 Hello, everyone, and welcome to the 4.30 to 5.00 p.m. session of the 2023 Open Simulator Community Conference. In this session, we are pleased to introduce the presentation, How Generative AI Will Change Content Creation and Coding. Our speaker is Maria Korlov. If you were at the previous session, you know that Maria is a published author and the editor of HyperGrid Business since 2009. Please check out the website at conference.opensimulator.org for speaker bios, details of the sessions, and the full schedule of events. The session is being live streamed and recorded, so if you have questions or comments during the session, you may send tweets to atopensimcc with the hashtag pound OSCC23. Welcome, everyone. Let's begin the session. Thanks, Lear. Thank you for having me. I love being in front of everybody here, in case you guys didn't see me in my previous presentation. So, this is about the stuff that I do outside of OpenSim. So, I've been the editor of HyperGrid Business since 2009, covering OpenSim in virtual worlds, but before that, I've been a technology journalist. I started as a technology journalist during the early days of the dot com thing. I was working on staff for Computer World, and I was covering the digital transformation of the industry back in the days where we had all these AI magazines, not AI magazines, Internet magazines like Business 2.0 and E-Company Now that were several inches thick, and you learned about the Internet by reading print publications, golden age for technology journalism, and I've been covering enterprise technology ever since. So, I'm still writing for Computer World. I write about AI for Computer World now. I write for Network World. I write for CIO Magazine, which is Chief Information Officer, and I write for CSO Magazine, which is Chief Security Officer, and these are the biggest enterprise technology publications on the planet, and I've won the top awards in my industry for my coverage. In fact, I won a national award for my coverage of Whirlpool using self-driving robots in their factories. So, this is my shtick. This is what I cover, and for the last year, it's pretty much been generative AI all day, every day at my job, and I'm talking to CEOs. I'm talking to CIOs and CSOs of enterprises who are trying to figure out what's going on, and I'm talking to experts and the people who are inventing this stuff. I'm going to conferences. I'm going to be a keynote speaker at an AI conference in March. If you happen to be in Peru, I'm going to be speaking in Lima. I'm going to be the keynote speaker there, and I speak at online conferences and all that stuff. So, that is my day job. And it has been an insane, literally an insane year. The stuff that has happened this past year is incredible. It is the fastest pace of change I've ever seen. So, before I go into the AI myths that I'm hearing the most from creative types, I want to just give you a sense of how fast this is changing. So, back when the internet first came out and we had America online, people had to buy modems and computers and get a phone connection to the internet or a cable connection to the internet or get the internet from their work or from their college. It was hard to get online. It was hard to use it. If you wanted to put up a website, you had to learn HTML. There were a lot of barriers to entry, and if you were living overseas, it was even harder to get online. Many, many countries, rural areas did not have internet access. Companies were slow to adopt the technology. Everybody knew they had to do something, but they didn't know what they were doing. They hired a teenage kid to be their webmaster, to set up a website. It was a very slow, gradual transformation. And the thing that the internet hit most was information. So, the transmission of information, the control of information, that was disrupted by the internet. So, if you are an encyclopedia business, you are severely, severely disrupted and you either went out of business or you figured out a way to adapt. Similarly, Genitive AI is going to revolutionize the sector of expertise. If you are an expert in something, if you know how to do something, if you can take the information that's out there and show you how to apply it to your own particular case, if you know how to draw, if you know how to paint, if you know how to write computer code, if you know how to write news articles or how to edit books, all this kind of knowledge the Genitive AI can do. It does it badly. It does it worse than humans in many, many respects. The early internet was also worse than what it replaced. The early internet was worse than encyclopedias. The early internet was worse than physical books. The early internet was worse than your white pages and your yellow pages. It was worse than all the technologies it soon came to replace because it evolved quickly. Genitive AI is also evolving quickly. But it doesn't have any of the barriers to entry that the internet had. When ChatGPT was released a year ago in 10 days on November 30th, 2022, it became the fastest growing app in history. 100 million people from around the world immediately started using it because as long as you have an internet connection, a smartphone, a computer, anything, you can get to ChatGPT. You can use it in pretty much any language, anywhere on the planet. And if you're in Russia and you can't have access to it, you immediately got yourself a VPN so you can log in. It was literally the next day after it came out, all the Russian websites were talking about how to get into it because the advantages of it were so clear for spam and hackers and everything else. It is really, really transformative because everyone on the planet at this point either has a smartphone or knows somebody who has a smartphone and they can access it or borrow it. And everybody can now, almost anybody can set up their own private AI. There are over 200 foundational models out there right now. Competitors to ChatGPT, many of them much smaller, many of them open source that are available. There's some small enough that can run on a phone. There's some small enough that can run on your computer. But with cloud computing, anybody can run even the biggest of these AI foundation models by just renting cloud service space and any company can do it. There are zero barriers to entry in AI. And it is also evolving insanely fast. So usually the technology doubles in capacity every 18 months to two years. I mean, that's been the Moore's Law historical trend. AI is doubling in capacity every few months, if not faster. The latest AI app that I'm using, Claude, can now process 200,000 tokens, which is about 150,000 words, which has two average length books or one epic fantasy book. So you can upload an entire book that you're writing and ask it to identify plot holes, identify your major themes, to identify lack of character motivations. It can do all