저자: BillGross · 외국인
요약 보기
AI는 인터넷·클라우드·생성형AI로 지식 접근 비용이 0에 수렴하며 역대급 변화를 일으키고 있음. 2025년 AI 투자는 하루 10억 달러 규모, R&D 중 AI 비중 1%→20%. 뉴럴넷 레이어 수 비교 시 챗GPT≈쥐 수준, 인간 뇌 약 1000층으로 이번 10년 내 초인간 수준 가능성 언급. AI는 긍정·부정 결과가 공존하므로 저작권 침해 등 부정적 결과를 줄이는 데 기여해야 함. 저작권 문제 해결을 위해 AI 사용량을 측정해 콘텐츠 제작자와 매출 50/50 분배하는 Prorata 시스템 소개.
다듬은 글
thank you very much it's a real honor to be here we are living in unbelievable times imagine that we discovered an island with a billion phds on it who were willing to work all day long for the cost of power that would previously have been unimaginable but I think we will actually see something like that this decade and that's because we're seeing something like we've never seen before so how do we take this unbelievable tool of Ai and make the theme of the conference a reality a positive future well if I look back on all disruptions that I've seen in my lifetime I really feel like this will dwarf anything we've ever seen while I'm covering these four that impacted me try and think of the things that you saw in your lifetime where when you saw them you felt the world will never be the same again the first one of those for me was when I saw an IBM PC came out in 1981 I went down to my computer land store in Pasadena bought a $5,000 machine with two floppy discs and 256k of ram but I really knew nothing would be the same after that it completely democratize Computing and the next one of those for me was the Netscape product in 1994 Netscape went public in 1995 with 30 million users think about how small 30 million users is today but that was unbelievable that 30 million users could connect with one another it was a dramatic change it completely disrupted and democratized communication then the iPhone mobile Computing each of these things changed the world so dramatically this now affected billions of people not just tens of millions of people like the Netscape product but each of these things completely was a turning point but now for me the last one the day I saw chat PT in November of 2022 even the simple thing that you could do to just have it spit out a poem or write a theme was so incredible even though some people called it a stochastic parrot it still was very very very impressive and it's going to lead to so much more change in just the last few years and in the coming few years than we've ever seen before the reason these disruptive changes are so big is because whenever the cost of something approaches zero everything's going to be dramatically impacted with the internet the cost of distribution approach zero with the cloud the cost of storage approach zero and with generative AI the cost of accessing and maybe even creating knowledge is approaching zero and that's why it's so transformative and the rate at which it's happening is staggering too you've all seen this but just put some numbers to it last year there was a billion dollars of investment every single day in AI a billion dollars every day that's the pace of this revolution AI is going faster than the internet or personal computers went right now you can see over on the right the rate of adoption to get to 40% of all people using a product and it's way way faster happening in only two years and in 2020 1% of global spending on R&D was for AI last year it was 20% 20% of all spending on R&D was toward AI so that's why this adoption rate is so fast and why it's going to be so transformative but in what ways will AI impact us I broken it down to these three categories with the help of Roy bahad from Bloomberg the first category is like looms which automate a person's work so how would that relate to AI take some repetitive task like data entry LOM AI could replace that the way a loom replaced knitting or sewing slide rules are another thing that advanced human capabilities by making a person faster how will AI be like that well that will be AI like GitHub co-pilot helping people code faster because all of a sudden you have this tool by your side that help you like a slide rule did for math computations but the final category of AI disruptions are like cranes and those extend what a person can do a crane allowed humans to do things they couldn't even do before without that power well the AI analogy to that is what Alpha did with protein Discovery and there's many many other things that AI is enabling us to do that a human couldn't even do before so now going back to the few personal aha moments that I had with AI the first one I told you was when I saw chat GPT putting out text that almost fooled me that a human was behind it but I knew it was just a sarcastic parrot just having lots of statistics on Words and putting together sentences that made sense even if there were hallucinations but it became much more than that in the last few years Alpha fold was another example where you could predict every protein with AI but then I saw something that completely changed my mind it blew my mind it was just last month actually I was giving a presentation at the fortune brainstorm AI conference in San Francisco and I was practicing my talk so I had recorded my talk my rehearsal in otter and I had a very crude transcript of it with even with mistakes but I took that entire transcript and just pasted into chat GPT and said can you give me any suggestions on how to improve this talk and it came back with 10 bullet point lists of detailed items that I could do to improve my talk I have the first two of them there where they come back and say the opening provides a good sense of historic disruptive moments but you need to get to your main point sooner and I was like how the heck did it figure that out how the heck from looking at my text which it had never seen before this was not a sarcastic parot anymore this was doing some analysis of something new that I handed it and that completely blew my mind it showed how reasoning was beginning to emerge much more than just probabilities of words and it's not just me this is Steven Johnson an author who did the same thing very similar thing I saw him post this on Twitter where he said the best example I've seen of L ofm seeming to understand was the experiment that he used which was he took a manuscript of his book which was not yet published so no one had ever seen it before there's no way chat GP could have crawled it or anything like that and he asked it can you identify some key places where I deliberately withheld information to create suspense for the user and it found them so it read this book that it's never seen before and it understood it the way a human would understand suspense and identifi those locations so it's really really powerful where we're going to see as this progresses is unbelievable so how