OpenAI Finally Ships Its Superapp, Meta’s AI Price War, ChatGPT Cheating At Brown

0:00 Opening eye super app finally arrives. Are all AI apps starting to look the same? Meta undercuts the Frontier Labs on pricing and Ivy League students use AI to cheat. That's coming up on a Big Technology Podcast Friday edition right after this. Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional coolheaded and nuance format. We have a great show for you today. OpenAI's long awaited super app is finally here and it's starting to look like all AI apps are going to look the same. So what does that mean? Meta comes out with a new model MSpark 1.1. It's almost as performant in some areas as the Frontier Labs, but in some areas it's 25% of the cost. So has the price war arrived? And finally, we will discuss the Brown students cheating with Chad CPT take-home test for the midterm.

0:53 Everybody gets a 100, it seems like in class at Brown University for the final and it is it's failure city out there. So, what does this mean for for our youth? Joining us as always on Fridays to do it is Ranjan Roy of Margins. Ranjan, great to see you. >> Good to see you, Alex. I'm glad you as an Indian person I'm glad you clarified it was Brown University and not the Brown students. But >> there were there were so many jokes about that on Twitter and and like people being like you know who tweeted this study like why do you got to bring race into it and I was like be careful >> Alex remember >> and of course I messed it up but yes Brown University let's move on until we return to that. all right. So, I'm not even there's no segue. All right.

1:42 Let's talk about what happened this week in tech. OpenAI finally unveiled its Claude Co-work competitor and its desktop super app. This is according to the information openai as part of its effort to attract more business customers announced a new agent called Chad GPT work which taps into corporate data to automate the creation of spreadsheets and presentation and can also handle more complex tasks like like financial forecasts and conducting research chat GPT work is open's answer to anthropic's popular cloud co-work product which the startup has used to expand the market for AI coding to nontechnical users open also unveiled a desktop super app that marries chat GPT with Codex and the new chat GPT work offering. This reflects OpenAI's recent realization that Codex is better than Chat GPT and handling long running tasks that involve multiple steps and require the use of external tools. U let's just start here Rajan. It seems like all I mean we've talked about this on the show a bunch but like now you can finally see it with your eyes. all AI products seem to be converging on this one use case which is that like you know you might have some chat but really AI is there to get things done for you. how are AI companies and we talked about this a bit with MG on Monday but how are AI companies going to differentiate themselves if they're all offering effectively very similar version of the same product? Well, th this is front and center in my life working at Writer and you know like delivering enterprise AI and what we've been selling for a year now natural language driven agentic workflow building which sounds very buzzwordy but is basically this is that kind of marrying of what codeex has done in the command line but in a more in a more chatbased interface. So have definitely been thinking about this a lot. Played a bit with chat GPT work was definitely underwhelmed. It felt a bit rushed in terms of like how claude co-work was kind of slowly built out and now has been integrated into the main claude interface is like a separate button which is another topic whether it's getting too convoluted and messy.

3:53 But I it's funny like I go to a lot of conferences and now every booth is starting to look the same. I've been in meetings where someone's like, "What do you do? Don't say enterprise agentic AI." And you're like, "Well, it is enterprise agentic AI, but like it's making it easier for business users to actually build agents." Like it's actually crazy how cursor just released something that's actually more in this direction. Notion has built an entire ecosystem around this very same thing. I think like the exciting part of it is that I feel validated. I started talking about this last October. It's this is every product going forward. And I think where I think the differentiation is going to be what we focus on is sales and marketing across enterprises. But like the actual like context layer, intelligence layer, how you actually build out the systems that help build that within an organization. That's where it differentiates. differentiation is going to be and I actually think it's good for the more verticalized companies and it's bad for anthropic and open AI.

5:05 >> Wait, sorry, explain that and and do it without making a rider commercial. >> No, no, no, but but dude, this is my life right here. This is like all I think about and read every day and we have to deal with >> Take Ryder out take Ryder out of it. >> No, but explain what the differentiation is going to be. >> Yeah. Yeah. So I mean it's difficult when this is >> I'm challenging you here.

5:27 >> All right. All right. Let's go. Let's go. So if you are a marketing organization within an enterprise there are different levels of like the foundation beyond brand. When you say brand everyone thinks like oh just put a tone of voice document or something like building out skills that actually and or a foundation of knowledge. So when people go to build agents, they're actually doing good work or correct work using this stuff out of the box.

5:58 Everyone, no enterprise actually sees any kind of value realization like they don't things don't work well. So like actually building the system. So if you okay if you take you're creating an event and like if the big next big technology summit like making sure that the eventbrite or what system do you use for it? Luma >> the Luma like making sure the connectors all work really well and out of the box for customers connecting to various CRM systems so you can actually remarket having your messaging and and like whatever kind of guidelines around how you like to communicate already built in. So like those are the layers where there's going to be a lot more competition and we can get into price wars and the the cost side of the equation, but actually making it so again you're a large consumer goods company like how your product knowledge is ingested into the overall system makes every agent either work or not work. And does that make sense? I'm curious.