this already. The pace of change that's happened over the past year has been absolutely insane. So let's talk about some myths that I see creators have. So I'm involved with a bunch of writing groups and there's artists. I've run a science fiction magazine, Metasteller, and we have hundreds of contributors to our publication. And so we have artists and other people on our board who are part of it. And some of the main misconceptions that I'm seeing is that AI is cut and paste. It takes stuff that's out there and just assembles it together like a little jigsaw puzzle. And so it can't produce anything that's original. It's just copying what's out there already. And this is absolutely not true. The way that the generative AIs work is they learn from what they see and then they extrapolate from it. They create new stuff based on the patterns that they've seen. They forget their actual training data. The actual training data is not stored. It's the patterns that they discover that are stored. And in fact, we just saw one of the court cases about copyright infringement. A big part of it was dismissed because the people suing open AI couldn't produce any cases where something was copied exactly from an original artwork or a piece of text. It generates something new. If it overlaps with something that's out there already, it's... That wasn't like the way it's intended. And if it happens anyway, that's one of the AIs that had a problem with this. It was printing actual lyrics to songs. They just added a filter to filter the actual copyrightable stuff out. People are saying it's just autocomplete on steroids that it remembers the relationships between words and just gives you the next word in the sentence that's most likely. And that isn't fact how it used to work. So the old machine learning that we've had for the past 10 years does work like that. You've bought shoes previously on Amazon. Maybe you'll also like to buy these socks. It's a simple straightforward prediction based on past events, predict the future events. The new generative AI takes those patterns between words and then looks at the patterns between those patterns between words. Then looks at the patterns between those patterns between those patterns and then on and on like that. And the final result is pretty close to actual reasonable thinking. And in fact, if you ask an AI a complicated question and you tell it to think it through step by step, it will give you a better response. And if you ask it to go back and reevaluate its thinking and see where it went wrong, it will go back and review its thinking and actually identify the problems where it went wrong. And if you still don't think that AI didn't have like original thought, upload a piece of work that's new that you just wrote. And like for example, myself, I have a whole bunch of unpublished novels. I've got like nine novels finished that are ready to be edited and published. And so I uploaded one of those models and I said, what are the plot holes in this model? And it went in this novel and it went through and found all the plot holes. It was not ought to completing this because that novel didn't exist anywhere on internet. I had just written it. I hadn't published it anywhere. It wasn't repeating what somebody else said about my novel. It was coming up with original criticism of my novel and this is a frightening, a very frightening thing because obviously I make my living writing and editing. So yeah, I'm like really super worried about this. So people are saying that it can't match humans because it's based on human models. And so it can only do as well as the average human. So there's a program called AlphaGo and this is like a couple of years ago now. Google created this. They were the original inventors of this whole generative AI although they didn't do anything with it. And AlphaGo was trained on all the human games and so it could play as good as well as humans could. And then they had AlphaGo play against itself. And the thing about Go is you can't brute force an answer. You can't just think about possible permutations and pick the best one because there's too many of them. It has to understand the patterns, how the game works and it played millions of games against itself and it invented new strategies that humans in their thousands of years of playing Go hadn't come up with. So yes, AI can definitely match and surpass humans in many areas. In fact, in a whole bunch of tests, including creativity tests, AI, these AI models have beaten the average human or even average professional in their field. So if you are in a creative profession and you've been dismissing AI as something that you don't need to worry about for these reasons, those reasons are already not true and they're gonna be even less true with every month as the models improve. All right, so I talked about AlphaGo. So it created training data by playing against itself. AIs can also create training data by interacting with humans. So every time you talk to chat GPT, it learns from that interaction. Every time you tweet something on Twitter, Twitter learns from that interaction and it's gonna train its grok on it. And I'm told that I have to speed up a little bit. Okay, and it could interact with the environment. So it can actually run tests and see what works and what doesn't. So if it wants to learn math, it could actually interact with a calculator and see how good it is at making calculations. So there's a lot of things that an AI can do where it's not limited by training data. So some people think AI will kill art and I like to look at the example of photography. So when photography was first invented, portrait painters were freaking out. Photography isn't real art. All you do is press a button. Literally you point a camera and you press a button. You don't need years of training and how to draw and it makes a perfect picture. This is like the work of the devil. We shouldn't have it. And yes, a bunch of portrait painters lost their jobs as a result. But we still have human portrait painters. The human portrait painters often use photography to create reference images so their clients don't have to sit still for hours. But photography also became its own art form and photography also gave us television and movies and all sorts of magazines and all sorts of other things that would not have been possible without photography. So even though you're just putting your finger on a button and clicking it, even if you're like putting it on a button on a drone and a drone is taking those pictures, it is still considered art and you are the artist of who creates that photograph. So there's legal issues around AI. So on training data is the big one. If you're an artist, Adobe uses only licensed data for its training and it pays its artist. It sent out the first payment in September. It pays people when their artwork is used to generate new art