does this do this well neural networks worked neural networks sort of invented 50 years ago were very very simple I'm showing a five layer neural network right here with a bunch of connections where you give some input stimulus and some outputs and you get some weightings and you build this up a five layer neural network can't do very much and many people abandoned neural networks and didn't think they would work that powerfully until recently when the computing power that could be brought to it was so large that the neural network got bigger now you see this five layer neural network and you see the connections as you add more layers the connections grow super exponentially so it's hard to make something that can handle large neural networks but now we can so how do you compare this to other forms of brains or thinking well a fruit fly has about a 5 to 10 layer neural network equivalent uh Chachi PT has 126 layers so imagine that five layer thing expanded to 126 layers with all the billions of connections that occur when you give it lots of information and have it form those connections it actually does lead to something pretty astonishing now a rat has between 100 and 200 layers so maybe where we are right now is at the level of a rat that's where chat GPT is which is mind-boggling actually but look at where we're going to go our brain has approximately a thousand layers so dwarfing what Chachi can can do today but everything is growing exponentially we're getting more and more compute power Nvidia just announced new Chips just last week at CES in Las Vegas it's really really moving fast so it's not inconceivable that we'll have multiple thousand layer like neural networks that can cooperate together and have superh humanlike capability and I actually think we will have that this decade so that's a opportunity and a challenge as you heard from Tristan earlier there are a lot of challenges when you have that much power I want to talk about that I think that because of this AI might actually be underhyped with all the hype that's around it it might actually underhyped but both the good and the bad I think we'll have ai smarter than Nobel Prize winners this decade I think we'll have ai even this year that can work with your mouse your screen your keyboard and interpret things that you're doing and feedback and take actions on your behalf I think we'll have large AI data centers which we already have but they're getting bigger and bigger and more powerful that will act like a country of geniuses like that country of PhD or that island of phds I alluded to at the very beginning but where is the challenge here well every new disruptive technology brings unintended consequences when radio came out there was a huge positive of instant mass communication but there were unintended negative consequences propaganda tool for World War I World War II broadcasts were used in very negative ways to take advantage of the negative power of radio nuclear power unbelievable energy source but least to nuclear weapons nuclear accidents there's so many things that in fact almost every technology has positives and negatives fossil fuels might be one of the biggest ones it lifted so many people out of poverty it gives people warmth Comfort convenience travel maybe all of most of us are here because of fossil fuels but we've had Wars over en energy resources climate change weather disasters I just flew in from Los Angeles where we've had those horrific fires the fires happen to happen The Identical week that the planet crossed 1.5 degre sea of warming so it's unbelievable the unintended consequences of fossil fuels no one thought when people started powering the planet with fossil fuels that we'd have all these negative consequences as well as the positives so what about AI I tried to put together a list of AI consequences in three categories I listed some of the obvious positives education and cures and Innovation possibly climate correction productivity easing tasks leisure time there's many many positives of AI I listed some of the negatives copyright theft misinformation bias you heard a lot of these same things from Tristan earlier I agree with all of the things that he's working on and then unended negative consequences things that you might not even think about like the Rogue AI or pollution that comes from AI power the power required for AI obviously propaganda but bioweapons and other things that might be accidental unintended consequences of this powerful AI so I feel many people are working on the green column the green column is powered by capitalism that's going to get taken care of I don't think we have to worry about the column on the left what I think we need to worry about if we want to have a positive future is worry on the other two columns so I'm really urging people to look at the other two columns and say where can you make a contribution that could possibly mitigate the red column in the middle or the yellow column on the right so I took a look at that list and said where I where could I do something and I I focused on one thing copyright theft I said I think I could do something to manage copyright theft and in fact this started a year ago a year and a few days ago the New York Times suit open Ai and I came to dld in January last year I started thinking about how could we make a system that would be more fair and not have copyright theft not have shoplifting of content in AI human creativity is an unbelievable resource that should not be squandered by AI systems that don't give compensation to the human creativity behind it and that human creativity takes every form it could be words from journalism and books it could be images animation movies music all of those categories I feel need compensation so I set out to do something about that I took a look at the industry problem where generative AI is powerful and Unstoppable but it's stealing content I really feel it's stealing content from content creators so I came up with a system I'm calling prata I'll tell you how that one works that measures the usage of content in Ai and then shares Revenue 50/50 with the creators who made that content and I saw this funny tweet on Twitter uh this person wrote I have been told to stop stealing muffins from the bakery unfortunately it's the only way to keep my lucrative muffin stand in business everyone is fine with this in in fact investors love this because I have no cost so it's incredible how this was meant as a joke but this is actually what some of the AI companies are actually saying if you take a look at some of the quotes on the left here Mustafa from Microsoft says everything on the internet is free and I can use it in my a