7:03 >> Okay. Yeah. But let me push back a little bit because all right first of all like I was able to use a lot of cloud co-work cla code and chat GPT to set up the event. So for instance claude code out of the box made the website. Now I gave it some references of what I wanted it to look like but it built a great website. It embedded Luma for me. Right. So it did that.

7:26 >> Yeah. And then hold I'm not I want to finish. So then Claude Co helped make like the prospectus for the for the event that we sent to potential speakers and sponsors and of course I I refined what was in there and helped give it reference material but it did a good job there. Chetchup BT designed you know with me of course you know all the graphics for the back of the stage right this was like a generative AI production. We went as hard as possible into Genai as possible and then the scheduling for instance you know was done in a single threaded chat with within chat GPT where like I would say okay we have this change and it would move the scheduling. So we didn't need any bespoke software there. We just basically needed like some references for the AI to go off of and then it was able to take it from there. So now imagine you have 221,000 different websites that you manage. You have 17 different regulatory agencies that you actually are responsible to. You have six different large brands that each have their own like entire architecture.

8:34 Now try to do that in the system that you just did. That's to me where the battleground's going to be. And that's like every action you took right there is it like that's like the beginning of how this all works. But that does not work at large organizations in that way. >> So let's go back to what David Karp from no Alex Karp from Palanteer was saying. David Karp is the Tumblr guy. Alex Karp from Palanteer was saying now >> nobody knows where he went. He just sold the blog to Yahoo for a billy and then he disappeared.

9:07 >> All right, let's talk about the relevant carp. on good relevant >> Alex carp >> relevant Karp. So basically what he was saying is you know he was he was at CNBC talking about how he was going to build this open model based you know based off of Nvidia that you would effectively use with his Palunteer consultants to put the AI into action right so is that kind of what you're saying Ranjan is you need a version of what Karp is saying and and is this maybe you know we've kind of joked about how open AI and Ananthropic are bringing the consultants in with their you know so-called forward deployed engineers which by the way is also a Palunteer thing. and so is you know is where we're going the customization of these tools for the purposes that you spoke spoke about with the assistance of some of these decidedly non-military but military sounding forward deployed engineers. You tell me.

10:10 Yeah, it's apparently I read once Alex Karp said he got the term for how French restaurants have the waiters work in the kitchen to like truly understand the the dishes before they go out and sell them. Even though I know that's what he said is so good at marketing. >> He's like legitimately reading what a consultant is and he's told you that his consultants are like fine French dining. >> I know. No, no. And it was like a military term. He's so good. Nothing against consultants, but let's why do we have to like we're we're we're looking for an apartment right now. And it's like some industries like the wine industry, the real estate industry, and I guess Alex Karp have the most audacious verbal flourishes to describe very simple things like you can't just have a standard two-bedroom.

11:03 It has to be a gloriously situated, you know, mini palace with, you know, two layers to rest your head. This is what what Karp is doing. >> Well, I mean, I do consider myself a bit of an AI sumelier, so you know. >> Okay. So, >> ending the show now. Ending the show. >> That's it. That's the next wave of job in this in this economy. 50% of white collar jobs wiped out but AI sumas are going to be the future. So for deployed engineer so that actually is exactly it but that is the high lift inperson layer of this and that's how things have worked to date. What I'm saying is and I'm seeing this and I'll avoid the writer commercial part of it but like this is across competitors as well. the Sierras of the world or like anyone who has like a very clear Harveys of the world like they're building their platform but is being differentiated. I mean, we say verticalize, but to me, it's less about verticalization and it's more about expertise within a domain and being able to actually integrate that into the overall platform. And that's what's going to happen. And it's going to be part FDE, AI, Sumelier, whatever you want to call them, people going.

12:22 No, I'm sticking with that one. I'm sticking with that one right now. It's going to be part that, but it's also going to be reflected in products more and more. That's where I see this going and that's where I see the differentiation and I think like claude for science already is kind of that. So like even the labs have their their versions of it but like that's where the next battleground is going to be and I called this battleground so I'm telling you where the next one's going to be.

12:50 >> Okay. So more verticalized style applications of these catchall applications is where the differentiation is going to happen. >> And I for some reason I don't like the word verticalized or verticalization cuz like in traditional SAS it was such a specific thing and now it's going to be more about the type of work being done rather than the type of company it is. >> >> explain that. So like if you're like a sales and marketing person, you'll use one type of tool and if you're like a researcher, you'll use another.

13:24 >> Yeah. Yeah. Which I don't know this is I'm still thinking through this one, but like the way people's minds go to when it's like Salesforce. It it wasn't a vertical it wasn't like for all sales and marketing. it was for CRM and then they had to buy more companies and try to like put together this larger offering versus now the AI is built for that person and that function rather than the like type of company it is.

