OpenAI’s Risky Browser Bet, Amazon’s Mass Automation Plan, Clippy’s Back
Channel: Alex Kantrowitz
Published at: 2025-10-27
YouTube video id: dYa2bz89W20
Source: https://www.youtube.com/watch?v=dYa2bz89W20
Open AAI is in the browser game. We go hands-on and tell you whether it has a chance. Amazon has an ambitious plan to automate hundreds of thousands of jobs. Meta does major AI layoffs and Clippy's back. 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 nuanced format. We have a big show for you today. Lots of ground to cover. We're going to talk about OpenAI's Atlas browser. We're going to talk about Amazon's mass automation plan, or should we even call it that? It's a fascinating story. We're going to talk about the the layoffs at Meta's AI division fair. And of course, Clippy has returned, which we will herald at the end of the show. Joining us as always on Friday to do it is Ranjan Roy of Margins. Ranjan, great to see you. >> Good to see you. Clippy's back. Clippy's back. I cannot wait though. They are called Mo or Mo. We'll get into that in just a bit. Mo as in Microsoft co-pilot. Mo get it very catchy. But I am I'm looking forward to our clip. >> That works. That works. That works. >> I've heard worse names. >> I I've actually That's not a bad name. I I'm going to give it to them. >> Okay. So, I think Atlas is a great name. And Atlas, of course, is the name of OpenAI's new browser. Uh it is an AI first browser meaning that there's a side panel where you can browse or talk about the web pages that you're on with chat GPT and you can also start to ask it to do things like book me a flight. Then it will ask you do you want me to actually go ahead and uh access these websites with your logged in accounts or logged or as if you're not logged in and then away it goes. It will do actions for you. Of course in the browser wars many companies have tried to unseat the incumbents uh but few have succeeded and by few I mean none have succeeded. So Rajan both of us have had a chance to download Atlas and play around with it. What is your first impression here? Do you think that it's going to be a legitimate competitor to Google Chrome and Safari? >> I think AI browsers are going to be the future. And even when I say first impressions this week, I used DIA from the browser company. I've been testing it for a number of months. I've been using Perplexity Comet. So to me, the Chat GPT atlas it not it wasn't anything that exciting in terms of like what is interesting about an AI browser. And again, having that kind of like right panel, being able to rather than copying and paste text, which I do all the time with a chat assistant, having it all right there and ready to actually be kind of like ingested into your me like chat assistant's memory, being able to ask questions, summarize things, all of that. I I think there's no doubt in my mind that that is going to be the future. I I'll say though the action mode do stuff for me and we've talked a lot about thought partner versus actually doing things. I had not seen anything that interesting at least in my usage so far though I have not been asking it to do too much and giving it too much sensitive information and we'll get into why in just a bit. What about you? >> So I have definitely enjoyed using it. I think that this is the future of the browser. The question is, is it going to be something that OpenAI can ride and unseat the incumbents or that the incumbents will effectively adopt into their products? And this idea of having an AI assistant or an AI chatbot in your side window and then giving it some capabilities to go surf the web to me, you know, it would be great and really meaningful if your competition wasn't Google. Yeah, maybe there's a chance that it unseats Safari in some way, but uh I you know trying to go against Chrome is going to be really really tough. Uh because Google does have the the talent to build this into their product. They will they already have Gemini baked in uh to some extent. So many things on the internet work as with Chrome as the default. Like you and I today we're recording on Riverside. It's a podcast recording platform. Um it doesn't work outside of like Chrome and Safari. So you basically have to get all this compatibility built in from the ground up. Although I do think actually Atlas is built on Chromium, which is an interesting >> which is an interesting uh um sort of wrinkle in all this that it's built on Google's open-source browser technology. Uh but then there's other there are other things that are just not great there. Uh for instance, the New York Times seems to block it. So, if you want to go to ny times.com and then like ask the the uh bot in the side panel of the browser to like navigate the site for you or to uh speak with it about the stories there. It just can't can't do it. Obviously, New York Times and OpenAI are engaged in legal battles today. So, uh there's obvious strengths. I do like I do agree with you. It's going to be uh a feature in browsers moving forward, which is exciting. Uh but but maybe this isn't the one. >> Well, I I think to me there's two levels of how you can use an AI browser. Again, the first one is just summarize some stuff for me, take some information that's on a page. Uh like I I very regularly with chat GPT now I have to travel a lot for my work. I'll kind of like paste in my flight info, hotel info, meeting schedules, and then ask it to kind of make me itineraries. That's a lot of manual copy pasting that now already I'm able to just say pull my flight info from this page and just pull it into your system and into this project in memory and it'll remember it. So so that layer I think is is already working very smoothly. I think the the bigger issue though is I I I actually had luck when I was doing my taxes this year and this was with Dia from the browser company like fill in this form. What do I need to do for this box? That was kind of my first aha moment of like having to fill out a really complex annoying form and actually it going and doing all that work and then providing me suggestions in that right panel of what I should be potentially filling in. But I didn't actually have it fill in the form for me. I had it kind of analyze it, understand it, explain it to me, but still I was I was adding in the numbers. Like to me that's still that next step. I have personally not done anything yet where I've like had to go out and I don't know run an entire workflow or do some entire I don't know take take a take some kind of meaningful action for me. Did you did you have it actually buy you a flight ticket or a a new a new pair of pants on Amazon or something like that? >> Well, Lord knows I do need new pants. Um, but I will say >> listeners cannot see Alex is wearing a pretty cool new blazer today. So >> that's right. We're uh we're powder blue here today. >> Powder blue >> in honor of our descent into winter. Um, we I just saw one of those tweets that it's not going to be light uh at uh after 6:00 p.m. until like May uh in in New York. And I guess like this longitude or or latitude, I don't know which which one it is, but >> latitude. Latitude >> is latitude. And uh you know that makes me sad. So I had to bring in some some light colors for uh for some contrast. But um one Okay, I'll tell you one thing that was interesting to me. I did go to try to book have it book a flight for me and um there were limited flights from New York to Berlin and I was curious why. So, while I'm having it like go and try to book the flight, I'm now talking to the chatbot about the reasons why it might happen. Like, there was this cyber attack at a bunch of different airports, and I was curious whether that had something to do with it or not. Uh, so I do think that there's this like really interesting