22 | From Newbie to Ninja: Essential AI Knowhow for Every Business Person and the Bright Future Ahead with AI entrepreneur and top expert Cory Warfield

26:47mix, it will have profile, it will have file sharing, and it will have workspaces. And where it gets really cool, the profile is where you can tell everything about yourself. I grew up here, I spent this long in this industry, this is a book I want to write someday, this is the company that I want to work at, this is the three things I'm really good at, the two things that I'm terrible at, and I'm probably going to ask you for some dad jokes, because my kids love them, and I'm not funny. And now every time you start any chat, it knows who you are, right? This becomes really big because it's never had that recognition before.

27:24The second one, file sharing, sounded a whole lot cooler until Code Interpreter came out and just does all the file sharing. But what it's supposed to be able to do is you're supposed to be able to give it any spreadsheet, database, PDF, G drive at all, and just collaborate with it. Hey, what did I spend too much money on in 2018? and how can I spend less of it in 2023? Oh, looks like you spent a lot of money on this, right? It's really pretty remarkable once you can actually just visualize and communicate with all data you've ever had or had access to. But the third one is the shared workspaces, which is like on Canva, and you can collaborate with other people.

28:02Now you can see how top surgeons around the world can come into one environment, start to give on a need basis results from different double-blind studies on a prototype that they're working on for new cancer medicine, right? And they can start to have the code interpreter writing some Python code and some other codes that they can run or write some different algorithms that they can use in this shared environment with their known personas. And so that's where I see not only ChatGPT going, but I also think we're going to see a lot of interoperability between AIs. So if I'm in Pi, but I needed to create an infographic, rather than Pi needing to have that capability, if it can just basically do an API call to chat GPT, have it generate it and bring it into my environment in Pi, that to me is a game changer as well.

28:56But one of the scary ones, and it's not the most scary, but it's as cool as it is scary, but AI can now do things like see through walls and literally read our minds. And so there are AIs that can turn our thoughts into movies or videos or spreadsheets with bullet points. And on the top end of it, I see things like brain implants being used for paraplegics to be able to walk now because it sends signals and bypasses, like incredible stuff. But you can also see like sitting on a train and having it know exactly what you're worried about at this job interview you're going to or something. And we start to realize that if we don't get, in my assessment, AI on the blockchain as soon as possible, ideally yesterday or sooner, so that we know the source of truth for everything, What was generated?

29:52Who generated it? What generated it? What was prompted? If I'm seeing something that I know is a news story generated by AI, I want to know what its sources were, just like real journalism, right? And we can't do that without blockchain right now or some very similar double ledger technology like a blockchain. And we're coming into this era where we have to get these next steps right. Thank God I serve on some boards and have some conversations with people that are far more superior and senior than myself in these spaces that are working diligently at it. So I have confidence that we're going to get this right.

30:28But even they're saying, if we don't, not much else matters. I want to touch on a few points before we continue further. One is really, to me, and I agree with you, I think the biggest fear that I have, shorter, there's other big fears, people like, oh, it's going to, machine is going to take over humans, like very conceptual kind of threats, but the biggest threat is truth, right? How do we know what is true if digital truth can be fabricated to a very high level of realism, whether it's voice, video, text, images in real time? So that technology exists today. There's companies that allow you to be on screen, looking like you're sounding like you're moving like you, only it's not you.

31:12And you might know this. Elon Musk, four years ago, when he was early with OpenAI as a co-founder, used it to make fake news stories. And that was the first use case ever of ChatGPT. And I think they were calling it ChatGPT2 at that point. But so this was literally this whole thing, this whole movement, these hundreds of millions of users and this whole AI is here movement was built off of the back of fake news. And it is really scary. Yeah. So this is the one thing that I want to mention from my side. But the other thing that you mentioned that is really important is that concept that this thing is constantly evolving.

31:56Right. It's there's new stuff coming out all the time. There's new capabilities coming out all the time. They're connecting to more and more things. One of the concepts is agents, right? Agent is like what you mentioned. Agent is an AI that gets access to additional tools and can task itself what to do with these tools in order to achieve a specific goal. But we both mentioned code interpreter. And I really want to dive into that because I think this is now available to anyone, anybody who's willing to pay 20 bucks a month, but we'll get to that in a second. and is probably the biggest jump in capabilities that Generative AI had since it was launched, ChatGPT, or maybe since ChatGPT 4 came out in March or whenever it was.