13:52 Does that make I'm still I'm still workshopping this one but I don't like >> Okay. So so I want to go back to our our sort of meta debate that we've had on the show for like four years at this point which is >> >> is it the product or is it the model? I almost decided to make the whole day about product versus model, but I decided to put the gas on that, but I'll put the brake on that. But now I'm coming back to it. Okay. why? So, so I think what you're you're you're sitting in one of these whether it's not verticalized or specialized type of companies, right? You're sitting there and you have that point of view. If I'm the lab, what I'm going to tell you is the consultants are a bridge. These specialized products are a bridge. I know >> what's going to happen is, you know, I'm speaking again like if I'm in the seat at the big AI model companies, my model is going to turn into AGI and at that point the model will have enough intelligence that these bridges and we've called it scaffolding in the past. all this stuff is going to matter much less because the pure intelligence will be able to take on this this specialized work with much less handholding and prompting. Let's go back to your example about the 117,000 websites or whatever it was. All right.

15:12 >> Yeah. >> You you're right now it takes a lot of effort to feed that. But maybe in the future when these models take their next leap, it will just be let's say a day of saying to your model like let's say I'm going back to you know my summit right go ahead and crawl everything we do get our voice get our branding and then you know you come up with a plan and go ahead and execute it and you could just do that at scale because the model has that much more intelligence.

15:39 >> Do you know why >> why doesn't it go that way? >> I know. Do you know why I love this? the debate has not been settled because that is the it still remains the entire debate. It's like it is it's product versus model cuz like I'm sure within the labs the assumption is burn money to invest in more capable models and they will subsume all scaffolding I like as like architecture might sound too buzzwordy scaffolding it's like the things that are required to make agents work today I'm sure they all think one year from now two years from now none of that will matter and it's a it's an interesting one because in reality like I mean the progress that's been made in the last two years the things you had to do two years ago all went away. So like you know like I remember parsing a PDF the simplest thing two years ago you had to like define the tool and like potentially upload a tool like a Python script or something like that versus now like all that stuff is just intelligent like so the the debate lives on and it's going to be the central I think to the next one to two years of who wins and who loses.

16:52 >> Okay. So now let me take us let's say three years into the future assuming one version not the necessarily necessarily the version that will happen but one version of this future where that vision does play out right so you have your AGI and it's in codeex and it's or the new chat GPT app and it's in anthropics app and meta's figured it out some way and Apple has it in the iPhone for whatever reason. no I should be I should be nicer to them. They're making progress. Thank you for I appreciate there. I still haven't downloaded the developer iOS 27, but I've been meaning to. Yeah.

17:27 >> So, okay. So, we get to that point where where like what's the value? Because it's going to be four companies that are going to be doing the same thing and we're already starting to see signs that the premium is going to be on lowering costs and does everything just eventually go to zero. For example, Sam Alman talking about GBT 5.6 this week said it was 54% more token efficient than others, right? Than than than its previous model. So like the if you're a client and you're thinking about your ROI, well the I is going to matter a lot the investment of the return on investment calculation and you can get a higher ROI if the investment is lower and if you have all these products doing the same thing or similar versions of the same thing. it seems like there's a chance that even though you're you're providing an extremely valuable service, it's a race to the bottom.

18:23 >> Pricing. >> I'm still about the R, I think. okay. This is our next one. You're you're the eye guy. I'll I'll be the R. >> Bring on a third person for the O. Like, what's your role in this podcast? I just do on. Actually, I'm just >> Ron John does return. Alex does investment. I'm I'm here for on. >> Why you here? Well, someone had to do on. someone had to I mean it is required.

18:48 so so I think I think cost becomes important. I don't know like on this cost and we're going to get into Zuck's comments like it is almost comical to me at points how dramatically the conversation now shifts. It was one Uber quote like even I mean telling you like talking to seuite people six months ago no one brought up tokconomics and cost and now it's on everyone's mind. Meta's coming in hard as like making and I listened to the Bos episode around like you know like they don't like they have more capital they have more cash they have more profit. So like actually investing in these things, they have more strength versus some of the other frontier labs in this case. but still like we're not there yet cuz people don't have this stuff working at scale outside of software like it it hasn't.

19:47 It's there's bits and pieces and promise but I can tell you definitively like this we are so early on this that like to even think or worry about cost optimization before you've actually figured out how to make it work well I think is like this is more like everyone in the small groups within these companies is realize that cost will be a factor and it wasn't before but I feel the pendulum again just keeps swinging in too far each way. Just just let us work people. Just let us work.

20:21 >> Okay. There's two separate things here, right? First is is this cost discussion overblown? Maybe to some degree, right? But you also have because you were right that 2026 is the year of agents and I give that to you once again. Thank you. the agent workflow is that much more token intensive that people who previously were using generative AI and didn't really care, you know, what it cost cuz it wasn't costing a lot. You know, now we're seeing it 10x this year and are starting to worry. Like end of the year last year, Enthropic was at a $9 billion ARR despite the fact that it is a flawed measurement. Now I just saw maybe they're at a $69 billion ARR in five months. you know, they've 7xed or more, 8xed, right? So, that's why these costs are starting to become real to people. And that's sort of the driver of this of this discussion is there are people within companies who are like spending this and telling leadership it's justified. And that was an easier sell when it was 17th or 1/8 of the cost than when it is now. Like leadership is actually going back to them and being like, we need to see the productivity increase and the return. But you know that's one side of it. The other side of it is so that's a discussion we we'll continue to have. But to me the more pressing thing here is that whether that discussion is merited or not we are go we are literally going to be in the middle of a price war here between these companies and again it comes from the centralization of the AI product experience into this like co-work u cloud code type experience right the super app experience and the fact that there are some companies that are quite motivated to to drive the price down And you mentioned Bos. So let's go to Facebook. so Facebook this week they introduced their Muspark 1.1 model.