capability where you have the action and the sort of uh information both at your fingertips. Uh, and I think that could be really interesting. We I did see though RS Technica had an editor basically go and pressure test it in a bunch of different ways and I was fairly impressed with some of the things that the bot was uh supposed to do. Should we go through a couple of these cases and we can talk about whether they're impressive or not? >> I actually loved these examples. >> Okay, so this editor, by the way, is a complete madman for what he had this bot do. So the first thing he asked it to do was go to uh 2048, the game 2048. If you don't if you don't know what 2488 is, you kind of mix uh you you you collapse these cubes into each other. Each has a number that like numbers will collapse into bigger numbers. So if you have a two and a two and you move them, you get a four and then you have to get another two and another two together and another four and you move converge the fours, you get an eight. That's my best explanation of 2048. Although I imagine most of our audience has probably played it. >> I I have not and your explanation didn't clarify it for me, but but I can tell it's a tough one. I can do it. >> This is a public service. So, you should, by the way, if Ronjan, if you travel a lot, I guarantee you 2048 is on at least some of the uh inflight entertainment systems of the of the airlines you fly on. Just go to games. Okay. But anyway, >> very quick note, I I was flying back from London this week and I I ended up I've already watched every movie that's available. I started playing solitire on the inflight entertainment system and I realized I hadn't played that in like two decades or something like that. But I I don't know if for for any of the older listeners there, there was a time when that was the only game that was really available on the computer by default and everyone in the world was playing solitire and I still remembered how to play it. >> Still still enjoyed it. >> Don't engage in mind sweeper eraser here. All right, you can take your solitire, but if you really just believe that mind sweeper doesn't have a place in the history books, uh I got nothing for you, Ronchan. Okay. Can I can I continue this or should we keep talking about continue? >> All right. So So here's what the editor said. When it came to the actual gaming strategy, the agent started by flailing around, experimenting with loop sequences of moves like up, left, right, down, and left, down. After a while, the random flailing settled down a bit with the agent seemingly looking ahead for simple strategies. The board currently has 32 tiles that aren't adjacent, but I think I can align them. Basically, this a this agent went and the on the browser went and started to figure out this game uh and ended with a score of 356 after a couple minutes. Um but after this editor prompted it to keep playing, it ended up with a score of 3,164. Um and the editor says that's pretty similar to the score I was able to get in a test game as a 2048 novice. I mean, the fact that this bot, you know, with access to the screen was able to get this, I won't call it the most complex game in the world, but a decently complex game, uh, figure out the rules seemingly from, I don't know, maybe from scratch or not, and and play to a novice level to me was quite impressive. Well, I I I actually would be curious, is it, and maybe this is too granular, but like I'm I'm very curious. Is it learning in the moment and on the fly or is it actually going out and learning the rules of Play 2048 like from other websites or from the website itself? Because it actually is a difference in terms of like how these systems will learn and how they will navigate. Like are they using contextual knowledge or are they actually able to look at a graphic and understand it in that moment, do a move, then have to take kind of like another screenshot, understand that image. I'm actually not sure, but I I think cuz the latter would represent like a pretty insane level of intelligence, let's say. >> Yeah. One thing I'll note is that AI has always not always uh AI has been trained to play games. So this is something in the toolkit of AI. Uh but it does seem at least from this explanation that it sort of figured out what it what it was doing on on the fly. Basically going from repeated patterns and then learning uh to get better as it went. So >> all right. So games games are okay. Check. >> Yeah. I think this is impressive. Okay. The other next thing this editor tried to do is uh he tried to get the uh agent on the Atlas browser to go to a radio station um find the songs that were playing and then and then turn those songs into a Spotify playlist. I I don't I I'll just cut to the chase. The the bot was really able to do this. I mean, the only limitation here was the fact that it didn't want to work forever, uh which I understand, but it was able to navigate to the radio station web page and build a playlist. I just think that's such a cool thing. Like if this is something that's going to be operating in the background to like go to local radio stations and have it build playlist is impressive. Oh, >> I I think this one was very impressive. But also, if we kind of extrapolate what that means it can do. It's an always on browser that is doing something. It's able to actually find a bit of information within that and then go into a completely separate system and take some kind of action. So, so I think this actually is like a good representation of lowrisk. It's adding a song on a Spotify playlist. If it screws it up, it's not the end of the world. But like actually taking information from one system and putting it into another, you know, like authenticated system. So, so all right, Spotify check. Not bad. Not bad. on that use case. It the editor also had to go through his emails over the past week, collect all the contact information, name, email address, phone number for PR contacts uh in those emails and add them to a Google Sheet spreadsheet. And not only did the bot do this, uh it added the relevant company name, which uh this editor did not ask for in the prompt. It had 12 rows of well- formatted data for 12 different PR contacts. Uh the the guy says it stopped well before it had a chance to go through all 164 emails returned by that initial Gmail search. But that too is impressive to me. Well, this is one actually I think this raises an important question around AI browsers. This is the kind of work that I still it blows my mind that Google for how much better Gemini standalone has gotten like why this stuff does not work in Gemini and regular Gmail and sheets and how they don't work together well because come on guys like this should be table stakes for Google itself when it's only it's in its ecosystem but I think it raises a big question of like I've been thinking about this for a while like this browser takeover idea this kind of like I'm going to use a browser. I'm going to use the web as it is. But if you think about it, a problem like this is anformational one. So it should be solved by like within Gmail itself or with the Gmail API or with model context protocol and not a browser kind of clicking through robotically your Gmail. And so it's still amazing to me that this system it seems impressive and useful, but to me these are the kind of use cases that don't actually make sense for for an AI browser, >> right? Well, the question is, do you need a browser or is it just an extension? And why is this a browser? And so let me just put it to you because um I think this is like a good opportunity now. We've seen a little bit about what it can do. Uh we of course know the strengths and limitations. Do you think five years from now we're still talking about OpenAI's Atlas