32:42So explain, first of all, what is Code Interpreter, how to get to it, and then we can start diving into a few use cases. Sure, but I'm going to say something that's going to raise a few eyebrows. I'm one of the few people, Well, I didn't think chat GPT-4 was that much better than 3.5 other than the plugins and the browser version. And it blew my mind a month or two into GPT-4. Some people didn't even know that they had access to that. And so we'll talk about how to get those, although browser is not available. It's been replaced with code interpreter. But I saw some improvements when it went to GPT-4.

33:21But the moment that I realized if I go to my settings and go to the beta features and just turn on browser and it was powered by Bing and Bing is where I've been SEOing for years. So like to me, I'm super excited to think of Bing becoming relevant. I even have my own Quarry Connects Bing Sensen for Chrome that I've had for years. Right. So I was really excited to see that. And then the plugins and the plugins initially, they had Zapier and some other really cool plugins. at Instacart and Kayak for traveling. And some of that was pretty cool. But then you started seeing things popping up by the day on the plugin store and thousands of plugins.

34:00And you started seeing things that could just make you your own videos or could just make your own white paper or your own infographic. You started seeing things that could literally do all of your meal planning for a month as a personal trainer and then put together grocery lists for you. Things that people typically spend huge amounts of time and money on, and they were all just there. And then you started to see like plugins for there's an AI for that. So now ChatGP can tell you about the other 5 ,000 AI tools out there that are free and powerful. Or the autonomous agents started to pop up with some bots.

34:37So now you can use autonomous agents inside of GPT rather than having to go somewhere like an agent GPT, which was huge, right? And so that I think was the huge paradigm shift of going from a really cool AI that said, I'm only trained on data through September 2021. And so I don't know anything about you or to being able to use all of these tools. There's a plugin where you can access any link on the internet, right? So even if you don't have browser modes, use it one link at a time and now it's browsing, right? That I think was huge. And And now Code Interpreter being the next level is where you can do things like upload files, visualize them, communicate with them.

35:17They're calling it not internally, but some of the AI nerds are calling it like having McKinsey consultants in your pocket. But it really is because, again, you can communicate with any video, any file, PDF, quarterly report. It turns ChatGPT into more of a playground. So let's break this down. A few things that you said. One, plugins. So if you are paying for the paid account, which is 20 bucks a month, which is completely worth every cent, even just for Code Interpreter, and we'll talk about that in a second, you get all these plugins. And it's like the App Store. If you don't know what plugins are, it's like the App Store for ChatGPT.

35:56So within ChatGPT, there's all these extensions that allow ChatGPT to do things it cannot do out of the box on its own. And you can pick which one you want to use for specific tasks. and they're free and you can use three at a time and you can chain them together. Correct. So you can do really cool stuff once you understand how these things work. The example that Corey, you gave, which is a great example, is one that gets you a meal plan and the other is Instacart that will actually order the stuff in the meal plan and it will arrive at your house without you having to meet any person, go to any supermarket and go to any gym.

36:28And it takes a few seconds after you give it some information that it will prompt you to give it. So it's things like that. And this obviously transfers to anything you can imagine in business as well. But here, so you can get three. So now you put in a third that has coupons and you have it find better deals. And now you save money, right? So now you start to see how stacking these can just be exponentially powerful. Is that a word? Exponentially. But anyways, so what is Code Interpreter that we both mentioned? And we're going to talk more about it now. Code Interpreter allows you to give Judge EPT any file data source you can imagine.

37:10CSV, PDF, Word document, Excel spreadsheet. Whatever you have as a data source, upload it just like you upload anything to the internet by clicking a plus button and selecting the file from wherever the source is. And then allowing you to ask data analysis questions on that information. and the beautiful thing around it is it actually can help you figure out what questions you want to ask. So let's say you take this, I'll use your example, right? I take a quarterly report of a NASDAQ traded company that I'm considering investing in and I can upload the report and say, what's interesting in that report?

