AI Keynote: Six Bold Predictions About The Future Of Artificial Intelligence
Channel: Alex Kantrowitz
Published at: 2023-11-23
YouTube video id: ayxxzzxO1Yg
Source: https://www.youtube.com/watch?v=ayxxzzxO1Yg
I'm going to feed off your energy so let's go back and forth here my name is Alex canitz I'm the host of Big technology podcast and I also write big technology you can find it at Big technology.com I speak with people in the AI field every single day and based off of those conversations I'm about to take you through what I think is about to happen in the AI world over the next year so you all remember last year November 2022 huge moment chat GPT is released and as you can see from the headlines on the screen nobody really had any idea what this thing was what it was going to become what were the benefits what were the liabilities well fast forward two months from then and Chachi PT becomes the fastest growing consumer product in history more than a 100 million people in two months use that thing and all of a sudden we're like wait a second something's happening here today you search Google will generative AI right with the whole field that chat GPT has ushered in and you have so many questions will generative AI a lot of people are worried about jobs we'll cover that but still a year in we're all here because we're trying to figure out what this is going to amount to let's get to our predictions my first prediction is that open AI which is the leader in the field today is going to lose the throne that's right open AI will lose the throne it will still factor in but it's not going to be the leader in the field that it is today now I see lot of skeptical looks in the audience so allow me to make the case first things first open AI did what it did with an amazing strength but that strength is also vulnerability and what I'm saying is that open AI took publicly available models the Transformer mod model in particular that was invented within Google and made the world's most popular product right it said we're going to use this model and we're going to make a chatbot that takes this pre-existing technology and turns it into something people are going to want want to use that's great but the problem is that now a year later all these other companies have taken this publicly available data and said we're going to make our own too so now instead of just having gp4 standing alone as a model you have anthropic which was founded by exop aai employees and by the way they didn't steal IP you have Google's in the market and they're researching what everybody says is going to be an even better mod than GPT 4 in Gemini so stay tuned for that Facebook is in the market with llama 2 and then you also have Elon Musk coming in with grock right I I think all of the um models on the screen with the exception of grock are going to have a a chance of really making some waves now and I'm sorry Elon Musk but what happens when this technology is commoditized prices come down and uh event after event we see open AI talk about how everything's getting cheaper people are going to build more and yes it's exciting When You're Building Technology and it's good for everybody building but unfortunately when the prices come down that means for open AI that the models compress and when the models compress you become a less powerful company now listen you've had the models commoditized the models the the margins compress and what does that mean it means that open AI is in the hits business they have to not only sit on Chachi PT and sit on Dolly 3 and make millions and billions of dollars based off of these products and the models underlying them but they have to continually be building new models and new products in order to stay on top they actually had one that they tried to release earlier this year and it never saw the light of day and when you're hit in the hits business taking money in and pushing hits out is never so easy as people in Hollywood right you're not going to get an amazing film just because you're spending money on it and I know there are probably some fans of the new Amazon Lord of the Rings uh uh series that they have on Prime but that's a prime example the other thing that open AI is facing is the open source threat what you're looking at on the screen is a Google Engineers words and this person said listen we have our own models but the open-source movement is growing so quickly that we have no mo this Google engineer was honest about Google and they were honest about open Ai and why is there no mode there's a three-step process that most companies are building when it comes to building with generative AI the first is you start with open AI why open AI has the models that are easiest to use they have apis they are well known and they've been doing this for longer than anybody else but then something happens as you grow you go from the plug- andplay solution to creating a product that you can put any model in and it will work exactly the same and that's what Michael mcnano of VC at Lightspeed Venture Partners is telling me in a recent edition of big technology podcast so you go from the API and then you move to a way where any model plugs in because one model might be better for your product you don't want to be wedded to one company's model forever and then finally you go to open source because open source allows you to customize uh in a way that you can't by just plugging in to an API oh and then there's this guy now Mark Zuckerberg saw that Sam Alman and open AI built the uh fastest growing consumer product in history in chat GPT there's one thing having spoken with Mark Zuckerberg that I know he doesn't like and that's when other consu other companies build consumer products that do better than him and you're I see some people laughing but get in the cage with Mark Zuckerberg and then we'll see what happens right this guy does not like being in fights that he thinks he might lose or or he might uh you know take a punch in the face and therefore he's going after open AI with that open source model that I spoke about earlier llama 2 which is the leader in open source right now and some more