browser? >> No. No, I don't either. I said I said that like pretty quickly. But the reason is OpenAI has been on fire with product releases, but they're releasing a lot and a lot of it is kind of halfbaked. And in this case, like I don't think the browser element is going to make or break their business. I think it to me even the way I see the world going is I actually think websites become less relevant. Clicking around a website doesn't become that primary interaction that an AI has with that the underlying information. Again, when we talk about travel booking, yeah, it's nice that a browser is going to go to Expedia and click around or whatever it is, but like to me, there's no reason it shouldn't have the the more efficient way for it to work is to actually just directly be connected to some flight database of information and like be able to interact with it and take some action as needed versus it it's just clunky and inefficient to have to navigate a website for an AI. So that to me that part is the one thing that doesn't make a lot of sense. And then I also and we're going to get into security again, but like I think this is going to be a huge issue for this entire concept. And I think OpenAI is probably the least trusted in terms of like working on security relative to like whether it's a Microsoft or a Google or even an Apple or whoever else. I think OpenAI is going to lose that battle any day of the week. So what do you make briefly about the fact that OpenAI does create many of these apps with a lot of buzz and then we have seen them fall off. I mean uh Sora I don't know what the longevity of Sora is at the top of the app store charts but I'll tell you it has been replaced at the top of the top charts. Uh do you know which app has replaced Sora on the top of the app store charts? >> I do not but I'm very interested. Alex is holding up his phone. Dave's Hot Chicken. >> No. What? >> Dave's Hot Chicken is number one now. >> Yep. It's tender, juicy, and made to order. It has a really actually a great mascot. It's a chicken actually. >> Oh, I I I know uh listeners, longtime readers of margins will know that fried chicken is something that I uh spend a lot of time thinking about and and and eating and interacting with. And uh >> Dave's Hot's pretty good. It's pretty good. I'm actually >> I have to say I'm looking at the the app home screen here and I think I see why Dave's Hot Chicken has uh displayed Sora. I'm counting there's at least four fried chicken patties and it looks, you know, to my counting maybe eight buns. So maybe it's just a a quadruple decker chicken sandwich and the internet could not handle that. Do you know who uh which musician is a major investor in Dave's Hot Chicken and actually made a big a lot of money when it was uh when I think it was some PE firm bought him out for like a billion dollars. >> Dave mean that music? >> No, that would have been too easy. Drake for whatever reason. >> Drake. >> Drake. Drake. >> Has to be Drake. >> Drake was uh one of the big guys in Dave's Hot Chicken. >> Okay, but let's get back to the matter. We digress. We digress briefly. Um, so what do you think about this? Like is is OpenAI obviously OpenAI has shown that it knows product and the flagship product is doing incredibly well. Chat GPT 800 million users although maybe 750 million now because Meta I don't know if you saw this they removed 1800 chat GPT from WhatsApp for whatever reason. Um but but so they're they're obviously good at product. Uh but they they I guess it's good. I my personal perspective is it's good they're taking all these shots. What do you think? I think it's it's a tough one because like Sora I I mean I've gone back on there. I still again with my son I think I can tell you the most recent Sora creation was he wanted to be playing a trombone that farted and I I went with it and that's that's use case number one for Sora in today's environment is just stupid fun ridiculous videos that you quickly make you laugh at and then you forget about like on my feed there was nothing interesting or new like those first few days I feel like Sam Alman was all over there there was there's a bunch of other, you know, there's creativity happening. I have not seen anything. So, that is a perfect example of like the technology is impressive. The app and the experience is just kind of thrown out there. There's some hype and then you move on. I think it either there is some real master plan of how all this is going to come together or they're just throwing a bunch of stuff out and they have so much money that they can just product develop the hell out of whatever they want and then they're just going to anything anyone does in the AI space they'll throw out some kind of lukewarm version of and I mean maybe it lands, maybe it doesn't, but do do you think there's a plan here or >> not really but uh I I I mean they're going after big markets, right? So sort of big big market going after the social media apps like Tik Tok. Um obviously if you get a hit browser, it's worth taking the ch the chance because then you sort of control search and you don't have to pay for defaults like Google's had to with in in the mobile space. So, I mean, I was speaking about this with Panos Pane uh from Amazon this week about uh the the you know, do you want to be really refined in your product strategy or do you want to throw spaghetti at the wall in your product strategy? I am I love the spaghetti at the wall strategy as long as you have the money to back it up. And so, I think that this is the right move from go for it. making a great browser like it still takes work in the maintenance of it and like kind of like again if if the interaction of every web page you load is it's being pinged as as like an AI crawler AI browser and that's going to cause some issues that you're not anticipating like all that investment can go out the door if you're not actually spending a good amount of time on it so so I think I don't know it feels like that chat interface. They dominate. They own that space. Everyone is in there. Like just basically like killing the web off and letting all interaction take place in chat GPT as a strategy to me would make more sense than trying to just be everywhere at once. >> Oh, they are trying to do that too. But you do make a good point and it's worse if uh if you end up using this browser and you open yourself up to security problems. So talk a little bit about that. >> All right. So this is my favorite actually. Did you ever in terms of browsers, did you ever uh use Brave? >> No. >> Uh quick quick history. This one was personal for me. Uh Brave, it was a it was a company. It actually came out during the first crypto boom in like 2017. And they had a coin called Basic Attention Token. um that basically the whole idea is like you as a user would get paid in a cryptocurrency to look at advertisements. I was like obsessed and fascinated by this whole idea. never really panned out, but they have a browser. Um, it's it's I think it was one of the founders of Mozilla initially, but anyway, they uh released this whole research where they actually went out and they created the what they're called prompt injections. And the idea is you're an AI browser, you're looking at a page, you're like playing 2048 or you're on a local radio station trying to figure out what song it is and there's unreadable text to a human eye like that's like in the same color as some graphic on the page that is instructing the LLM to do something malicious. And again, just simply having it take some kind of weird action is like uh they're able to do. It's again like uh I think they had it just go out onto another website and show that this would be possible but you can imagine starting to see or even like one of the examples