Summary
Notes
Transcript
doing class is I just opened up my notion page clicked audio transcription click one button and what's happening real time is it's actually recording my voice the full audio and as you can see on the screen it's actually transcribing real time too so I'll try to zoom in a little bit so you guys can see it clearer So, you can see as I'm talking, it's literally doing real-time transcription. Now, here's the problem.
If you have the Notion app, it's free. You got to pay for the AI Meet note-taking version, which is a little bit more. I think it's like 10 or 20 bucks a month. But it does the real-time transcription. You can actually chat with all your content on Notion. If you forget everything, the good thing is you've got a phone, you can record it automatically and transcribe. So step one, I want everybody to cheat. Now here's another device.
It's called OMI. This has a lot of flexibility. It's kind of creepy if you're in the state of California. I think it's illegal to record without other people knowing that you're recording. So obviously, if you're a jurisdiction, you figure out. I'm not an attorney. You figure out what's the rule. So I generally say in our company, every meeting should be recorded.
unless somebody explicitly states that it's not going to be recorded. So we are in the policy of record all the information. I'm regulated by a lot of industries. I'll give you some background and context. So I went to Georgia Tech, studied computer science, studied biomedical engineering, and have never used my degree ever. I went straight into being an entrepreneur 30 years. I think it's 20 different companies.
I've sold three companies. Over the last three years, I've been in the Stanford program, which has been transformational. I'm also in the family office network. And I have about a couple thousand employees spread around the world, but nobody reports to me.
I've extracted myself out each one of the companies as a president, And I just hang out with them. And I'm going to walk you through also some of the questions you have where it's like, who's got a daily stand-up?
Well, worst case scenario, by the time we're, well, I don't know about in one hour, but everybody's going to create an agent. And we're going to... Give some examples of how to connect the agents to communicate with each other, how to have a data store.
So we will kick it off. So I'm a professor at Rollins in AI, and I've been lecturing at Georgia Tech for 23 years on quantitative computational finance and AI. We do everything in our company. So if you want to have case studies, I've had a hedge fund, a commodity hedge fund, got a couple billion, sold it. I have a direct lending fund in the consumer space. So if you want to, I think some people have like investment fund background and context, so we can talk about how we can create dashboards over there and intelligence layer.
It's going to be hosted at Rollins. Who's been to my monthly class? Anybody? Handful of people. So I will send out, just somebody remind me, every single month, actually because of the Stanford program, I've been teaching a free class every single month of every single tool and technology that's coming out. How we're applying it. I have a bunch of CEOs that talk about how they're applying it within their company. Completely complimentary. You can bring your entire staff. We have a bunch of companies who've brought like hundreds of their employees to do training programs.
All of it's complimentary. I just love teaching. I want to give out this information to everybody. Okay, so we'll start with building an AI agent. I did mention this earlier, who set up a relevance AI accounts? We got one person. Okay, so at least one person, ideally multiple people. What I'd like you to do is go to relevanceai.com. You can set up a free account. Is it the most advanced AI agent development system?
No. Is it easy to understand? Yes. So I, at this point, want you guys to actually have practical experience. We use this for enterprise grade, right? So it's not that it's like lower grade and stuff like that. It's just not as custom engine, right? So unless your team has actual development resources or AI engineers, this is probably one of the easiest ways. Just get started. Question. Yeah, yeah.
Okay, don't worry about that. As long as you have an account, I'm going to walk you through how to create an agent. So I'm going to give you guys just a few minutes. I want everybody to at least set up an account. And once you've got your account, you should see a page similar to this. You're gonna see just a area where you can create agents.
Free is okay. Free is fine. You should be able to create AI agents with the free version.
Sorry to bring Tony Robbins over here. You can do whatever you want, but I just want to have one per table at least, and they're in a use case. You can all create different ones, but I want each table to have one that you can work off of it and you can see the consequences. So what do you guys create? That sales prospect lead generator, okay? What's your night major? BDC agent, okay. And what was yours?
Chief of staff, that's a great one. Okay.
I've checked. Everybody, there's somebody at your table who's already created an agent. So I'm going to walk through. So you're going to click default. And what you're going to do is just give a title to your agent, okay? So let's see, I'm going to create Professor AI. Now, The reason why I like relevance, it's simple stupid. If you just say, "I don't want to think about anything," it'll do it for you.
This has been a game changer in my life. Comet browser COMET Browser So right now you're actually seeing Comet Browser. It's a perplexity tool. To actually download the browser, it's free. But in order to use the advanced features, you do have to pay for perplexity. Does anybody have a perplexity account? Great. I think there is one person at every single table that has a paid perplexity account. So for those of you who have a perplexity account, You can download, everybody can download Comic Browser, but I'm going to show you why it's mind-blowing.
