AI Security, Claude Styles, Doc Synch + MCP

Welcome to wake up with AI, the podcast where human powered meets AI assisted. Join your hosts, Chris Carillon, Niko Lofakas, and George b Thomas as we dive deep into the world of artificial intelligence. From the latest AI news to cutting edge tools and skill sets, we are here to help business owners, marketers, and everyday individuals unlock their full potential with the power of AI. Let's get started.

Good morning. Happy Monday, December 2, 2024. It is time to wake up with AI. George b Thomas, Nicolas Fakus here with me. How are you guys doing today?

Doing well, sir. Doing well. Had a had a wonderful holiday. Holiday full of full of news, full of updates, and, feature releases. So I'm doing really good.

I too enjoyed the time off. I did a little work here and there, to be honest with everybody. Like, let's just let's call a spade a spade. But it was the stuff that I wanted to do, but I also I also spent a lot of time just watching, binge watching. I finished my fear of the Grateful Dead or Walking Dead.

Sorry. Not Grateful Dead. That would be a that would be a unique band moving forward. But then also, the Daryl Dixon spin off of that as well. So I I'm nice and rested, and I'm ready to talk about AI.

Yeah. It was a good holiday for sure. I got to see Wicked last night. And, yeah, I love especially as it relates to marketing and how you get into people's minds and, like, stick. Like, I was introduced there, like, 20 years ago.

Right? And at some point, was listening to that album at least daily. Right? It's been a long time since then. I could still sing along with every song in the movie.

I didn't because there were other people around.

Oh, you should have. When I went with my daughters, they were singing, and I was like, this is wonderful. Now I will tell you, Chris, I did take my son over the weekend and and Nico to gladiator 2. There was no singing during that movie, by the way.

I was one of my top five movies, the

the original. Big fan. Haven't seen the this

one yet. AI news. Original. Big fan. Haven't seen the this one yet.

AI news, Niko. Let's get back in the swing of things here.

The news is honestly pretty thick right now. I mean, 4 days is a lot for for what's been going on. There's let's put it this way, the plateau that people thought things were at is definitely not that, right, in terms of like what it's supposed to represent. It was supposed to represent this point at which like, oh, maybe we haven't quite gotten as far as as we thought. Maybe we should have a much better grasp of how we're going to make the advance and that we should be we shouldn't have the the scaling issues that we do.

To be honest with you, I didn't think that they were going to run into any sort of scaling issues. Kinda doesn't seem like they have. It's just a matter of where you're scaling things, I suppose. The amount of power behind it hasn't changed. So, in light of all of this, the jump for AI security and safety has, like, really started to heat up, go through the roof.

And so much so, I look at these things in terms of how I gauge. Like, how asleep people might be and where everybody is at in terms of, like, knowledge on this stuff. There's people like myself. It's like I'm I'm scrounging for every little crumb of information that I can find related to, like, what's the latest and greatest. And then there's, you know, people who are, like, they, like, stand by, like, watching, you know, talks and things of that nature.

Then there's people who are kinda, like, dipping into it every once in a while all the way down the line. You know, it's it's been pretty interesting to watch, like, BBC jump in there real quick and then followed by, like, I think it was, like, MSNBC. I was seeing more of the financial sectors as it was relating directly to stock jumps. But then eventually 60 Minutes jumped on board and a lot of other bandwagons. And so finally this morning, and I'm sure it happened yesterday, Fox News.

They had their first full blown segment surprisingly with Sam Altman for, like, 10 minutes. And I say surprisingly because I think most all of us should know by now that Elon is the right hand man as it is for AI for the White House for the next 4 years. So it was just a little surprising that they went with Sam over Elon, but then again, OpenAI is a little bit, you know, just a tiny bit more popular and famous than than Grok and x AI. So

Just a small bit.

Smidgen. That's definitely getting stepped up. But I think the biggest development to come out of last week is the jump with regards to Quad. And, like, what is happening just with, like, what are the advances that are going on there? Why is nobody talking about it?

What what is being done?

My god. There there are so many things where I'm like, Claude, you are stepping up your game, sir. I'll let you continue on, but I have some thoughts around the Claude. There's three things particularly around Claude, but but go ahead, Nico.

So one of the bigger aspects of what's happened lately is that Claude now has styles that you can choose from. And it has some default styles and then it has some additional styles that you can create. These now this is outside of if you've been using Claude, you should know that you also have a means of telling it what type of work that you do. So if you go into your settings, you'll see what best describes your work, and you can sort of tell it what that is. And then you also now have this personal preferences for Claude.