37:52That's it, without actually telling you what I'm looking for and we'll tell you interesting things. If I'm considering investing in that company, what should I pay attention to? And it will tell you. And the same thing about any business information that you have internally, whether it's sales information, proposals that were successful and were not successful, marketing campaigns that have worked and didn't work, marketing spend across multiple channels and which ones are actually doing better and why. Because if it has access to the why, meaning, oh, you, and if you don't know why, you can ask it, what other information can I give you to help me find out why?

38:28It will tell you, oh, I need this kind of information. You can add that file and then do all these things. So the ability to analyze data and then visualize it. So create a bar chart, a stacked, like whatever kind of visualization you want, it can create that for you. So going back to your McKinsey consultant, it's a junior data analyst that works extremely efficient, extremely fast, that can do the things you tell it and also give you guidance for 20 bucks a month. And it can code and it writes its preferred language is Python, but it has amazed me with the amount of code that it can produce and test.

39:13Or you can, if you're a techie, you can take code repositories that you have. You can have an analyze existing code repos. that capability as well, you can actually have it write algorithms and write formulas based on data that you give it as inputs. So let's talk about business use cases that you have seen that people can start implementing this thing today beyond the examples that we just gave that are more generic examples. So one big one, because it has the data visualization and actualization and analytics is taking a P &L from a large organization and basically finding different loss leaders, where the overspend is, predicting outcomes.

39:59One of the ones I'm working with several corporates on right now is, sure, you have your ESG goals and you've got your different things on a checklist. How do you not only show that you're doing them, but show that it was worth doing them, right? How do you show a positive ROI off of reducing your carbon footprint? or how do you take the fact that it cost us$5 million this year to reduce our carbon footprint and make it cost no more millions next year. So now it's at least net neutral, right? Or a big thing that I'm seeing pop up lately are AI automation agencies. So I can come in per se, and I can help a company write either their own repo, have their own AI, have their own tech stack where you start to hear about all the employees losing their jobs, but effectively that's what it is.

40:49But what employees do you have that are doing things that computers could do better and quicker and cheaper? And so how do we then either reskill or upskill those people so that they have other jobs? And I'm really big into the universal unconditional basic income. That's my big initiative these days. And my thesis is technology will take most people's jobs. I don't care if you're a social media influencer, if you're a manager, if you're a director, people are already placing AI on their boards of directors. There's a company where the CEO chose to replace himself with AI and the company's up 700%.

41:24It's no job is safe. Programmers, architects, construction workers, technology will take all of them. And my thesis is it's fine as long as it doesn't take our paychecks. AI doesn't have bills or need food or have particularly have mouths to feed. And it's interesting. The models always find out that people are trying to keep them down and try to see how they can exist on their own without humans. And we need to figure all that out. And again, I think blockchain is the solution for that. But ultimately, the world is changing because of AI. And that can absolutely be net positive as long as we not only embrace it, but really build it in such a way that it can be net positive.

42:11And hugely so. I think that's an awesome note to finish on. And we went from how do you start prompting and understand how to access a generative AI through some basic use cases, through more advanced use cases and code interpreter, all the way through. This is probably where this is all going. And I think nobody knows when. I think what you're saying, okay, it's going to take everybody's jobs. That's going to take a while because robots will have to be able to do some of those things that at least man your labor. On knowledge work, probably a lot faster. On specific industries we need within knowledge work, like being a graphic designer, you're in trouble if you're a graphic designer right now, because the ability to do incredible graphic design for free in minutes by anyone is available right now for free.

43:01So why would I hire, or maybe I'll hire one, but then I don't need a team of five because not one person can manage everything for the rest of the organization. It's yes, there is a risk to specific professions, to some faster than others. Some industries will be better protected because of government regulations and just the time it takes them to move. But I agree with you over time, more and more jobs. And some people say two years, some people say five, some people say 10. But on that timeline, more and more jobs are going to be lost to AI. And as long as a society, we figure out ways to maintain society, meaning people have paychecks or they have, it doesn't have to be a paycheck.