they gave is you can add comments onto a website. So not even one that you own but go on to other websites and add comments on Reddit or product reviews on uh like retailer pages and then trick the LLM into going and doing something. So just imagine if people are really opening I mean you just open up your Schwab Fidelity Robin Hood account and suddenly the browser can go do things that you have no idea that it's doing the it's able to take actions in the background as we talked about is cool that it's aggregating your Spotify playlist when you're not even looking at it. It could also be in I mean imagine instructed by unseen text on a website to go transfer your money to some other account like I mean this this is >> stock up oncoin. Yeah, go stuff. I mean, no, but think about like if it's that easy, people are gonna exploit the hell out of this. And yeah, the the more I was learning about this, the more I I did not log into anything sensitive when I was playing with Atlas. I was only having it just like even logging into Reddit. I was a little nervous around like like but definitely not banking information. I was not even logging into my Gmail while using it. >> Right. Okay. I just want to uh sort of put a bow on this because we've talked a lot about like these different agentic capabilities and this is a word that keeps coming up on the show. Um, we've had a number of listeners reach out and ask, you know, why haven't we talked about the Carpathy interview on the Dwaresh podcast where Andre Karpath is basically like it's not the year of the agent, it's the decade of the agent and the technology today is just not ready uh to sort of take that mantle and it's going to take about 10 years. U to me I so let's just spend a second on it. I'll say to me that totally tracks with what we've been saying on this show that this is going to be it isn't going to be overnight. Uh there are going to be places it works and places it doesn't. And the real question is um whether it match you know is a mismatch with the funding or not. You put all these all this money in and you have to wait 10 years. Uh you could lead to some bad aspects and people were saying that oh this this is the interview that's going to pop the bubble. Um doesn't look like that happened this week. The S&P 500 is up a point and a half. So >> alltime highs, alltime highs, >> alltime high. So the Carpathy the interview didn't pop the bubble. But I am curious what you think about his perspective on the fact that it's going to just take a much longer time for the agents to work. >> Oh, I loved that interview. I mean, I think he was it a lot of people, I think, misread it as overly bearish. I took it as incredibly realistic. like the way he even talked about the change management side or like basically organizations and people aren't necessarily ready for what the technology can provide and in many cases the technology is not there for a lot of the promise that's being made or as we're talking about right now like larger systems are not ready and we're going to talk about Amazon and robotic automation but like in so many of these domains there's a lot more than just can I make one quick run and show that this is possible. And I think he was really getting at that. And I think I to me it was an incredibly realistic perspective on the industry as opposed to a bearish one. So I'm I was pro Carpathy on this one. What about you? >> Same here. I like what he said that he's actually much more optimistic than most of the skeptics, but also some of his realism compared to like Sam Alman saying that this stuff is going to cure cancer in his most recent blog post um I think is like taken as pessimism. But I I we've both, you know, I think from the start of our conversations about this stuff, um we've tried to be as realistic as possible. And so yeah, I'm going to definitely, you know, praise someone who's being realistic about it. And I think I've talked about it before, but uh Carpathy has his own YouTube channel and they've the videos he puts up there have been uh terrific in terms of just explaining the basics of large language models uh and where things are heading. So all right, I definitely want to get to Oh, go ahead. >> And I really want to put a bow on this segment as I have just learned. Do you know why Dave's Hot Chicken app is the number one app in the app store? >> I haven't gotten that far. Apparently, in honor of Drake's 39th birthday two days ago and through this week, they're giving out free sliders for Drake's birthday. And that if you download the app, I I see a video here in New York 56th and 8th, which I've actually been to that Dave's Hot Chicken. There was a a bit of a mini riot out there of people trying to get in and uh get their free sliders. So, cheers to the marketers out there who could still cause a mini riot by just giving out free sliders if you download their app. >> Well, allow me to say it loud and clear. Uh, happy birthday to you, Drake, and congrats for being the reason why Sora is no longer number one on the app store. You did it, man. You did it. >> You did it, Drake. Thank you for just bringing us back to what what tech should be, a brand tricking you into downloading their app for some chicken. going back to the roots. So yeah, on the other side of this break, we are going to talk about this crazy Amazon story. To me, the biggest tech story of the week, the New York Times headline, Amazon plans to replace more than half a million jobs with robots. That's coming up right after this. And we're back here on Big Technology Podcast Friday edition. Major story in the New York Times this week. Amazon plans to replace more than a half million jobs with robots. I'll read the beginning of it. Amazon executives believe the company's on the cusp of its next big workplace shift, replacing more than a half million jobs with robots. Amazon's US workforce has more than tripled since 2018 to almost 1.2 million people. But Amazon's automation team expects the company can avoid hiring more than 160,000 people in the United States. It would otherwise need by 2027. That would save uh 30 cents on each item that Amazon picks, packs, and delivers to its customers. Uh executives told Amazon's board last year that they hoped robotic automation would allow the company to continue to avoid adding to its US workforce in the coming years, even though they expect to sell twice as many products by 2033. That would translate to more than 600,000 people whom Amazon didn't need to hire. All right, couple things here. It's not that they're going to replace the people as far as we know that are in the workplace in the the um the warehouses today or fulfillment centers as Amazon likes to to call them. It's that they the uh coming robotics would basically alleviate the need for them to hire more people. So this headline they're going to replace the um the people I think is a little bit uh extreme. Um but I'll just say this one thing and then turn it over to you Rajan. Um we are seeing the robotic technology increase at an unbelievable pace and of course a lot of that has to do with uh years of research behind it but also better uh AI models within these within these things. And for Amazon the real problem has always been uh picking uh putting the items in the the rack and then picking them out and then putting them in a package. That requires real dexterity of the human hands uh that robots have not been able to do. They're very good at shuffling uh racks uh back and forth. That technology has really improved tremendously uh in recent years. And um whether that's going to uh uh happen or not, I don't know. But part of this story says Amazon would like to see 75% of its fulfillment center workforce automated. Uh which means that they believe that this technology