models and Sam Alman says we actually need copyright materials for free to make our business model work and I just think that's ridiculous there are many many businesses that make their business model work thank you it is absolutely ridiculous that you have to get something else from free from other people's work to make your business work and of course this has led to lawsuits the New York Times has sued open AI News Corp has sued multiple other companies Universal Music Group has sued many companies that are stealing music as well and in fact it came out uh the New York Times read this article that even the open AI researcher who left admitted that they were stealing copyrighted works so there was no hiding of that in fact even this week it came out Mark Zuckerberg was brought a whole list of pirated books that that were stolen on a Russian site and they brought them to him and they said should we use these in our llama system and he said yes and he said but scrub out the copyright information and the ISDN numbers take that out so it's just ridiculous that the companies are doing this and getting away with it and I think this copyright infringement really has to stop so why is it possible in this area and it's not in the other areas so I'm covering some other areas where people create content and they are compensated in the App Store in the Google Play Store on Spotify Spotify took in $16 billion of Revenue last year paid out $12 billion to artists still is a profitable company and it's 100 billion doll market cap so it's completely possible to pay money to creators and still make a viable business YouTube YouTube is a half a trillion dollar Company YouTube started out by pirating copyrighted material and many people gave them takedown notices and it was a cat and mouse game for the first year or two of YouTube's existence then YouTube started sharing Revenue 50/50 with creators and it EXP loaded their business actually got better when they started sharing Revenue because that now empowered creators and YouTube is now a half a trillion dollar business almost double the size of Netflix because they share revenue of course Netflix shares Revenue with all of its creators Apple music shares Revenue with all its creators so why in all these businesses can you share revenue and not with generative AI well I think the reason is that it's not as easy to count with Spotify you can count the streams with YouTube You Can Count the views with apple music you can count the downloads of course you can count the downloads in the App Store so we set out to try and figure out a way to count what is being used in generative AI now it's trickier because in generative AI what you're getting is a remix you're not getting just one PE person's work you're getting a compilation of multiple people's work so we had to come up with a way to unmix it to be able to take a look at the output of generative AI whether it's a picture or a movie or animation or text and break it down into who wrote that first who came up with those ideas first that are in that as components in that answer and by what percentage and we came up with a way to do that and we reached out and many many Publishers were interested in that so we started at dld last year so this company was actually almost formed at dld one year ago we reached out to Publishers and hundreds of Publishers agreed to participate in this 50-50 system where their content would be compensated we'd now signed 400 Publishers in the first year and now we're launching a search product called gist it's called gist doai it'll be like open AI or like perplexity or like even the the Gemini answers at the top of Google but with one critical difference all of the content that's used in the answers we've been given explicit permission to use it and we're sharing Revenue 50/50 and you can see in the answer at the top there's a attribution bar that says this answer is composed 36% from this Source 25% from this Source 15% from this source and that's how the money is shared because the algorithm figures out exactly who to give credit to and exactly where to share the revenue so we think that this is very powerful not only for text but also for images New York Times also showed this example where they asked AI to create some images and it used copyrighted materials throughout that and we asked meta to create an image of a of a masked superhero and it created this image and you can tell pretty much that there's Spider-Man aspects to it our algorithm re uh re unixes it and says this is 90.3% from Marble and 6.2% from DC Comics and that's where the revenue should be shared to those two entities in that exact ratio so the whole idea of this is to enable Fair attribution we want to give payper use so that people make money every time their content is used and my dream for this would be that we have a planet where anybody with intellectual property in their brain anybody with an idea anybody with creativity can create something register it and whenever it's used in AI engines they'll get a check a royalty check every month for the content that they create that's really what I'm shooting to to make happen with this when you look back on the list of the other things that were on the chart I'm only tackling that one right now the copyright infringement of generative AI but if you look back at that list there's so much opportunity and it really is the best time in history for making a positive impact on the world because almost every aspect of life is going to be touched by AI so there's almost every aspect where you can make a difference with it and your brain power can make that impact and think back to what happened when the printing press was invented in 1450 and the people who grew up with that when they were 10 and 20 years old were cernus and berardo and Michelangelo and even though they didn't write books they were impacted by the societal change of that printing press well all of us are going to be impacted by the societal change of AI and what are the odds that we're alive in this exact decade where that is happening this is the decade right now we're all alive likely Nardo and capern were when the printing press came out so there's hundreds of ways we can make a positive future with AI I want to try and help make that happen if I can help make that happen in any way I'd love to work with you so feel free to reach out to me at any time just Reach Out out to me at Bill pro.
and I'm happy to send this presentation with you as well but reach out to me if you have any ideas and how we can minimize the negative impacts of AI maximize the positive impacts thank you very much you've been a great audience but
Comments 0