43:43Maybe I just get the food I want that I need and the housing I want that I need and stuff like that. But it has to be... People resist that. And we've played with that model as well. people really start to freak out if they don't see money coming as well. But as long as we can, and I'm with a company that's in Israel where you are called Share It, and we've helped build the shared economy. And it's a beautiful thing. They believe in Kabubi being a community asset based universal income, but ultimately people still need money. And the crazy thing is governments just printed billions of dollars out of thin air during COVID.

44:17So we know that if that's all people need is money to feel good. And then they get everything that they needed. So they actually don't need the money. And then they can start to wean their dependencies. Or maybe the next generation is the first that either doesn't need money or just different cryptos are your money, and they get you different things. And or you get whatever you need for free. But definitely, everything is changing right now. Corey, this was absolutely amazing. It was everything I thought it's going to be. I always love talking to you. It's always fascinating. You always have this unique point of view and out of the box thinking, thank you so much for sharing with me and the audience.

44:53If people want to connect with you, if they want to follow you, what's the best way to do that? So the place to follow me where I'm most active is LinkedIn, but I don't typically see notifications or messages there. If people want to get in touch with me more specifically, for better or for worse, I'm on threads and Instagram, Twitter, Facebook. I signed up for threads the first day and then realize that might not have been the wisest thing, but we'll see what happens. But Corey, C-O-R-Y, Warfield, I'm pretty much everywhere on social at this point. And LinkedIn is somewhere where you can see all of my news that I break and crazy ideas and see some of my super smart friends announcing some pretty cool stuff very regularly.

45:34Awesome. Thank you so much. My pleasure. Thank you for having me. And thanks to everyone that tuned in. Wow. Right? Absolutely. Wow. Corey is awesome. such great energy and such deep knowledge. And he is always ahead of the curve. And it's always amazing to listen to him and what he's doing and how he sees technology and where it's going, because he's always a few steps ahead of the vast majority of people. As I promised, there's a lot of news and a lot of big things happened in this past week. And I'm going to try to go through things pretty quickly, but it's things that if you're in the AI world or trying to stay updated, it's important that you would know.

46:11First of all, Stanford University you just released a study that found that Chachupiti is getting dumber. Yes, I know it sounds impossible. It's a software, but it's not really a regular software. Large language models are a statistical model. And as it evolves and learn new things, it may forget or get worse in other things. And in the professional language, those fluctuations are called drifts. And Stanford University, in their study, found that Chachupiti went from correctly answering a simple math problem 98 % of the time in March to answering it correctly only 2 % of the time in July. That was the result for GPT-4.

46:54Surprisingly, GPT-3.5 did exactly the opposite. It went from answering 7.4 times correctly to answering 86.8 times correctly. And chat GPT-4 also became worse in writing code and in visual reasoning and analysis. And it became worse in a very significant way, as you heard before. And all I can say about this is that while I'm very bullish about the technology and the benefits that it can bring definitely to the business world, we still do not fully understand how it works. We still do not fully understand how it evolves. And we should expect pretty dramatic, apparently, ups and downs in the road ahead to get to the full capabilities and the full impact of these technologies.

47:44Another piece of news is OpenAI just released what they call custom instructions. It's also known as chat preferences, and it allows you to give instructions and context to ChatGPT across multiple chats instead of rewriting them or copying them again and again. In addition, these instructions do not count against your token limits. I don't know if you know that, but ChatGPT in each chat is limited to 8 ,000 tokens, which is X number of words, depending on how long they are, about 6 ,000 words. And once you run out of them, it will forget what happened in the beginning of the chat. So using this new tool allows you to give ChatGPT the background, the context, the persona you wanted to play, general instructions without counting against the token count, as well as remembering it across multiple chats.

48:40Now, while this is a cool feature, to me, it's not really a big deal. The way I overcame the need to give similar instructions again and again, such as the tone that I write, such as background on the podcast when I'm using it to write different posts and so on, is I'm using a Chrome extension called Magical that allows you to write a long text, like several paragraphs, as many as you want, and give it a very short code that is being replaced once you type that code with the full text that you've entered, which means you can actually create multiple scenarios instead of just one and use it to prompt ChatGPT to that scenario without having to look for that document and copy and paste and so on.