is really progressing. So let me turn to you Ron. What did you think about this story? And is it is it as scary to you as it sounds to me in terms of the uh impact on employment? >> I think it's as scary as it sounds in terms of like the impact of unemployment. I think it is inevitable again having been in the retail world and like pickpack ship is what you're describing and seeing what robotics are already feasible and you have it correct that like like actually like receive like you know a person reaching into one container pulling into another understanding which product it's actually going to they are actually going to pull putting it into a robot that then takes it to a different part of the warehouse that's how things already work to replace that person who's doing the picking and the packing into like the as part of the process is it's such a different skill. Robots I do think are moving in that direction. I think the where I see the speed of it being very difficult is like uh warehouses or fulfillment centers are still built for people to do that work. So, are you gonna have a And this kind of gets back into like Optimus and do robots have to look like humans and have the same dexterity as humans versus you actually build the fulfillment center specific to what robots are best at? Um, but yeah, and which we debated in the past, I think, but uh but overall, I think it's it's telling and I to me there's no doubt that this is what a fulfillment center will look like 10 years from now. And I think we we do like to critique some of the PR around this stuff. Um so before we get deeper in uh there's these lines from the time story that are amazing. So this is uh you know something where like it's clear that Amazon wants to buy the community off. Amazon is so convinced this automated future is around the corner that it has started developing plans to mitigate the fallout in communities that may lose jobs. Documents show the company has considered building an imi an image as a good corporate citizen through greater participation in community events such as parades and toys for tots. It is so funny because like you know you see these program you see oh here's Amazon it's a sponsor and it's never quite clear that there's a line between yeah we're going to you know automate jobs in your community or or raise the standard the the um raise uh the prices of living because of the salaries we pay engineers and it's just clear that the company blatantly sees these community initiatives as like um PR tools to blunt the impact. There's also uh the terminology that they talked about using. Yeah, I this this really just made me want to fall out of my seat, but I was on an airplane when I read it and it would have been impolite. Um the documents contemplate avoiding using terms like automation and AI when discussing robotics instead and instead use terms like advanced technology or replace the word robot with cobot. >> I mean that one it doesn't it doesn't bother me as much. I think I think uh it's I honestly would rather they're not just throwing in AI for the sake of AI and actually talking about what this what it is. But it I will say in this market and environment, it's almost scarier when a company is like worried about their technology being so good that they're afraid to say AI versus every company that's just stuffing AI into whatever it's saying for the sake of it. So, so maybe that that actually is a testament to how far along they are, >> right? But it's it's manipulative. That's what I'm trying to say. It's like, you know, it is a specific language exercise and community initiative meant to try to blunt the image of what it's actually doing. Um, and I just find that to be uncool. But well, so my question is this is one of the big especially for a retailer this forecast that they will potentially sell twice as many items. If no one is employed, no one is buying these items. Like this is the the kind of catch 22 I see for an Amazon specifically like for potentially other tech giants that are not just selling consumer goods to everyone to like the mass of the population that if there's a significant impact on employment they're not selling twice as much stuff like there's no way >> maybe probably I mean hasn't that been the problem in uh in the US in particular for so many years we've gone from a place where if you work on the automation or if you work on the assembly line of a car, you should make enough money to buy the car to now that you know you you sort of you you might work on the floor of Walmart. You don't have enough money to buy the car and Walmart then goes to the suppliers and demands such cheap prices that they send the jobs away or automate them. We've just been in this declining uh cycle of lower wages. um and greater consumption and just eventually it's yeah it's all it squeezes um the squeezes everything to the to the bottom of the barrel at a certain point and I think that's where we stand. It's probably big reason for all the unrest that we have in the US right now. Yeah, I mean that's a much bigger discussion I think but but I think the there's already been a lot of pressure on consumption around how like the models of employment but we America's done pretty good so far on the consumption side of things. Um but basically we're going to have to invent an entirely new form of credit. It's the only answer so everyone can keep buying stuff. I think that you should just take >> I say that in just to our listeners just to make sure. >> Yeah, you should take credit from the uh Nvidia Bank of America and then just use that to fund infrastructure and then >> and your own personal consumption and buying random on Amazon. Yeah. >> Then when the AI bubble bursts, there will be no >> cascading problems. I'm glad we're not doing that. I'm glad Mr. Beast might be a bank one day, but Nvidia doesn't seem like it's on that trajectory. So that's good. uh Amazon statement. So they said the the documents, this is all in documents. The documents feed by the times were incomplete and did not represent the company's overall hiring strategy. Amazon also said that it's not insisting executives avoid certain terms in that community involvement is unrelated to automation. And to that I say sure. >> I'm sure as well. >> I'm okay with that. Amazon. So, let's just talk lastly about um about what's going on. So, there is a facility in Sher, sorry, there's a there's a facility in Trevorport, Louisiana that is a template. Have you been there? >> Uh, no, but I had a friend from Shreveport. That's the only reason I know. >> God bless your friend. Thank you for your friend. Um, it's a template. >> Saved Alex on this one. Jacob, >> thanks Jake. Big thanks to Jake. Jake, go get yourself a a chicken sandwich for free. >> Hot chicken slider. >> Yep. All right. So, anyway, it's a template for uh this Shreveport fulfillment center is a template for future robotic fulfillment centers. Once an item is there in a package, a human barely touches it again. The company uses a thousand ro robots in that fulfillment center, allowing it to employ a quarter fewer workers last year than it would have without automation. Next year as more robots are are introduced, it expects to employ about half as many workers there as it would without automation. So I just want to say one thing. Uh I wrote this book about workplaces. Uh the workplaces of the big tech companies um including Amazon was the first chapter. The book is called Always Day One. Okay. So it's like clearly like Amazon inspired. My belief was always that these companies would find places for people whose jobs they automated to do different things. Um, but the one place that I really could never uh fully wrap my head around on whether that would happen was the fulfillment centers. And