Help to figure out what model to use. OK. You can select any model. But guess what happens whenever you ask a model what model to pick? They're all fucking biased. So I'm going to use Sonar, which is Perplexity's model, and it won't be able to choose itself. They're all fucking biased.
Say I've never used relevance before and I don't know how the hell to use this tool. This is like a new tool that has like a thousand different things to do. Well, Common Browser will take control over the computer, over your browser, and it will click all the buttons and set it up for you.
that it doesn't allow you to do jack shit. That's the problem with this law. Like, it wants you to approve every single five seconds, every single thing that you can do. What is the downside of having a system that has that power over your browser?
So what I said was, if you wanted to download it, just search for Comet browser. It's a free download. You download it. It's just another Chrome skip. It just creates a chrome scan of a perplexity model and it allows you to use this thing called assistant right on the right side to control. I think it's a great cheat code, especially if it's a brand new application that you've never used before. All right, so what I want you to do is take five minutes and I want you to go select Select the model. Let me close this.
I see some lights there. What's the question? Okay, no question. Not yet. All right. Somebody, okay, you guys are creating agents, right? At least somebody at the table is creating an agent, right? Please. So it says create with AI. You can just tell it exactly what you want it to be. You said Olympia Pope there? You can actually do research, say I can create an AI based on Olivia Pope. And it's gonna actually do research and create the framework, the base framework for Olivia Pope.
So here's how I cheat on prompting, you know, This is like the easiest way to do it now. Let's say... I'm going to copy this. This is a terrible prompt, right? It's like one of the worst prompts of all time. Take this prompt and optimize it. For whatever model you want. So if you're using like, say you're on Claude and you're using Sonnet 4.6, all you would do is, in this case, let's Pull up Claude Sonnet 4.6.
So they've actually built in now all the models. If you're like, I don't know how to do a great model, or a great cron, Use the model. To get it to optimize the prop for your model. And we talked about, do you guys remember what zero and multi-shot prompting is? Zero at multi-shot prompting. Okay, whenever I just push enter, then zero shot. You will get up to 40% better performance Out of a problem, if you'd ask it at least twice, to improve the prompt itself.
Ask it to do that. So it's created a prompt here. It created an optimized prompt. Give me a-this is how I cheat. Give me a numbered letter. List of suggestions. to improve the prompt. in a Y, N, or edit So by forcing the agent to give a numbered orList.
There's a model. It's a mathematical model. It's called Aristotle. Everybody can write down Aristotle. Aristotle is a math-based model. It is not based on languages, it is based on mathematics. There's a new model, Vlad Telyev, who's the co-founder of Robin Hood. He's building a physics-based model. That means that anything that it outputs has to correctly work in physics. The math bottle Aristotle has to work perfectly mathematically.
It's the professor who's like, hey, show your work. That's one of the other tricks that I use. At the end, if you're like, "Hey, I don't know if this is correct," At the end of all of my agents and skills, I get it to do a math curve. I get it to force The eval, remember we talked about evals? force it to prove that it is correct mathematically. Even if it's language-based, it has to show it logically and prove itself. Those are your evals.
I am telling you. You will not make mistakes if you go through that process. One mathematical, if you use Aristotle as a math-based model, And two, you have to prove it. Have it prove itself. Do a math proof. The older models, you can't do that. Every single model right now, all your top models, they will all work that way.
They will force the proof backwards. So let's look at this. So it said... So let's do it again. It said, take this prompt and optimize it. So I took that BS prompt that I created And it created an optimized prop. Of course, you can edit if you'd like. But that's a pretty comprehensive prompt that it created. Then I said, Give me a number list to improve it. So what did it do? It went through, it said add this, remove that, add this, keep that, remove this.
So I go back and forth, ChachiBD is better now, Claude is better tomorrow, Gemini is getting better the day after. With a tool like Relevance, you can automatically update every single model every single time that it comes out. In fact, if there's a new model, it'll actually reach out to you and automatically ask you, do you want to update to the latest model? And I'll do that automatically for you.
I don't know why the hell they come up with these complex acronyms. It's just a knowledge base. That's it.
I took the audio transcription and I plugged it into the knowledge base. Just literally, well, I mean I had an automated sequence in Claude, but the point is that copy and paste of the transcript. So it will not hallucinate at all. If Jonathan said it, It will repeat it. If Jonathan did not say it, it will not repeat it. So everybody's concerned about hallucination with it coming up a random shit. You've got to create a In the knowledge base, information you can upload and it will not hallucinate.