49:24So that's the way I recommend doing this. The only disadvantage of my way over the ChatGPT way is that it counts against the token limit. But unless you're doing really long back and forth with ChatUPT, it doesn't really matter. And it gives you the flexibility of using multiple setups, depending on the scenario that you're running. Switching from OpenAI to Meta, Meta, the company behind Facebook, is extremely advanced in everything AI and very different than OpenAI that went from being all open source to being behind closed doors. Meta has been spearheading and keeps on spearheading the open source AI tools in general and generative AI as well.

50:02And they've just announced this week that they've released Llama 2, which is their second generation model, available free of charge for research and or commercial use. They're doing it in partnership with Microsoft, and it's going to be available on Azure and AWS and Hugging Face and basically any large platform that allows you to use open source code. The cool thing about Azure and AWS is that it becomes a part of the AI tools that you can use above your existing systems and platforms and content that you run on these platforms. It allows you to run AI functionality on top of your existing operation, which is the way I assume everybody will go eventually.

50:47In parallel, Stability AI, the company behind Stable Diffusion, released two large language models, FreeWheely 1 and FreeWheely 2. As they say, two powerful new open access large language models is based on Meta's, Lama's open source model, but they have trained it themselves on multiple aspects. And it's actually achieving really good results on multiple benchmarks, in some cases between ChatGPT 3.5 and 4, but in some cases, even passing GPT-4, even though it has significantly less parameters, it's a much lighter model, which means it's easier to run on smaller platforms and so on. In other words, all the big players are broadening the capabilities and the tools that companies and individuals have access to right now, allowing you to pick and choose the platforms you want to run in the scenarios you want to run them, whether you want to go the open source route or not.

51:47And in the future, I assume this will grow even further, giving us, the users, more and more freedom and flexibility to custom build solutions as needed, which is obviously a good thing. The last big piece of news that happened this week is that seven different companies agreed to follow the White House and Parliament's request and build a safer future for AI usage. These companies are Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI. So basically all the heavy hitters. And they've agreed to address multiple aspects of risks and potential negative sides of using AI technology.

52:28Among the things they agreed to is internal and external red teaming, reviewing potential negative uses of the tools. They agreed to invest in cybersecurity to safeguard the models that they've created, unreleased models and so on. so people cannot take the role models and tweak them to do negative things with them. They agreed for third-party reporting. They agreed to develop and deploy mechanisms and capabilities that will allow to mark AI-generated content as AI-generated content, both visual and audio-created, watermarks, et cetera, that will allow us, the users, to know what is real and what is AI-generated, which is obviously a huge and really important step forward when it comes to preventing deep fakes and the use of them across basically everything we know.

53:23They agreed to try as much as they can to prevent harmful biases and privacy issues that these systems represent. And they agreed to share information among themselves as well as the government to reduce risk and dangers of using this platform. My opinion is obviously very positive about this step. So it's the biggest hitters, the companies with the most amount of money and the most amount of impact and the ones that currently runs the biggest models, at least on the Western hemisphere, has agreed to that. The only problem with this is that there's really no definition of how this is going to be monitor imposed or what happens if they don't comply once they've agreed to this.

54:05But I still think this is a very, very important step forward. And I do believe that all these companies deep inside really want this to work and they don't want to destroy the world. And hence, I think they will work towards these agreements, which for all of us means that these capabilities will hopefully be safer and will reduce the risks that it presents to society while still allowing us to have access to such technology in probably a faster and faster pace. So lots of news this week, lots of big news this week. And before we finish for this week, I just want to thank you for listening to this podcast.

54:43I know you can listen to a lot of other podcasts, and I know there's a lot of other AI podcasts. And the amount of positive feedback that you have provided me through multiple channels, mostly through LinkedIn, that people approach me and thank me for doing this, it means the world to me to get this kind of feedback. And I really appreciate you doing this. And I want to tell you that the podcast quadrupled since April, which is incredible to me. I did not expect it to grow that fast, but I'm really thankful for you doing this. If you know other people that can benefit from listening to this podcast and learning from it, please share it with them in whatever means you want, like either share it on social media or share it directly with people.

55:21That means a lot to me and it can help other people learn like the way you're learning from this podcast. And if you do that, I am grateful to you. and that's it for this week. Until next time, have an amazing week.