and it's the only big company that does it. Get images also uses only licensed data. So if you're an artist, I recommend that you use Adobe Firefly. That's the one that we use at Metasteller. It's only uses licensed data and it has a free plan and the paid plan starts at five bucks a month. Then the copyright. You cannot copyright AI produced work. So that is currently the standard in the United States. If an AI produces it, you can't copyright it because it's considered not to have human involvement. But you do have to type in a prompt and then often you might give it like a sketch or something to get it started and then you interact with it and modify the results and make it go a little bit more like this or change this part. So there's a lot of human effort already involved. But also with photography, all you do is press a button and you can still own the copyright to that. And with photography, if two people take the same picture, both of them have the copyright to that picture. The Chinese courts have already ruled that you can copyright AI generated stuff. And I have a feeling that because of corporate pressure, we're going to be able to copyright AI generated work very, very quickly. So if you're an artist hoping that this will protect you that you can't copyright AI stuff so you don't have to worry about AI, I don't think it's going to last long. Will it take jobs? It will take jobs, but it will also create jobs. So there's an Ernst and Young study a month ago and they asked a couple of thousand CEOs about if they plan to adopt generative AI, 99% of them said that they would or are already doing it. And the reason they're doing it is because they're already seeing immediate business benefits. It's speeding up computer programming, it's speeding up creating marketing content, it's speeding up all sorts of stuff that corporations do. So they're already seeing business benefit from it. Generative AI is the single biggest business technology adoption that I've ever seen in the history of covering enterprise technology. So as I already mentioned, the way the internet changed access to data, generative AI will change access to expertise. So how do you survive as an artist? And I'm getting this from the consultants that I interviewed. So consultants are totally in the crosshairs, their business model is about to disappear. So all the consultants are freaking out. PricewaterhouseCoopers is spending billions of dollars to figure out what their mission is gonna be in the age of generative AI because they do accounting and other advice and generative AI is gonna be able to give that advice better and cheaper than they are. So the first thing is you educate yourself on what's happening out there. You experiment with what's happening out there and you figure out where you can use it and where you don't wanna use it. So for example, if you're a portrait painter, you might use photographs as reference images. So if you are an artist, you might use AI to generate a bunch of marketing stuff to promote your human art. And finally, you lean into your humanity. The same way that organic farmers lean into the fact that they're organic and how has its own Facebook page, you lean into that. If you're an artist, you go on YouTube. You go places where you can show that you are human and you make that a major, major selling point. Okay, so for OpenSim, if you are a content creator or OpenSim, the two big use cases are marketing and content. Marketing is absolutely the big one. It's the one that OpenSim grids have the hardest time with. You should be using it for marketing right away. And that means that I use Claude II for text generation. Claude II is run by Anthropic, founded by expats from OpenAI, based on constitutional AI. And it does not use your stuff for training data. The basic plan is free and you get all the features. You can upload 150,000 word book. That's like a huge, thick epic fantasy novel and you can work with it. It's as good as GPT-4. And again, it does not use your stuff for training data. And it's not being sued by everybody. An Adobe Firefly for art, it's not as good as Mid Journey, but it's getting up there and it's trained on fully licensed images and pays its artists. So those are the two right now that I recommend, but there's like hundreds of others. Use it to write website text, press releases, emails to your customers and clients and members. Use it to create illustrations for your marketing designs. It also suggests ideas for everything, for ideas for social media posts, ideas for like literally everything that you're having trouble with. It can give you advice and help you work through things. You are so good. You're almost at AI yourself. Look at that link. That's wonderful. More content for Open Sigma. So first of all, speed of development. This was March, 2022. This is what AI art looked like. This is the state of the art in AI art. This was April. So it looks a little bit more like a cat. This was July. This was when Mid Journey started winning art competitions. And this was November. This is the pace of change of AI and the way that image generation changed last year. Text generation is changing this year. Next year, we're gonna see video and 3D coming. But the same fast pace of changed. If you're not using it because it's not good enough, learn how to use it right now because it's about to get good enough like literally any second. In Open Sim, where did it go? In Open Sim, you can use it to create in-world stories, NPC dialogues, textures, 3D objects, scenes. You can use it to code interactive elements and scripts and debugging your code and to write documentation for code. It is really, really good at this and getting faster and better at it literally day by day. I've seen some estimates that it's doubling an ability every two months. It's an insane pace of change. And AI is designing its own chips and improving its own code. So it's like bootstrapping itself at this point. It is insane. This past year has been one of those insane periods in human technological inflection. People are saying it's the invention of electricity, the industrial revolution all over again. And I don't blame them. I've been covering this for decades. And yes, this is the fastest pace of change I have ever seen in my life. So absolutely email me if you have any questions. And here is the way you can get the slides online. Well, thank you, Maria, for such an informative and interesting presentation. As a reminder to our audience, you will want to check out the conference that opensimulator.org to see what's coming up on the conference schedule. Our next session begins at 5 p.m. in this region and it's called Managing Student Emotions in the Metaverse. We also encourage you to visit the posters in the expo regions for the sponsor and crowd funder booths. Thank you again, Maria, and to you, the audience. Thank you for having me.