And this is all, like, beta features if you guys are signed up for 4 beta features with Claude, then you'll have this. You know, know, for myself, you could see that, essentially, I I do some coding in HubSpot. So I want it to automatically understand that anytime I'm working with code or doing something related to custom coded actions, that I want Claude to immediately create a to create an artifact that I can just save them to projects that we're working on or copy and paste whatever the the answer is to wherever I need it. And so now, most especially, whenever I'm asked requesting information specifically, requesting something that is in the form of, like, oh, you know, user needs pile of information. That comes out in an artifact now as opposed to just a normal straight quad response.

And so, you know, of course, I told it I wanted to be straightforward in things and just, you know, telling it to be a little bit more personable. And I love using the phrasing with it or being experimental with it and saying, like, wherever you feel like it, you make the preference. I don't I don't wanna have to necessarily tell you when to when to and when not to use my name, so to speak.

So I just wanna throw some things that are are on my brain around that piece that you're talking about because mine, while you were referencing coding, mine is around, like, writing. And and mine says marketing as well, but the first thing is I've I've contemplated on creating a Claude account just for superhuman framework, not a project, a Claude. Because then I could be like, oh, we're human resources. And the context of human resources do that anyway anyway, I I digress. But, you know, in mine, I have things like not clickbait, but perspective shifting truth bombs that demand attention and drive action.

Right? I've got things like content architecture, one clear h one, logical h two sections that support the main topic. H three subsections for enhanced scannability, meta description that drives clicks, and alt text for all images. Like, description that drives clicks, and alt text for all images. Like so, like, there's these things that you can put in.

I I want everybody to understand, Nico, how important what you just said to what they're gonna do with their AI, you know, assistant or projects or or cloud in general. So okay. I'll shut up.

No. No. You're right on because that's the one of the things I was I was just about to bring up is that the adoption rate of what's going on with this. So you you probably have all worked with, GPT a little bit, or hopefully have worked with it a little bit. And so you understand there's custom GPTs, there's custom instructions, there's all that kind of stuff.

But probably one of the greatest hang ups has been writing style and getting it to write and sound like you, and getting it to legitimately write and sound like you. And one of the things that I've noticed when it comes to Claude is when you're building out styles, there is a build time. It's really minuscule. It's not like it's not a lot. It's like a few seconds, but there is some there is a level of reasoning that is going on underneath there, I believe, with the information that you're giving it.

And I don't know if that goes so far as to, let's say, build out its own sort of subset or something like build out its own persona or something like that in order to to help with that or if it's just, you know, a software thing that it has to do. But it feels more like it's taking what you're giving it into consideration and at the same time, the output seems more consistent with whatever with what it is that you're requesting. So by default, you're gonna be given 3 different styles, concise, explanatory, and formal. And if you've been using Claude lately, you'll know that the use has gone up so much so that most every response has been forced into being concise in order to just handle the amount of traffic on the servers. That speaks to what I was what I was going to say earlier which is that the adoption rate on using Claude at an enterprise level has increased 30%.

Not on ChatGPT, the enterprise adoption of AI LOM systems has decreased on OpenAI's front by over 10% in the last quarter and increased by 30% for Anthropic. There's a lot of core reasons why that's the case. 2 of which are the most important in my opinion. 1, they lead the field in safety. They they literally set the standards.

They are the most rigorous and stringent with their own model and it is the forefront of what they do before they do anything. They're the only team I know of. I'm sure that Google might have some lying around but they don't talk about it. They're the only team I know of that has psychologists on board that are actually working with and working on how the model is going to interact and react with people, given what the developments are that they're making. 2nd to that, the model, by default, does not train on your data.

Doesn't train on anything you upload. Doesn't train on anything you build in a project. You are, like, you are and I don't know how else to put it. Like, everything is just data safe, I suppose. You could think of it that way.

It's the most trustworthy I can I can very honestly say that? It is the most trustworthy language model that you can use By way of, like, data safety. Right? I'm not saying that it doesn't hallucinate or anything like that. Every model's capable unfortunately, we've gone from a point at which the model was hallucinating because it didn't have the information to a point now where the model could hallucinate because it's capable of bullshitting you.

See, you gotta watch out for that too. Because there's been plenty of examples where you could tell a model, hey, what's the color of the sky? And it'll say blue. And you'll say, actually no. It has no color because, you know, it's a refraction of the blue light that comes from, the the water in the ocean.

It's not really blue. It's really it has no color. It's transparent. Oh, yeah. You're right.

Now, that is correct. That is a scientifically correct answer that I that I corrected it with. But everybody in the you know, everybody knows sky's blue. Like, we just refer to it that way. Right?

So you could tell it at that point, like, you know what? Technically, it's prismatic. It's black, sometimes. You can't even tell. It's just this weird thing.

It depends on the spectrum of light you're looking at it in. Yeah. Yeah. Actually, you're right. You know, it can it can definitely go along to get along.