22:16 This is according to Bloomberg. and they are going to do something to strategically bring this market down. Here's what Mark Zuckerberg told Bloomberg. Since this is not an open- source model, I think this is the first time we're doing a real a serious API and the pricing is going to be very attractive and aggressive is the Bloomberg story. the the API will be used to collect fees from developers. API pricing is roughly 25% the cost advertised by other top models from OpenAI and Anthropic. I mean Zuckerberg has said there's some some good margins that the labs make some good margins here and so he's like well we have the compute we have a model it's almost as performant as everything else we are going to go a not not a half not a third quarter of the price of you know the models it's competing with now obviously it's not at the same level of intelligence but I just want to hear your perspective on what is the consequence here if all AI pricing starts to just crush you know, as this commoditization era kicks off.

23:25 >> Oh, I think it's a massive I'll use the term headwind liberally there where it's like I mean this changes the entire battle. It does and it already has like again becoming model interoperable whether like having the right model for the right task everyone is I and I I don't think that's unwarranted. I think everyone is rightfully thinking about that in any normal like development of a new economic cycle or whatever like new industry you would think okay this is all pretty normal it's you know we're new technology kind of the economics of it are being understood the technology itself the application it's going to cost a lot at first and then price is going to come down like that's all pretty standard I think what that I mean the two companies that affects the most are OpenAI and Anthropic and like and again that wouldn't be a problem if it wasn't their like actual cap tables and just the way they've raised money but it is like if everything is about the near-term and rush to IPO for them I think that's a big issue.

24:39 >> Yeah. I mean I think Zuck would personally be thrilled to play the spoiler here. And by the way, didn't I say last week that Zuckerberg was going to come out and complain about the concentration of power among the Frontier Labs. >> Oh, Zuck. Love it. >> And what he could do about it. And we, you said, you heard it here first. Within the next week or two, you're going to hear Zuckerberg make his attack. And he did. And this is the quote I was looking for. The price from some of the other labs is very extreme and has very high margins. We think that there's a real ability to offer frontier or very high level intelligence at a much more affordable cost. So knives out right and again this is because power has consolidated between within open AI and anthropic. you know the there was inevitably going to be a player who's going to come out and say well screw that let's even the playing field right and that by the way goes to my again this idea that this is this could commoditize and you know lo and behold Zuckerberg raised his hand you know the week after we said he would and he Leroy Jenkins his way right into the into the competition. I mean, when I think about people that must hate concentrated industries and power and high margin monopolized areas like platform advertising, I think of Mark Zuckerberg.

25:55 I mean, you know, he, >> you know, it's it's funny to make that joke. The reason why Zuckerberg has as much power as he does is because when he's seen a threat to his business, he has often masterfully thwarted it. Whether that is you know, copying stories from Snapchat, copying reals from Tik Tok, AI has yet to he's yet to be able to do that. Now, obviously, they have this quest to build personal super intelligence, but I think you know, with that taking longer than expected, as he said last week, you know, the other option is you know, just just run in there. Now, he had he had open source models, right, and they were free. so, he's charging for them now. So, is it really that different? But I think what what he's doing is very interesting. He's saying, "All right, I've spent all these billions and I'm going to at first use it to commoditize the model layer." Like Bos said, and Bos agreed with you, it's the product layer that matters. and so for for Meta, the idea that AI would be costly does not serve their purposes. they want it to be free so the product can be built on top of it whether it's theirs or others and in particular you know they they also want to be able to use this as Bos said as a negotiation leverage when they rent models from OpenAI anthropic and Google >> well I think for them it's almost it's like a two layers it's one like I think you had made the point that just like an existential threat that chat GPT competes with social media that could compete like people could be spending more time talking with their and chat GBT rather than scrolling Instagram.

27:42 That is a threat and like kneecapping the companies that are coming after you on that would be kind of like you know classic Zuck and kind of amazing. but then I think also it's like you said if personal super intelligence whatever that may mean but like something around being that like for everyday for the consumer for like helping you manage your life and everything not enterprise whatever. I think that is another area that the cheaper everything gets the better for them. And again they how many is it 4 billion people use their products a month? three bill whatever it I mean >> in that neighborhood >> everyone stopped counting I feel like it's basically the world >> well they saturated anybody on the planet with a phone >> yeah you you own distribution so like being the one to actually be the front door to all AI which is interesting like as I'm saying this like they really could be competing more head-to-head with Apple going forward I think like if Siri iOS 27 becomes a bit of personal super intelligence. Ambitious statement there, but like they could be competing a lot more directly soon.