I do I mean I I think Amazon's going to accomplish its goal. I think they're going to get to that 75% automation. And I think that is going to be, you know, I don't know. I think it's going to be a clear problem. Well, I think >> because it's a major employer, >> but the way you what you're describing, I think the real reason there's so much kind of like hype and unrest around potential job losses, it's no longer in the fulfillment center. It's going to be in the corporate offices that you'll actually see. In the past, in the corporate office side, that's where it always was. You know, you can uh you can give them something else to do. There's going to be new growth opportunities. To me, that's why this is such a different story right now is because that's where the potential job loss is, >> right? But I think what the Amazon story is showing is that the potential job loss might actually be in these fulfillment centers uh as these robotics get get uh more advanced. So, I mean, we could see it in both areas really. >> Well, I mean, and speaking of white collar jobs potentially going away, did you see the uh open AI investment banker story? >> I sure did. And as someone who has sat on the floor of a banking institution, I would like to hear your perspective on it. So tell us what happened and give us your thoughts on whether this is going to be the end of the junior banker. >> Well, and and also I I was on the sales and the trading floor side, not the investment banking side, but even more so, I think the investment banking profession has certainly been ripe for a bit of disruption. And basically OpenAI apparently there's something called project mercury where they've hired over a 100red ex bankers to train their models to basically you know like do that tagging work good answer not good answer you know start helping like the models understand different transaction types IPOs like you know what are the documentation look like what are the right ways to approach it I think to me like >> wait they're they're uploading full full financial models. I mean, >> yeah. Uploading their own financial model that the models can learn from this stuff. >> I think I mean this is it's an interesting approach. So, honestly, the most shocking part of it to me or the funniest part was apparently they're only getting paid 150 bucks an hour, which is not it's I mean, it's not a ton of money for especially for if you're in the banking world. That one surprised me a bit. But overall like this is I mean this kind of work repetitive work that's kind of just aggregating and kind of like routinely synthesizing information. That's the stuff that LLMs were made for and that is what bankers would stay in the office till 3:00 a.m. doing. So I think I I kind of I enjoyed this one. I think we should all should we all celebrate project mercury. >> Uh maybe I see this is the thing it's still it's still so unclear about where this is heading. Uh I'll give you one example, right? Okay. So like what happens when this stuff is automated? Um, so I was in Mountain View this week as part of uh one stop I had on a longer trip across the US and I got a chance to speak with Yosi Matias who is the head of who's the head of Google research and um I asked him I said well do you think AI you know basically we talked last week about how Google had this model that came up with the hypothesis about cancer cancer cell behavior and that was actually confirmed in living cells to be true which is unbelievable. So I asked him I was like do you think AI which can go out and do this stuff is going to replace researchers or how does this sort of net out? Um and so what he said is uh we're actually going to need many more researchers in all disciplines. The only situation where you're going to need less researchers is if you assume that we practically have answered all unanswered questions. I don't think anybody here would think that we are only understanding a tiny bit of what we need to understand. Using alpha fold as an example, AI automated the decoding of protein folding. It led to the need for more protein researchers because once you have that foundation, you can actually work on bigger problems and not just decoding the structure of protein. So that's sort of like my question is like is is every uh occupation going to have the same thing? Like if you think about the vast space of research that has yet to be uncovered, you can automate, you know, all the work that researchers are doing today and you have so much more to do. And so does that hold in like places like investment banks or even in in companies where they're automating customer service? Uh does that hold and do they actually have the same situation as the researchers where they're going to need more people as opposed to reduce the amount? I'm still I'm going to say no because in this case again you're going to be there's going to be less of a need to raise capital because no one's going to have any money and there's going to be no businesses growing. So bankers don't need to go put together a pitch deck and kind of like issue some new to buy some new debt or you know like raise some new equity on one side of it. But also the correlary of that would be potentially there's so many new financial products that are being created thanks to AI that it somehow maybe creates that new wave of credit that we were talking about a moment ago. But that doesn't sound appealing to me either like getting more potential financial products. So, I don't I don't I don't see the correlation between the like protein folding and kind of like cellular research and bankers being able to automate an S1. >> Hold on. I want to I want to turn this back to you because uh are you saying that we're going to have this technology that's going to basically advance productivity so much that and of course this is the standard argument but are you saying we're going to have this technology that's going to advance productivity so much that it's going to be able to automate investment junior investment banker jobs um and that will lead to uh society that has so much less money that they won't be able to consume anymore? Well, I mean, not as direct as, but it it all falls in line with Amazon as well. They're if you're you're automating away the warehouse jobs, people don't need to buy more stuff, so you're not going to sell twice as many products. You're automating away more white collar work, so the amount of companies that actually need to raise capital becomes less. Like, I mean, I I I think that's the weird balance that in this whole kind of job loss discussion is ignored. It's like what are the follow- on effects of not having the job then the business itself doesn't keep growing. So yes, that is what I'm saying. >> Okay, I will take the other side of that. Uh but of course uh we'll only be able to know in time. Okay, one last thing about this because I think it goes along with our job. The interview process for these bankers is fascinating. Have you So So this was the craziest part of the whole story to me. The application process for project mercury involves almost no human interaction. Uh the first step is a roughly 20inut interview with an AI chatbot which asks questions based on the applicant's resume. The second phase test candidates on their knowledge of financial statements. The final stage is a modeling test. I mean this is just like reading those two paragraphs to me is insane. the idea that you're going to be able to like conduct an interview and test an applicant's financial knowledge. And again, this is coming from OpenAI. So, they have the jump on how to use this technology best before everybody else. That is just fascinating. They've literally replaced the human interviewers and they have they're trusting these