Not, no, N-A-S or 12? Has anybody heard of parallel.ai? Parallel.ai will transform your life. I swear to God, all my legal bills... So I'll tell you how we do it legal now. We drafted using Clod plugins. We can show you guys that later. but it's generic. Parallel.ai, what we do is we can do research on every single jurisdiction And so it extracts all the case law, all the regulations, all the statutes for those domiciles and pull it into your law, into your contract. Automatically.
And they have a free account. Parallel.ai. You can create your own account. They give you like $80. You will not run out using it regular. It'll take you, I've been using this for like two and a half years. I still haven't run out. The amount of credit, it is the highest accuracy and the cheapest. It is better than any other research, full stop, period. So if you're going to do any research, parallel.ai, we'll show you how to connect it to your other tools.
But if you're going to do research on, say, Tony Robbins or RIAs or how to do, If valuations of real estate, and you extract out best case, Parallel.ai, full shot, is the absolute best quality, zero hallucination, highest resolution search that you could possibly do. And it's fully integrated with Claude and all these other tools that we're talking about. So let's see what the artifact came up with. What was the question over there?
I know we have somebody close to Washington. Anybody know? You need to get spousal consent. That's a very specific question. regulatory for the state of Washington. If you use cloud plugins, there's no fucking way it's gonna tell you that that is how it is. So you can use Claude to draft it, and use parallel.ai to do the research, and then put the contract in there based upon actual law. I'm going to show you that. I'm going to check out on you then. Don't worry about it.
You could add a picture, you could add a text file, you could add a Word document. Just add something to your knowledge base.
My entire life has been changed because of connectors. Anthropic invented the MCP connector. I'm going to show you how to do it right now. My screen is a little bit small. Actually, let me pull it up in...
And it's like, it's like, Five seconds to do this. So you go into connectors And see it's got Asana, I've got Confluence, I've got DocuSign, Gmail, Google Drive, Hax, Microsoft, Notion, Parallel.ai. So now I can just chat with all the tools and use Claw as kind of a central repository and it does all the work for me.
I want everybody to at least do one connection. So you can do the browse connection. For parallel.ai, it's a custom one, but there are hundreds upon hundreds of pre-developed connectors that is already pre-integrated within Anthropiq. I'm still scrolling. this gigantic list of connectors.
What you're doing is you're creating context rot. Has that blanket longer? The stuff that's in that brain gets rotten. They call it context rot. So what you need to do is create new chats. But then the question is, well, how am I going to keep creating chats if it's not going to remember where all the stuff is? The only way to do that is you've got to take the stuff outside of the window. Take it outside of the chat.
Cloud Codex allows you to work with files on your computer. So this is like level one. Instead of keeping it in chat, you can dump the context into a folder.
So with... Projects You can use an existing folder on your computer or create a new folder on your computer. So as it's working, it'll keep dumping the files and the context into the folder. And so you never have to worry. You can just create new chat after new chat after new chat and it's working in the same folder. And that allows you to not get stressed out about, hey, am I getting context wrong?
Am I filling up the memory of the AI? And this automatically solves that problem for you. You don't have to stress out about it all. So what I'd like you to do is go ahead into Cloud Cowork. And I want you to create a project and just select any folder. I would highly recommend using an empty folder. You can store it wherever you want. We'll say new folder, YPOS Stanford 10. test folder. And I'm going to select the folder. And that's it.
And now everything that I work in that project is just gonna dump all the files into that folder right on your computer. So if you have Google Drive or you have OneDrive or you have iCloud or whatever it is, you can put it on there and you can sync all your file folders. Remember this is like level one, there's way more advanced concepts like Obsidian and all these like, or RAC databases and stuff like that.
Does anybody know Databricks? Oh, you haven't filmed. Okay, let me film me doing this. So I don't chat anymore. I use dispatch on clients. that's connected to my company computer. So now I can chat all the time with all my files. So every single file that's on your computer, instead of just doing regular chat and losing all the context, I can chat via dispatch. Go ahead. It connects with your computer.
You don't have to spend $5,000 or $6,000 on it, but you now never have to use chat again, and you never have to worry that the context is going to rot. Because when you chat with it on your phone, And you click dispatch right on your cell phone. It says looking for a desktop. Literally, this is the desktop I got it to connect to. And now I can chat directly with cowork on my phone anywhere in the world.
Good. Well, in co-work, it's dumping the files onto your machine. So instead of stuck in chat world, say for instance, some other company comes out with some incredible technology. And you're like, what the hell am I gonna do? Literally one of my forum mates was like, I've got everything in ChatGPT. How the hell am I gonna get all this shit over to Glot? Well, I mean, we got it through Codex, but it was this long process.