29:02 >> Oh yeah, they they certainly will, right? And this, by the way, you know, in in the you know, in a in a roundabout way, you know, there's, you know, we've now given all the explanations for like why they're happy just to cut off the economic benefits from the other labs for other purposes. In a roundabout way, this might actually help their AI efforts in general. This is from the information. Now, what Meta is doing is no way to make money in AI, but that clearly isn't meta strategy. It knows that the cost of AI has become a paramount issue for many businesses. So much so that they're trying various methods to reduce it, including the use of open- source models or routing some AI work to older and more cost-effective models. In that environment, Meta presumably thinks that by undercutting everyone else on price, it can persuade users to at least try out its new model and potentially hook them. If that approach works, Meta could jack up the price later. So, it's like it's like the the you know, maybe it's using the old I don't know, the drug dealer model of economics where I give you a taste, you become hooked, and then I say, "All right, you know, if you want to keep using it, it's going to be a little bit more more pricey."

30:11 >> I mean, Anthropic and OpenAI both, I think, have definitely been anthropic more than anybody went that route. So, so yeah, I think everyone gets that like and again overall consumption of AI and compute will exponentially increase and it's going to make its way into more and more parts of our lives and the ways enterprises work. So I think like it is interesting everyone is trying to find where they fit best in that equation under the assumption which I do agree with that it will happen. So I guess it's good. Before everyone was just kind of riding on if the explosion happens, it's just great and like you will you will be valued at a trillion dollars or whatever. But now everyone's starting to actually try to like map out where do they fit in that future.

31:07 Yep. Okay. Okay, so before we go to break, you know, one of the underwriting themes or under underlying themes on this show is always has always been you know, we believe that there's real technology here, but we just wonder whether there's a business here. so given what we've discussed for the last 31 minutes, what do you think all this means for the business prospects of OpenAI and Anthropic? >> I mean, it I I don't see how it's good.

31:36 I I really and I know I like compete against them in some cases. So like but but I'm just being like like I can you map out in a world where models are interchangeable and interoperable and cheaper and cheaper. how that could be good. The where the only way I see it being good is if they still maintain that model that subsumes all the verticalized functional offerings or are so good that somehow you just stop using social media and then like that becomes your personalized super intell like I mean they're still making the bet.

32:19 That's why it's kind of almost like weird to me that when Sam Alman starts talking about 50% 4% tok more token efficient that's not their game. Their game has always been we are going to build AGI and then that's why we win and it feels pretty binary. So >> tell they've also said they want to build intelligence too cheap to meter. >> Wait, I thought he said it will be like electricity and metered. >> I guess they've had mixed messaging.

32:48 Okay. No, no, but but explain to me what would be your take on in a in a world where >> AI gets cheaper and models interoperable and application layer etc. How do they win? >> I don't know. I mean I think I think it this this is always been the problem here has always been you're building on something that it's very difficult to hoard and I don't know if there's an answer to that yet. So they've they've started to build products right and you know maybe get to go back to the laptop example from last week you know you're building something that's like a laptop can be used by personal folks folk can be used by people for personal reasons can be used by people for business reasons and so like your your you there can be multiple laptop makers but ultimately we see what happens they compete based off of cost and based on cost and you know Apple's you know doing well in its MacBook business and there are some others but you know it's it's not like worldbeating businesses. So that to me is is like the real question here in terms of where they go. But I mean the other side of it is you could say let's say we take your line right here that the product matters most. they both have built compelling products you know open AI which at GPT anthropic with cloud code and cloud co-work and those will continue to grow and they've been the engine behind their behind their growth and maybe that won't slow down even if the pricing power goes down a bit. So that's kind of the way I think about it. Totally unsettled.

34:30 >> Yeah, I mean it's I don't know the the next few months I'm just waiting to see an S1. I want to see the numbers, but it's definitely it's >> again all of this stuff in a one, two, threeear time horizon, it's just such a different conversation over the way this is going to play out the next 6 months. >> Yeah, that's the moment those S1's hit, >> it's not going to tell us anything because the battle like like what we're talking about right now, battle's just starting. It's not even we are just approaching the start line between because like it was four years.

35:06 >> It was four I'll just say it there were it was four years before these companies figured out the trajectory of what they could build and the form factor and now they finally centered around it. So now it's game on. Everything else was a windup. But do you think if numbers come out and they're atrocious and then that affects like stock price and then that affects like employee and researcher attrition and that like do do you think they are two separate things or do you think one could affect the other?

35:42 >> I don't know. I mean, of course, it can affect it can affect the future without a doubt, like we've talked about this, but ultimately, you know, it's all prologue. They would argue and I would argue, too. >> The businesses they've run, they've run up until now are not the businesses they're going to run from now on, >> except for maybe Anthropic. We have a preview of that with Anthropic. >> That's not That's not how S1's work.

36:09 They're backwards looking. >> That's what I'm saying. I'm saying that like it's they're they're well >> it's not that's why I'm saying this the S1 in this case isn't going to tell us anything unless it's entirely narrative based. >> All right. All right. Fine. SpaceX showed us the alternative. So remains to be seen. >> Dude SpaceX trading below its first day. opening price 148 right now. Opened at 160. >> Yikes. still valued $714 billion market cap.

