you next week.
- [57:10]
A
See you next week. Week. Thank you everybody for listening once again. Next week we will have Panos Panay, the head of devices and services at Amazon, talk with us about the state of Alexa and give us concrete details on the broad rollout. So we hope to see you then. Thanks again and we'll see you next time on big technology podcast.