I think that's a better way of putting it. But, that being said, these that I think that's why these these tools and these features work so much better with quad is because of the fact that they are tailoring what they do to enterprise. They are tailoring this for, you know, mass mass adoption at a mass adoption at a professional level. Let's just put it that way. Now gbt is definitely gaining mass adoption at a at an I don't wanna say amateur level, but just on a street level, I guess.

On, like, your average user level. Right? Like, so on the the kids going to school and that kind of stuff. But in terms of, like, who's using it in the office, like, administratively, it's probably going to you know, this is this is the Microsoft Apple of our time. Let's put it that way.

So here's here's one thing I because I wanna I wanna skip into a thing that was eye opening for me around Claude versus g p. So one of the things I I loved about or love oh, interesting that I used loved. Anyway, one of the things about GPT is when the Google Drive thing came out, and I could, like, take a document, get into a conversation, and just roll with it. And and here's the thing. When I was doing it over there, my mindset and by the way, my mindset could be wrong.

My mindset was that I was just basically uploading a document. It paid attention to what that document was. It used it for context, but then the story was kind of over. Like, that was it. Now I read something because, obviously, Anthropic, Claude now does, Google Docs.

And I read something, and it made me wonder, oh, is that how GPT is using it, or is this different? And then I jumped to the conclusion of, well, how many people who have used it there won't think it's a big deal here. And let me explain what I mean. If you go to support dotanthropic.com and then a bunch of words, you can Google this. It it literally says important notes.

By the way, important notes. Google Docs added to chats and project knowledge are synced directly from Google Drive, ensuring you're always working with the latest version. Meaning, if you add information to a document, it's going to look at the new information in the document, which then just says, okay. So, basically, what I hear is that's almost a little bit of, like, I can train it on my own information, have a method to my madness around the Google Drive folder and the documents that I'm gonna use and a way that I'm going to update those documents because then I'm updating my project. Important notes.

Synced anyway. I'll I'll be quiet.

I mean, I dig it because, like, I'm sure other people will pro might use it in this way. This this is the way that I'm not sure that a whole lot of people are doing it yet. I was doing the same thing with with projects prior to the Google Drive connection. I was using a project in order to save project output to that project, so like as you keep going on. Now that it's got the drive aspect to it, the book that I'm writing just sits in a singular file.

And so all the chapter output just goes right back into that file. So it is constantly in the know of where we're at in, you know, which chapter we're on and where we're at in the chapter of the book, and I don't have to keep, like, saving little clips anymore. I just boom, boom, and we're dead. We're good.

Google Drive is one way syncing from the Google Drive. Right?

Yeah. Yeah. So I'm not writing to it from Quanta. It's just pulling from it. Yeah.

And now with with the MCP stuff, you can

make it

There we go. There's the other things. Because what I found, I started playing with the the Google Drive right away. And it was like I want

I want everybody to pay attention to this. Right? We were talking about how long it takes to get, like, AI ready. Please listen to what Chris is gonna say.

If I could just share share my screen for this one. But what I was gonna say, like, definitely learning curve about how and when to use all these tools because I got excited about Google Drive too, but I'm also a guy that just tries to shove all the context into the knowledge base immediately. And one of the problems I was having and the difference between syncing to the Google Drive and just uploading a file, right, or adding a text file to the knowledge base. I'm not gonna show that here, but it was like an 8% difference of the space it would take up in the knowledge base, which makes sense because there's probably a ton of other information that it has to have to be able to sync it. But and that's one way.

The other update that happened last week in related to your discussion, like, it's almost like Claude and anthropic prioritized context at all costs where, like, you you get that tactical let's get things done feeling when you use chat gpt. That it's like, let's it doesn't matter the context. Let's just okay. In, out, in, out, in, out. Unless you just overload it with context, then it starts doing.

Whereas, Claude, even as I was testing these tools. Right? So the MCP allows you to connect to other systems. And shout out to to David Onrej because I saw this last week, hit the first barrier, and was like, oh, it's a developer stuff. I don't have any coding tools.

I, you know, I hit friction, and I stopped immediately. And his video showed up on my YouTube yesterday in 18 minutes. It's, like, exactly how to set it up. And, like, 3 times in the video, he's like, you just gotta push through. Just take a half hour, push through.

This is amazing stuff. Amazing capabilities in the hands of of people that have never had access to this stuff before, including myself. I didn't have v s. I didn't have Visual Studio on my system. I didn't have any code editor.

Not a developer. No GitHubs. And but was able to follow his directions and then use Claude to troubleshoot my way to getting it done, which felt so awesome. But it got me to a place of connecting to search, Brave Search, API, and GitHub. And I have my own GitHub repo now, which I've never had before with a page.