36:43 >> No, sorry. >> 1.96 trillion. >> Yeah. Sorry. Sorry. That's Elon's share. I I vibe coded an app. Is Elon Musk a trillionaire.com if you want to go over there? And he is currently a trillionaire, but it's at 1.01 trillion. So, he's right now is like right on the border. >> Well, prayers out to Elon. Hopefully, you'll get through this this difficult time. Hopefully we'll get through this break and on the other side we'll talk a little bit more about the cloud business that Meta is considering. and of course those students at Brown University using chat GPT to tweet.

37:19 We'll be back right after this. And we're back here on Big Technology Podcast Friday edition. All right, so the rumors that Facebook may start its own AI cloud business, there seems to be something to it. I shouldn't say rumors, I should say reporting. This is from Bloomberg. Meta Platforms needs all the computing power it can get, Mark Zuckerberg said. But in a market star for resources n necessary to run and develop artificial intelligence products, he's also considering whether some of Meta's AI infrastructure could be more valuable if rented to outsiders.

37:50 this is what Zuckerberg told Bloomberg, "The offers that you get for using the compute are so high that it may make sense in some cases to rent out or consider those kind of deals instead of your own internal uses." Zuckerberg said, "The potential to for a cloud business is certainly there anytime we want to build it." I just want to say this. So, we got this comment on the show, and by the way, I love the comments. Keep them going, but sometimes we'll want to respond to them. This is on Spotify. You're not thinking like economists. Opportunity costs. If demand for compute is outstripping supply, why not sell off your excess and make profit rather than struggle with figuring out what to do with it? rent it out for a profit and let others figure out what to do with it. Let me I just want to address this.

38:39 Why is let's say Anthropic willing to pay so much money for your compute? The answer is because they have built a product that even if they're paying you such high rates, they can mark it up and be profitable or at least build a business for the future. The way tech companies work is they invest early in a product and they end up, you know, building building something that people want and then profiting on it later.

39:12 It's like the the this idea of like, you know, the opportunity cost. There's an opportunity cost on your VC money, right? You don't see startups taking their VC money and loaning it to other companies and saying, "Well, we're going to make a profit here." It to me it just shows a lack of either imagination or or what's probably happening is success with the product and there's the gap between the winner and the loser where you have one company let's say Anthropic willing to pay you that much that you can't say no and then finding an even more valuable use for it via their product while you sit at home you know or in your office and try to figure out what to do with your compute. Does that make any sense? That's how I see it.

39:54 >> Yeah. But where I that actually all I think like as a logical line makes perfect sense. Where I still think that actually kind of brings back the whole like race to the bottom commoditization conversation even more is exactly what you described. Anthropic has a very expensive product they can charge a lot for and so they're willing to pay a high amount for compute and then Facebook is willing to realize that they can sell them that product at a high price because they're willing to pay. But if that cost structure like a cost battle happens then that goes away then suddenly anthropic has to default to cheaper models even within claude or whatever claud code like even within cloud the moment they have to start defaulting more to cheaper models then they won't be able to pay as much for compute which means that it becomes less of an attractive business. So like and actually what I love here is like kind of Zuckerberg is basically taking both sides right like the cloud business is if it's price is high and people are willing to pay then we'll just sell compute via the cloud business to anthropic if prices are low which we're also helping make happen on the other side then we'll kind of kneecap them put them out of business and make our personal super intelligence even better. So suddenly Meta is looking pretty good in this. Yeah, I just love that though Zuckerberg said it doesn't, you know, he told Bloomberg it doesn't mean Meta is overbuilt or has excess computing power available available.

41:31 >> I mean that you know it's either one or the other. You either have too late. >> Yeah, exactly. You either have excess computing power that you can sell or you don't. >> No, no, no. It's like what what's the wire line? There's always a buyer at the right price or something like that. I'm sure there's some stringer bell line that's in there, but it's like >> they they don't have you could argue you don't have access computing. You have that internal demand at a price. But when the Facebook is mass market massive like you know I mean they have high margins high profit on advertising but like the way they would deploy to customers you know who don't pay them like like that is a they would need that internal demand but at a low price for compute whereas if it's coming in higher then maybe you you just do it again you could he can forever argue you have four billion phones with like meta AI kind of jammed into Instagram and Facebook Messenger and stuff. You could just couple of growth hacks get everyone adding prompts and using some kind of compute and doing AI stuff very easily.

42:43 Half of Facebook feeds are probably AI generated anyways now. So like you could do it. >> Shrimp Jesus. >> So you're Yeah. Shrimp. Oh, shrimp Jesus. I got to say like we can get into the World Cup stuff. Like actually just AI content overall is getting pretty good. Did you see Peptide Seinfeld? >> No. I have to see that though. I've I've watched all the Harland memes. Have you seen them? The Erling Harland memes of like the Norse versus the British like a battle of old in modern the AI slop. This did you say this? This is the week that AI slop has really transformed to like AI majesty. It's amazing. Right.