people. They're going to put the financial models that they make into their into their AI model. I I weirdly like this part of it mainly because their job is going to be interacting and communicating with an AI. So, what better place to actually have an AI interview you? It's one thing if like you're interviewing someone for a job and you're trying to just like cut corners and having AI imagine like you're having a your voice through 11 Labs conduct the interview and faking it. But in reality, that person is going to have to interact with people. So that's a slightly different thing. But this part place, their job is to talk to the models. So why not start that in the interview process? >> I'm loving Project Mercury. I'm loving Projectur. >> I'm not even I'm just I'm not even criticizing it. I'm pointing out how crazy it is that they have the technology that's able to do this. >> Yeah, that's bananas. >> It's bananas. I I agree. But it oddly more so than Atlas, I'm going to give this is OpenAI's win of the week. Project Mercury. >> I'm starting to understand really why the Hot Chicken app is the top app. The sorry, Dave's Hot Chicken app is the top app. Here's what's happened. America's jobs have been automated. They've spent their days using Sora, but they are hungry. It's Drake's birthday. They need to eat. And now Dave's hot. That's the Look at our future. It's the Dave Chicken app future. This is this is where we're going. >> Listeners, go online. You will see people clamoring and like pushing each other to get some free sliders for downloading the Dave's Hot Chicken app for Drake's birthday. It just this story pulls together everything about job loss, automation, celebrity culture, hot chicken, and food trends. Everything in one tight package. >> That's some This is the future that the AIVCs want, isn't it? >> Is this what you guys wanted? >> Well, that's not too bad. At least people are getting free chicken. Okay. >> Yeah, but not you don't get your chicken unless you're fighting your way to the front. >> Well, you got to got to work. >> Nothing in life is free, Rajan. Nothing in life is free. >> No, >> no such thing as free chicken. >> Not even on Drake's birthday. >> Maybe there is. Okay. Uh, one more break. When we come back, we're going to talk about the cuts at Meta Super Intelligence Labs and then of course the return of Clippy. We'll be back right after this. All right, we're back here on Big Technology Podcast Friday edition, bringing it home. Meta has cut 600 jobs in its AI super intelligence labs after spending, I think, billions of dollars on AI talent. Here's the New York Times. Meta said on Wednesday that it cut approximately 600 jobs in its artificial intelligence division. Um Mark Zuckerberg has been on a hiring spree to stack his company with AI researchers. Uh but he has started to cut. The cuts on Wednesday did not affect the newest hires who've been empowered to develop super intelligence. The job cuts were aimed at cleaning up the organizational bloat that resorted that resulted from three years of building up Meta's Meta AI's uh Meta's AI efforts too quickly. Uh the layoff help layoffs helped uh layoffs aimed to help Meta develop AI products more rapidly. Uh interestingly though they did cut um some long tenur seemingly superstar uh AI researchers. Uh Yuan Dong Tian is one of them. Several of my team members and myself were impacted by these layoffs today. uh he's not somebody who actually started um you know who whose recent uh a a someone who's been hired recently. He actually was hired 10 years ago uh by fair and most of these cuts came in the fair division Facebook AI research or they call it fundamental AI research now the division started by Yan Lun to uh advance the frontier of AI research within meta Rajan what is going on over there >> well is Yan Lun going to stay do you think >> uh >> he's been he's been on this show a number of times right >> three times uh obviously yeah we like hearing from Yan I don't know. I don't know. I mean, Jan has stayed at Meta and previously Facebook for many years uh in through thick and thin. He's been one of the company's biggest defenders. So, I will say it doesn't seem like he can be happy about this. I mean, this is really his division. So, >> what do you think is happening here? I mean, I think this is just a very weird but also not unconventional complete like, you know, organizational refresh. They bring in new leadership. They bring in a bunch of new people, get rid of the old people. Mark Zuckerberg was clearly not happy with their progress and where they were before and he decided he wanted to do something about it. I felt like that detail that they concluded it had become overly bureaucratic was definitely very strategically leaked because I saw those words everywhere and so like they >> it's got to be a PR person on the phone. >> Yeah. Overly and it sounds good and it's like >> it's also like it's according to two people familiar with the matter I think. So, it's like you just get two I mean I don't know this is probably not what happened but I'm just envisioning like two PR people coming on the phone and saying it and being like you can say multiple sources told you this and then you >> and I mean they very clearly chose those words and then it starts to paint this picture and then everyone is like oh yeah like slow bloated um you know like no one likes overly bureaucratic things so suddenly it kind of now you're bringing the like hard charging Alexander Wang in the TBD group and like TBD lab. So I mean to me this is actually kind of like standard organizational politics. I think it just happens to be at a very large scale in terms of salaries and importance but otherwise it's not that surprising to me. >> Yeah. You could call it do department of Facebook efficiency. >> Dome dome department efficiency. Yeah. >> Don't you think you'd want all the AI? I mean, I get it, but you just spent like how much money? >> Can't be that big of a cost. You're still highly profitable. >> Wouldn't you just want all the AI talent you could get? >> It's a good point that like you would think just hoard I mean hoarding resources from competitors is certainly something our friends at Meta are are great at. So like I think uh it I agree it is actually kind of surprising that they weren't just say saying but that actually kind of I think is reflective that it must have been very political because otherwise it would be a no-brainer that they just kind of keep paying people and and make sure they don't go to other places. So that means that it must have gone down enough that they're like we need to get rid of a num number of people. What do you think about this move? In a sign of the escalating competition in AI, Meta said Saturday uh it would cut off access to non-meta chat bots like OpenAI's Chat GPT on WhatsApp beginning next year. This means WhatsApp's 3 billion users would no longer be able to use Chat GPT uh in the messaging app. I mean, one thing that's interesting is all these companies that profess that they're dedicated to openness and the advancement of the technology, like second gets real, they're like, "Yeah, not that open any." >> Wait, wait, sorry. I was a bit confused of this story. It said Meta is shutting off 1 800 chat GPT within WhatsApp. So, does that mean there is a large number of people like actually toll-free calling >> chat? I remember when this came out, we kind of joked about it, but is this one of those weird things that could actually have insane usage that we're just missing? Like people are actually toll-free calling chat GPT on WhatsApp. >> Yeah, I think that uh I could be wrong about this, but you could also just chat with it. Like it's a phone number you can also chat with, but I do think it's a very >> Kevin Kevin