So I can ask a question about Gmail, my Google Drive, my company information, my Notion information, my Parallel.A, and one single chat enter... All connected right there. It's a super brain connected. Can you have it bunched first? Yeah, so here's the problem with, you can, if you click it, obviously the power's just going to keep on running. Yeah. Because if you want to chat on your cell phone and you're in Thailand and your computer's in Atlanta, something's got to be on.
So, dispatch is not enabled automatically if you have an enterprise account. Your IT department has to enable it for you. All right, it is... Is anybody here that's supposed to go to Novo? Yeah, time for Novo. Otherwise Pascal's gonna kill me.
A skill is just a bunch of steps. That's all it is. Does anybody have SOPs in your company? That's all it is. It's just very detailed SOPs. And by having detailed SOPs, you're going to have one step, then two steps, then three steps, then four steps. But now, remember we used connectors? One step could be Go to my financial system and pull down the financial records. Step two, create a analysis of the last quarter's financial statements, analyzing it and breaking it up by sector.
I'm going to walk you through how to do this. Oh, I'm using Opus 4.8. Okay, let's see. So all you would do is Create a skill. Based on the following. Now, I'm not going to use Opus because that thing is going to take forever. So I'm going to use the dumbest model. Do not do this if it's real life. But what it'll do is, Cloud will actually create a skill. Now that you have a skill, see it says use the skill?
So this really is a work You got it. You're basically packaging that SOP into a thing called a skill. And that skill that is a step-by-step instruction. So anytime you run it, it's going to do it for you. So... I created a JP Voice. So when I text versus Microsoft Teams versus Slack versus one of my six email accounts versus whatever, we all communicate slightly differently, right? In fact, for every person in your company, you slightly communicate differently.
Did you type create a skill and it said save skill? You need to do like three steps and then say create a skill for me.
Tell me a new research. One of the workloads I have So who has too many memes with people they don't know or they don't remember, right? I think I met 1600 different unique people last year. Hi, I'm a AI agent. that goes on the internet, uses parallel.ai, extracts their LinkedIn profile and all public information about them, puts it into a single format of their profile, their family, their interests, their non-profits, everything in a single detail, and it automatically goes on my calendar, creates a calendar invite next to the meeting, the exact same time and day, and so before every meeting I just click the calendar, boom, I know everything about the person.
That's a workload that's been life changing. Right? Because I can't remember everything about everybody, except it pulls an email address and I ran the skill where it's go use parallel.ai to do research on this person. And I have an AI agent that has a very specific structure. And it uses the calendar app, puts it into the calendar as an appointment automatically. And then every time before I go to a meeting, I just click the button and it has a profile of everybody.
So I gotta find the email as a reference. Well, guess what? It pulls it in there automatically. Creates a profile of them, the candidate snapshot. Does anybody know what the top grading methodology is? It's just another like how to interview a person, one of the best practices out there. It did this entire analysis on this person I'm going to interview, scored them, what are the questions I should ask them.
Analyze their resonances It took all the information from Greenhouse where it's storing all the other previous interviews, pulls it into the analysis, the AI agent formats it the way that I need to do the interview, tells me the highlights and lowlights, preface the questions for me, and does it fully automatic. I don't even, I... Just have to click the button in the middle of the interview and it's already ready for me in the calendar.
And if they want to use this, here's the next thing, guys. Okay. Everybody create a skill. Anybody who wanted to create a skill, create a skill, right? Now you're going to be like, how do I do this automatically? Who is it that did the automated-oh, he had to head out. OK. Right here is what's called scheduled.
Get a new laptop that's outside of the company. By suggestion. Were you able to get the schedule? Okay, perfect. So what you're gonna do in schedule is create a new task. Oh, what's the schedule? Simple, stupid, right? It's just a reoccurring appointment, right? It's going to run this sequence. Consistently.
Before I finish up here, Okay, so I want everybody to do this one. And this is again only available in Clawed Co-Work. So right here, we looked at skills. So once you create a skill and save a skill, It is now always available in every single chat. So you created a skill over there. That skill is now available everywhere within Clod. So you just click the plus sign, and you click Skills. So in here, there's a tool called Plugins.
All a plugin is, a bunch of skills. That's it. Nothing special. But Anthropic has an incredible job of analyzing best practices of investment banking, of finance. And so what you can do... You create your own plugins, but I use their plugins. So, man, I wish I could. Okay, zoom in there. So this says legal. When the legal plugin came out, And the finance plugin came out.
Yes, they have WhatsApp connection too. Yeah, they have WhatsApp connector. Search online. If it doesn't have it pre-built in. Thank you. Yeah. All right, brother. You got it. No worries.
Yeah, of course. I'll send it out. I'll send it out. Awesome. I'll reach out. No worries.