43:22 >> Yeah. It's not sl like I think the word slot it's kind of fascinating again like watch peptide Seinfeld. It's like >> well done. It's George goes on peptides and suddenly good-looking and like kind of like jacked and like it's so well and it's not just like that it looks like them and it is Jerry talking. It's just actually good a good story and funny and in the Seinfeld tradition those World I think the World Cup is going to be when >> AI video found its moment and everyone realized like you can actually make good stuff like it's no longer >> gone are the Do you remember Will Smith eating pasta from two or three years ago?

44:01 >> Yeah. And we're we've come a long way come a long way. That's where the demand is. >> That's where it is. This week, my perspective on AI videos did change, right? I went from, "Oh man, I've been fooled." to like seeing so many of the England versus Norway videos that I started seeking them out and sharing them. like this week. I'm telling you, my wife, who is a very big Harland fan, as I think are many women here, who are watching the World Cup, her WhatsApp inbox is filled with AI videos from me where I'm just like, "Watch, laugh, send. I'm like a robot. Watch, laugh, send these AI slops." Not even slops.

44:43 >> It's not No, no, it's not beautiful AI art pieces. >> No, no. I mean, I think the word slop >> is needs to be like reminded. It's kind of like everyone uses vibe coding for in a certain way. I feel slop is overused as well cuz like slop is when it's bad. >> And man, videos are getting good. People are being creative. It's like they're being genuinely they're making good content just in a different a different tool.

45:13 >> Yeah. No, it's pretty cool. And you know, as OpenAI has gotten out of that business, there's an opening for Meta, but as Meta tends to do, the company just can't help itself. This from the Guardian. Instagram's AI image generator alarms privacy experts. Meta has sparked blowback from privacy advocates for allowing its new AI image maker to generate photos of users with public profiles by default. users users of Meta's Muse image AI tool released Tuesday can tag public Instagram profiles and generate pictures that pull from faces of people featured in these social posts. Instagram users are not notified when their posts are integrated into what the company describes as its most advanced image generation model yet. So basically, if you want your Instagram photos and videos to not be able to be used for this AI engine, you actually have to go and opt out. You tap the hamburger menu on the top right of Instagram. You go to sharing and reuse and you toggle the button that allows other people to reuse your content off. That's our PSA. but you know, it's sort of like gosh, Meta Meta almost had like a really good week in AI and had to sort of spoil it with this typical, >> you know, privacy sherking behavior.

46:25 >> I can go to anyone else's. I've not used this yet. I'm definitely going to go look at this now. So I can go use anyone's public profile and use their likeness to generate new images. >> Okay. I I don't want to say definitively yes, but it seems like that is that is, you know, either that or they can incorporate it. Bottom line is your stuff is ingestable by these AI tools and then apparently remixable. So, I turned it off for my personal profile, but I kept it on for the big technology podcast Instagram feed because I guess if people remix and reuse that stuff, we'll we'll be happy about it, right?

47:07 >> Meme us. Just meme us. >> Yes, >> it's we're ready for it. >> By the way, Meta, you know, it's so interesting that last week we talked about how there are basically two points of failure in this AI industry. And Meta, as it seems to always do, has just inserted itself right into the conversation again. And how much have we talked about Meta this week? You know, >> hours. They're back. >> They're back, Zuck. We just had July 4th. We're waiting for the next one.

47:36 Zuck on the hydroofoil American flag. >> Hydroil. Yeah. Bony. It would be an What would you call it if it's not AI slop? >> AI? >> I mean, it's just No, no, it's actually something. something I've thought about a lot. It's just it's just a video. >> Like it it is >> interesting. >> I think like where the whole debate around is it AI or not. Okay. If it is if it's pretending to be something real, that's if it's like a true deep fake in that sense and like it's trying to convey that it's real. I think then that's a deep fake. That's a problem.

48:15 But otherwise, it's when it's Peptide Seinfeld, it's it's just content that's pretty funny and good and yeah, it's just content or >> new era. New era. all right, so let's close out this week talking about the cheating at Brown University. so, so this is the story from I think Institute of Higher Ed or publication like that. no, I should really cite them. I'm going to cite them. This is from inside higher ed. Okay, there is this professor at Brown. For the first time since he started teaching welfare economics and social choice theory nearly two decades ago, Brown University economics professor Roberto Serrano gave his students a take-home midterm this spring. Quite a few students had expressed anxiety about being in a classroom after a gunman killed two students and injured nine in a December mass shooting at Brown. So, it was appropriate, he said, to allow the students to take their exams home. By the end of the semester, Srano regretted the decision. Dozens of students in the class likely used artificial intelligence to cheat and earn perfect or near-perfect scores on their midterm, he said. Sorrano, in turn, made the final exam in person, which led to more than a dozen students to drop the course and even more to fail it. His welfare economics class typically attracted up to 30 students by the spring. He had taught 86, an in an increase he attributes to the promised take-home exams. When the midterms came along, the average score was 96%.

49:43 Historically, that was 65 to 80%. So, the professor knew something fishy was going on, and he and his graders ran the test through Chachi PT. The AI gave answers that mirrored what his students had written, which were kind of correct, but very off with a very convoluted style. So, he said, "Okay, the final exam is going to be in person." Here's what happened. 18 students dropped the class. Nine stayed enrolled, but didn't show up to the final exam. So you already have 27 very acting very fishy.