chatbots that we're going to talk about in a moment finally we're going to move uh not finally we're getting ahead of myself we're going to move from Big models to smaller models big models inefficient smaller models efficient we're going to go from Arnold Schwarzenegger models to Danny DeVito models great movie by the way and then finally what you see on screen is a screenshot from open AI developer day that by the way I'm familiar with the format it's called a broh hug and it's happening between Sam Alman the CEO of openai and saan Adella the CEO of Microsoft something happens when you broh hug with a big Tech CEO it means that you are much more likely to work with them and much less likely to work with their competitors not to mention the fact that Microsoft owns 49% of open Ai and so that means that if you're a developer you're you to work with open AI you're probably going to want to use Azure but there are so many different Cloud platforms including the leader from Amazon that you're not going to be able to work with as easily now it's potentially a benefit to open AI because this uh uh sync with Microsoft could end up leading them to a place where they're able to build better models that perform better because they're this close but let me tell you something the best way to win in Tech is to be everywhere and open AI they might be everywhere but it won't be as easy to work with them because they're so wetted to Microsoft maybe not an immediate problem but watch this is definitely something that I think you need to pay attention to all right we're still in prediction one so let's sum it up the four problem or six problems that open AI is running into model commoditization margin compression they're in the hits business there's an open- Source threat we're moving to smaller models and they're just a bit too close to Microsoft I see some cameras so at least you know it seems like there's some some response to this argument maybe a little bit different than when you saw Joffrey on the screen 5 Seconds earlier so who's going to win who's going to benefit here look this is the old ancient meme of Jeff Bezos as the original book seller saying yeah you know I I sell books on the internet and then Jeff Bezos becomes this multi-billionaire and he says I sell whatever the f I want that's the evolution of Microsoft my perspective here is that Amazon will now become the everything store for AI models and that is going to put it in the lead of the arms race does that mean it's going to win the AI arms race I don't think so we don't know yet but is there going to be a moment where everybody looks and says wow Amazon is the leader of this race yes I definitely think that's coming why is the case first of all Amazon has a bring your own model approach instead of being wedded to an open AI Amazon says listen what I want is for you to work on Amazon web services it doesn't matter what you use because remember we talked about margin compression in the beginning of this Amazon's going to make its money not by using its own llm although it has one it's going to make its money when you're uh when you're using AWS to build your own products and as an ancient Amazon proverb goes no matter what model you use we will get paid and in fact that that proverb comes from an interview that I did with Matt Wood who's the VP of Technology at Amazon web services let me just read it to you he says we get paid by providing compute capacity to actually do the model training and for providing the access to the large amounts of storage that are needed to store the data and then get that data into the machine learning models do you hear this everywhere along the chain Amazon is getting paid does that exist for open Ai No and that's why I think Amazon is going to be the leader oh and by the way the research advantage that others have well Microsoft is starting to come at par because uh earlier this year here actually just a few weeks ago they invested $1.25 billion in anthropic which is that uh open AI competitor started by openai expats and they're GNA it's going to give Amazon this hook into Cutting Edge research that it wouldn't have had previously finally this is another screenshot from open ai's demo day this is Sam Alman open AI CEO standing in front of six different phone screens representing six different voices that you can interact with to talk to Chad GPT what this is showing is that open aai believes that we're not just going to interact with AI via text we are going to interact with them via voice who does that benefit it benefits Amazon Amazon has 500 million Alexa connected devices uh in homes across the world if this is the way that we're going to interact with AI models then actually Amazon already has the hardware in our homes ready to deploy employ it you see a s you see a slide like this from Sam Alman at open aai developer day you start questioning whether that company can lead forever and that's why again I think Amazon is best positioned to lead this race at least for a moment remember this is the company that loves artificial intelligence and loves robotics so much that Jeff Bezos taped his freaking Echo onto a Roomba okay they're not going to lie down and let others take this uh take this field it's just not going to happen Clearing House for AI models anthropic partnership 500 million Alexa devices take it to the bank Amazon is going to lead here all right now this is a great excuse for me to show some of my favorite nature photos I'm I am predicting that we're going to move from broad to narrow use cases of AI now this first photo here is a photo of me at the Grand Canyon and the Grand Canyon is a sight to behold it's wondrous one problem you have no idea where the hell to go because it's so open and vast second also in Utah is a narrow Canyon narrow canyons are little little less Grand and they're a little less breathtaking but you know exactly where to go when you're walking through a narrow Canyon and it's similar with AI Chachi PT just takes you in this broad area you open up that Chachi PT window and you look at it and you say oh I know this is wondrous I know this is cool I know there are