while specified shutting off 1800 chat GPT. >> Right. Right. That's what I'm saying. I think it's a phone number that you could chat with, but but maybe not. All right, we're going to have to look into this. I'm going to be calling chat GPT and then flirting and falling in love with them, but not on WhatsApp anymore. >> Trying to get it here. But um yeah, apparently uh it's going to be one step more difficult to fall in love, which is good because we we do know that according to the scientific research, when the bot plays hard to get, it actually leads to more deep relationships. >> That's what they're doing. Thank you, Mark, for at least adding a bit of friction into my flirting with Chie BT. >> That's right. Because now now my heart will flutter even more. Okay, talk about this this Reddit thing briefly and then we'll bring it home with Clippy. >> Yeah, just I wanted to we've talked a lot about kind of copyright. I think this is going to continue to be a story this year. Basically Reddit in a kind of baller move, they're suing a number of providers including Perplexity for uh like you know scraping Reddit data in unauthorized ways. And what Perplexity was doing is you know like if you ever search something in Google and add Reddit at the end, you'll get some amount of information in the Google search results without even going to Reddit. Um basically scraping those results. a a lot of companies have been doing as a service and selling to OpenAI Meta and Perplexity. But Reddit basically tested they had a a fake post that was only accessible in Google search results but was not accessible even by going on to Reddit or any other way on the internet of being accessed and it was showed up in perplexity results meaning they were still scraping Google for Reddit results and incorporating it into their own system. So, so I think to me Reddit is going to play this really interesting role, I think, because they do have one of the most valuable data sets on the internet for LLMs. Um, and they're kind of like a good upstart. I mean, they're a pretty big company, but they're still like kind of like if them battling the giants, I think is going to land well. So, like the New York Times suing OpenAI, I don't know. I don't think that kind of draws as much energy as Reddit saying it's our turn time to fight back. So, and then fighting back in pretty creative ways. >> I I think this will make me sound more cynical than I actually am. U but I think it's worth just pointing out. Uh what do you expect from an industry that is built entirely off of taking people's content without permission? Like it's sort of the fundamental thing at the base of this. >> I know. >> I'm not I'm not dismissing them. I'm just saying that like, you know, it doesn't surprise me. And like if someone's telling you don't crawl something, just don't crawl it. It's kind of kind of disgusting. Then go for it. >> Well, but that's why the way they approach this and tricked them or kind of caught them. That's why it shows that they know how to play this battle like or play fight this battle like >> so good. I mean, it has to happen. And you would think they would lose their you' think they would lose their um uh gall and to do this after Anthropic got hit with that massive lawsuit that could cost it a billion dollars. >> Yeah, but no one paid a billion yet. I think it until there's like an exorbitant cost to any of these companies, nothing's going to happen. >> No, you're totally right. Sad. All right, let's go. Let's move uh to an uplifting story to send everybody off this weekend. Uh this is from Mashable. Microsoft resurrects Clippy back from the dead after 21 years and the internet is wilding. Um, so basically if you use Mo, which is it's interesting. It's this animated face uh from Copilot. Here's how Mustafa Sullean describes it. It's expressive, customizable, and warm. The optional visual pre visual presence listen reacts and even changes colors to reflect your interactions making voice conversations with the bot feel more natural. Mo shows support through animation and expression creating a friendly and engaging experience. So that's mo I think it's interesting. But if you poke it enough times it will turn into clippy again. So clippy is back. >> Wait, what do you mean it turns into clippy? >> Like >> it's like an Easter egg. Yes, Easter egg. So, if you have Mo on your phone and you just keep tapping Mo's face, apparently it has enough of it and it becomes the great paperclip uh known as Clippy. >> And if you anger it enough, it will become racist like the Microsoft chatbot Tay if you remember the >> get spicy like >> destroy your relationship. >> All of the above. Microsoft Mo I think might have a dark side. I think Mo, keep an eye on them. >> We brought it up at the beginning of the show, but as I think about it more and more, Mo is just the best bot name. It's so good. >> It's so good. >> Yeah. >> It's sensible. It like makes sense. It's logical. >> Michael. And also I like the representation of AI being this kind of floating cartoony sunny yellow little character rather than you know like the chat GPT the the blue ball that kind of like hovers and floats if you use voice mode. Like I think Mo is a little cuter. Mo is a little more >> Oh my god. Just don't flirt with Micah though. It's a little cartoon character. Stay away. Stay away people. Well, Mustafa did tell me that like he believes that all AI is going to differentiate on the basis of personality. So, the fact that Microsoft has made this move is not surprising to me. And we'll see if he's right. >> But let's I'm going to go I'm going to go chat with Mo a bit because just being Microsoft, I'm curious how uh how authentically engaging Michael is as a character because they haven't had the best track record on this. though. >> That's true. Well, uh, maybe one day we will see Mo give away chicken sandwiches and then assert its rightful place at the top of the app store. >> That's our That's our future. And none of us will have jobs. So, we'll have to The only way you can get your food is by downloading an app. >> No, that and you have to beg Mo. >> And you have to beg Michael. >> You have to wait for it. You'll be like, "Micha, can I get chicken?" And it's like, I didn't hear the desperation in your voice. Michael, >> download this separate app and say please. >> Here's your tickets. Here's your tickets for your chicken. >> The more desperate you are, the more buns and patties you get. So, >> yeah, >> it's like, ah, that was, you know, it was only good for four buns and two patties. >> Welcome to our future, folks. >> Welcome to our future. All right, Ronjan. Good times. >> You know what? I can't tell if I'm enthused or depressed, but I'm happy that we talked. >> All this conversation, the only thing I can say is I think I'm going to go to Dave's Hot Chicken this weekend. Hopefully after the crowd's died down. I don't think I think I missed the window for the free sliders, but you know what? I'm employed. I'll pay my 12 or $13 for a couple of pieces of that Dave's Hot. >> That's right. All right. Well, Ron, great talking with you as always. We'll leave we'll leave it there. How about that? We'll come back next week and see how the sandwich was. >> See you next week. >> All right, everybody. See you, Ronan, and see you all. I'll be back on Wednesday with a conversation with uh Medium CEO Tony Stubble about how AI is going to impact the future of writing. Uh, also looks like I might be able to air that conversation I had with Yosi Tas uh from Google. So, that would be coming on Sunday or Monday. So, stay tuned. And of course, Ron and I will be back next Friday for a very spooky edition of Big Technology Podcast on Halloween. Thanks again and we'll see you next time on Big Technology Podcast.