50:12 Three got a zero. The average score on the final was 48.6%. Again this was for kids who had an average of 96% on the midterm. By far the historic low. Previously the average on the final exam never dropped below 68%. He has a great quote the professor at the end of this. We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is okay. This leads to a declining society to a failed soci society. We cannot choose to become idiots.

50:44 >> Your thoughts? >> I don't know if we can we might have crossed that Rubicon a while back as a society, but >> before AI AI didn't do that one. I I I've thought about this. I mean again education like should fundamentally change based on the tools that are available and like I would exams should somehow be like all right you all have access who's going to put out the best work and I think like the way tests and exams have always been structured it was less around like the information will and the calculations will get done but your understanding of it like your ability to kind of extract new insights from it. That's where the value is going to be. And like to me, actually, I gotta say like Serrano come up with new like to do stuff in the exact same way you've always done and then say it's a problem given how much things have changed. I think that reflects more on the institution and the teacher than the students. Of course, they're going to do it. I don't think it's cheating in that sense. I also love like could you imagine the three students who earned a zero how much they were sweating sitting there in the exam hall just like even the average score for it. Imagine you just don't know any of it. This is like the stuff people have nightmares for and like you're just sitting in there and you have no clue what's going on. and you're just staring blankly at that paper and I say this >> reminds me of my >> Yeah, go ahead.

52:24 >> No, I mean I I feel it's never been that bad, but I feel like there's probably been like maybe an exam I never did. No zeros in my time, but like where you're just like, "Oh I did not prepare for this." Now imagine that 10, 20, 30x. That's got to >> Oh, yeah. >> got to be some sweating. No, it reminds me of my childhood where I would bring a zero home from school and my parents would say, "Nice work, but you couldn't get a three."

52:54 >> That's a joke. High standards joke. You know >> what? Anyway, >> you know, it's a joke. It's like the typical parent, but no matter what number that kid comes home from school with, they say, "All right, nice work, but couldn't you have done a few points higher?" Anyway, it didn't land. God, I guess we're at that point on a Friday, aren't we? But, >> July, it's Friday. That's horrendous. >> So, quick no personal story and then, and then some thoughts about this.

53:24 Seriously, yes, I have gotten a zero on a test. or a one of five on AP physics, which I took AP physics. >> Not one out of a hundred, >> right? No, but I shouldn't have even gotten that one because I think that was the lowest you could possibly get. If there was a zero, I would have qualified for a zero. I took AP physics in my senior year of high school and realized I was completely out of my depth on it.

53:47 I mean, I remember a few things like specific heat, but other than that, I was toast. Like trying to calculate the trajectory of some object u based off of like the mass and the velocity. I'm just like, you know what? This is not for me. You guys figure this out. I'll take the other stuff. Anyway, so I show up for my AP physics exam and you have to sit there for at least two hours, two of three. I'm done in like 30 minutes cuz I'm just like I'm failing this one pretty bad. and you don't get college credit unless it's a four or five. So I sat there for the next hour and a half like drawing pictures of the spaceship spa the solar system and spaceships and stuff like that. And I handed it in and my my physics professor came up to me or teacher came up to me later and he goes, "Hey, Alex, I heard you were the first to finish the exam. He must have done really well. He He knew I failed.

54:34 He knew I failed. And I failed. Okay. So, that that aside, I've been there. I've been there. You know, students at Brown University who got that zero, I empathize with you. I feel I feel your pain. Okay. But I I agree with you before we go. I agree with you, Ron. You know, there's been this dialogue. I wish we had more time for this. There's been this discussion since we're offloading so much of our thinking to AI, you know, are our brains going to rot? I think it's actually the opposite. I think now that I've handed so like all the like lesser activities to AI, I'm thinking through much harder problems and I feel like my brain's getting stronger because of it, not weaker. Like I'm actually able to like really think about the tough stuff.

55:19 And so I think you're right that the professor needs to realize that that's going to be the universe that we're we have going forward. and this binary just doesn't fully capture what you're supposed to be testing for at a university. >> Yeah. I I always think back I know I can drive I grew up outside of Boston and Lexington, Mass and in basically like junior and senior high school drove a lot and I can navigate more where I grew up without a map than I can in the New York City area where I've been for ever or especially like I can't get to JFK without a Google map because I'm just so dependent on on maps and that's one part of my brain I outsource. And nowadays I'm not making slides anymore, which is the greatest thing ever.

56:08 you're just having the slides made for you so you actually think about what's in there. So I think there's there's I'm sure pitfalls and people who are not going to actually take advantage of this, but do better Brown ad administration, not students. I'm team students. >> Team students. I mean, the the disparity was amazing. Like one kid got a 95 on the midterm and a 95 on on the final. So below average on the midterm, obviously the best on the final. And then one kid got like a 100 on the midterm and a zero on the final. So go figure. All right. Great speaking with you as always, Ron John. And good to have you all with us again here in the Big Technology Podcast Friday edition.

56:48 We'll see you next time on Big Technology Podcast.