a lot of different directions to go in but damn I have no idea where to take it we're going to go to narrow models we're going to go to an Era where you're going to start using purpose-built AI models and that's again why in open AI developer day you had Sam Alman standing in front of dozens of personal gpts that are purpose-built to make certain tasks easier for you with generative AI broad to narrow the and and just think about the advantages versus the disadvantages broad uh you're inefficient because you're basically quering the world's knowledge every time you talk to this thing it hallucinates cuz it's just handling a lot of information and there are unlimited use cases small Bots they're efficient they're train they're working on less information because they know what you're probably going to do with them they have specialized training so people will upload information inside these Bots and they're much less likely to hallucinate because they know exactly where you're going with this and they're going to rely on that information base and there's clear use cases you're in there you know what to do you're not saying I know this thing is cool but I wonder where I should go with it we're already starting to see this in practice with legal Bots like Harvey which law firms are using to query legal material go through Discovery and ultimately make their lawyers more efficient and then the other the other uh image that I have on the screen here is the 28 different Bots that meta introduced in WhatsApp and messenger in their most recent roll out again let's go back to the theme Mark Zuckerberg wants to kick the crap out of Sam Alman and open AI he's not doing it with one bot he's doing it with many Bots and if you thought the Elon Musk Zuckerberg cage match was the main event no that was a sideshow him vers Sam Alman that's where the real fight is going to happen and he's betting on narrow and you can already see the data start to come out that proves my point that shows you that something like chat PT just is not as serious of interest to people as people thought was going to be I mean look at this data sessions down 29% three straight months of declining usage in the summer uh bleeding users says giz Moto now listen whenever I share this data people tell me well it's because students were out of school just wait till the students come back to school and it's going to be a different picture really so you're telling me that your technology shrinks when school's at a session and it grows when it's not that's ridiculous it has to be something that's broadly applicable and available and and that's going to be when we move to these narrow use cases students love it because they can type in any coursework it's just not a model for a product that has long-term success if we don't reach General uh generative uh sorry General artificial intelligence if we create AI That's on par with human intelligence then this all all beds are off but until we get there bed on the narrow anybody here see the movie free guy all right we got some hands it's a great freaking movie let me tell you what happens Ryan Reynolds is a non-playable character in a video game okay a non-playable character is also known by the acronym NPC and what that is it's the people that kind of walk around in the game they have no they have no dialogue they're not there to really interact with you you can punch him in the face sometimes and that's about it and in this movie Ryan Reynolds the free guy wakes up and he becomes sentient and he's like wait a second you know he starts to express the full range of human emotion and I believe that that we're not going to quite get there but we're going to get pretty damn close to free guy when we build llms into video game technology and that leads me to prediction four is that video games are going to be absolutely incredible with this stuff and I think when they get there we're going to move from an era where we pulverize MPCs to we empathize with them instead of running them over with a car we're going to have a conversation with them they might even be our friends and when we're working in these virtual rule these Virtual Worlds we're just going to get to a point where a video game is going to feel so much more like real life and that's going to make them freaking amazing this is not something that's coming out of left field I spoke with the founding product manager of Lambda which is Google's original chatbot and he said to me that they were exploring some use cases around how a technology like that could be used for powering NPCs within games the smartest Minds built the original Technology based off of large language models inside Google inside Google they never shipped it but they did come up with this use case and you better believe that these companies and video game developers right now are having conversations about how to make video games something that even non-gamers might find fascinating let me end this prediction section with a quote from the free guy he says I may not be real but for a second there I felt pretty alive um I wanted to put this in here a because I believe that every presentation should have as many images of Ryan Reynolds as possible but B because I do think that we're going to get to a place where these Bots or these NPCs won't exactly wake up but they're going to feel real to us Okay I lied one last thing screw the critics the critics believe that free gu is only 80% on Rotten Tomatoes wisdom of the crowds us the audience we say 94% if you haven't seen it go see it it's my dad's favorite movie okay prediction five that we are about to see the world world's greatest copyright battle greatest copyright battle last week at open ai's developer day Sam Alman said that if you get sued C to certain customers if you get sued for copyright infringement we are going to pick up your legal costs what he's saying effectively is like please go use other people's copyrighted material and we'll we'll pick up the bill if that gets if that turns out poorly for you we also have andreon harwitz who said recently that if you train I'm sorry training on copyrighted material