Right? But I wanted to say for the search, what was interesting this is where I understand the difference between 2 tools. Like, with chat gbt, I remember George coming back when when this first opened up and he, like, was able to search for superhuman framework and give it a URL. I don't even know if he gave it a URL, but it came back with just, like, a chat g p t came back with just, like, a verbatim. It's not verbatim, but it's a good summary.

Right? Yeah. I tried to do the same thing here, and it looks like like if there's just not enough information to confirm or, like, give it enough context, it's not gonna say, okay. This one site this one website says this thing. Here's the real information.

It's like unless there's enough context for it to understand that it's real information, there's enough context for it to understand that it's real information and that it's probably true, it's not just gonna give it to you, like, verbatim, which is awesome to think about, because the power of just being able to publish stuff online and then have it, you know, just go everywhere is definitely a concern. But just to kinda finish the loop here and this is possible here at MCP, it's basically agents. You can set up an an agent system where please do the following. Make a simple HTML page, create a repository, push the HTML page, add some CSS, make an issue, make a branch, make a pull request. All these things in GitHub happening.

And this is somebody I've never used GitHub in my life ever. I created a profile just so I could start to do this and end up with, like, this, you know, a basic web page. I didn't even give it anything to use. Like, it's working off knowledge base. I just said, like, make a simple page.

I didn't even say what kind. It's working from the knowledge base in the project, and this is ridiculous, man. Like,

I just

And that's, like, I I just again, I want people to understand Chris didn't know anything about Visual Studio, didn't know anything about GitHub. Didn't know anything about repos. Didn't know anything about how to install Node. Js, how to run it. Any of that stuff.

Didn't know any of that 48 hours ago. Does now. Not only does he, he's, like, gung ho. Like, he was just chatting to me last night about, like, oh man, I'm doing this, and I'm gonna I'm setting this up, and I'm gonna do this. That's the effect, guys.

Like, that's the effect of having knowledge of something that you didn't before, but the thing that you have knowledge of now is what you have likely wanted to know about for years, but have just either not had the time or the capability or whatever. And now that it's in the palm of your hand, you really like, this is the time to just absolutely just soak it up with every effort that you have.

Well, in the the rate of speed. Right? Like, I want the thing first of all, all of this is amazing. And and what it's doing to empower, enable humans, what I want everybody to understand is and Chris referenced this, the guy says, just push through for 30 minutes. You know how quick 30 minutes is?

You can't even watch an episode of your probably favorite show in 30 minutes. Like, most of the unlock and ability that Chris now has based on pushing through for 30 minutes. I just hope people understand that piece of it.

And that's where, you know, I was thinking about why. Like, I mean, you said it in the videos. Like, this is where everybody drops off like flies. Like, this step 2. Right?

And I was like, yep. That was me last week. Meanwhile, we use things like, you know, HubSpot and, I mean, websites. We use new things all the time where it's there's troubleshooting involved. There is learning.

Like, it doesn't work out of the gate, and you like, if you're if you have the motivation, you will figure it out. Try to find some motivation with AI. And but that being said, it like, videos like that are super helpful. That's why we're doing this thing. We're gonna be sharing more details about a December 16th workshop.

Our 1st AI wake up with AI workshop that that we're doing in a couple weeks, and that's all we wanna do. Like, we wanna share what's going on here because we know a lot of you out there are wanting to do things like this. And my switch from 48 hours, like, on Friday, I think I shared some. I was like, alright. I guess I gotta get the developers involved again.

And then now today, I'm like, nope. I mean, not for that stuff. I still love the developers, and I want to, you know, engage and get all their insights. But there's just these little things that, like, if it's admin work, how about we don't ask any any humans to do that stuff anymore? Like, let's do bigger and better stuff.

We all can.

Yeah. Everybody needs to realize that everything's changing. I mean, we say that so much, but, like, I'm sitting here looking at my desktop as we're recording this on Riverside, and there are there is a chat gpt desktop app pinned to my taskbar. There is a Claude, desktop app pinned to my taskbar. The third thing and they're on the far left side.

You know what the left side is? The important side. And right after that is oh my gosh. But you're gonna learn a lot about me, but right after that is Visual Studio. So I have the 2 AIs and Visual Studio as the first three icons so that I can wake up with AI and get nerdy right from my desktop, not even have to go to Google Chrome, not even have to just right here, right now.

Let's build. That's what happens when you wake up with AI.

That's a wrap for this episode of wake up with AI. We hope that you feel a little more inspired, a little more informed, and a whole lot more excited about how AI can augment your life and business. Always remember that this journey is just the beginning and that we are right here with you every step of the way. If you love today's episode, don't forget to subscribe, share, and leave a review. You can also connect with us on social media to stay updated with all things AI.

Until next time. Stay curious, stay empowered, and wake up with AI.