makes their Investments worth billions of dollars more than if they couldn't use it and they were deploring the copyright office please let us train on copyrighted material this is something that the AI movement needs to train on but there's another party here and that is the people producing the copyrighted material and let me just direct your attention to a recent article that showed that the Associated Press did reach a deal with open aai to allow them to train on their material by the way remember the whole broad use case right if you look at these at these interactions you're going to want to know what am I supposed to do with them news is amazing to solve that problem it can say hey these stories are breaking would you like to know more people are going to chat with them extremely extremely valuable to these developers and that's why open AI is starting to pay some of them sorry extremely yeah extremely valuable to developers that's why open AI is starting to pay some Publishers with the AP first to on them but they're not going to pay everyone and what happens when they don't pay everyone they're going to get sued and you already start to see that there are some Rumblings from The New York Times saying you're training on my stories now give me my money and let me remind you of a famous journalism proverb which is take my work and I'm going to sue you and I think you're going to start to see many many lawsuits and the courts are about to basically determine the future of this very important new wave of Technology ology by the way not immune to me like I said I publish on substack I was going through my settings I never knew this existed and recently I Came Upon This toggle that was op that had opted me in to letting Google and openai and other bots of a similar nature train on my material now as you can see this is a live shot from my substack dashboard I'm allowing them to train but more and more questions are going to come up about this and people are going to try to figure out how we get paid for helping these models uh uh live their potential by the way I love the Innovation I want to see these models succeed but there should be a fair value transaction andri Horowitz is telling us that this information they're training on is worth billions of dollars but you tell me that we can't get compensation for that it's not going to fly for me and I don't think it's flying for many others all right bottom line AI winter is oh sorry AI copyright winter is coming coming and the way this play plays out might determine whether we head into another AI winter itself finally good news on the horizon we're keeping our jobs people let's hear it for our jobs yeah who booed jobs don't Boo the jobs no but this is great I think the only way that we're going to see Consultants tell us that we are redundant is going to because going to be because our companies are failing not because AI has replaced us and is good enough to do our work why is that I personally have four reasons first of all AI makes us better at our jobs if you're a boss and your worker comes in and says I'm now 20 30 40% better at my job who in their right mind says great you're fired it just doesn't happen and so we're going to see people get better at their jobs bosses always want people to be better at their work and so they're going to keep them it's actually a retention policy for workers around the world as opposed to something different and on top of that the Bots can only handle a wide range of tasks you can use them for instance to get you information from a document but whose job is only to get them get information from a document and hand it to their boss even research assistants need to take meetings they need to figure out what the task is they need to go deep in the subject matter they need to be experts on the stuff going on uh previously so yes they can have some conversations with these Bots but it's not replacing them also change management is hard anyone that's been involved in change management inside an organiz ation knows this to be the case and then finally AI job loss is something that we're seeing from companies that are weak anyway they're going to have to do layoffs anyway and they go to the market and they say listen we have to do these job cuts and we're replacing our employees with their with AI oftentimes no they're not they're just trying to get positive Market momentum because they're going to be seen as a forward looking company it's just not the case okay be very very skeptical when a company says we're replacing our employees with AI generally what that means is we are weak and we're trying to find some excuses and a positive narrative all right really important stuff here because you're seeing it more and more this is a quote from the uh head of AI at the Mayo Clinic lab he's a neuro radiologist so everybody remember when um AI started making its way into computer vision and um they said it was going to replace Radiologists because it can read scans better than they can well that's not exactly happening uh Mayo Clinic has 11 AI models functioning in the clinic and they're still looking to hire Radiologists so again let's just be very skeptical of this narrative and you can hold me to it I was at uh uh I wrote about this in the Boston Globe in slate on big technology on my show we do account accountability when we get things wrong and we talk about it and so if you end up listening and you see that I'm wrong here write in but I I'll hold myself accountable and I hope hope you do too all right one bonus prediction the AI is not going to kill us okay not happening all right six b six B predictions on the future of AI let me recap them for you open AI loses the throne Amazon will lead the AI arms race we're going to move from broad to narrow use cases AI is going to make video games absolutely incredible the world's biggest copyright battle is on its way and God bless we're keeping our jobs thank you very much everyone you've been an amazing audience take a photo follow the podcast follow the newsletter and I hope you have a terrific rest of your websummit thank you for joining and have a great day