BONUS: How We Would Teach AI From Scratch in 2026

17:40Corey:Okay. Let's do Microsoft. Because I was just doing some stuff with Microsoft earlier. Subscriber-based pitching tips, AI integration for program management, Microsoft AI security focus, core weave plans. These are not like searches. These are searches of conversations I've had with my personal AI before where Microsoft was mentioned. It works really, really well. Also, you've got various apps you can use, but that's probably a little more in depth than you want to go today. But if we have time at the end, we'll go back to that. Deep research is a tool that allows you to ask a question and get it to go think for 30 minutes.

18:18Corey:You know, if what you're doing is, say you're doing a big report and you want it to go through like a spreadsheet and you want it to go gather industry data from a government site, stuff like that, and accuracy is really, really important, that's where you use deep research. It takes a little while longer. How long is good to wait in thinking mode? And once you see it circling, do you just get out or confront? it can take as long as a few minutes, Lexi. Sorry, I just saw you ask that question. It can take as long as a few minutes. It can be a little short. It totally depends on what you ask. They've gotten much more about kind of flexing that amount of time.

18:57Corey:It used to be that thinking meant X number of seconds, and then it got a little shorter, and then people wanted longer because they had harder problems and giving it more time lets it think more. So, you know, the truth is you can run it, tell it to go, and if it starts circling, you can go over to another chat and work on something else and you can have it notify you when it's done. So you don't have to just sit and stare at it. Usually what I will do is I'll spin up two or three chat GPT windows here and if I have two or three things I know I need to work on this morning, I'll go start each of those and then I'll go over and I'll drink my coffee while I go through my email and just let it work.

19:36Corey:You don't have to just stare at it. The truth is that's where the big productivity boost is. It's doing some legwork for me while I'm going through something else I need to do right now anyways. My day is busy and this has been such an unlock. I was looking at like even just Grant and I's production numbers from the last like 14 months. Grant, I haven't even told you this. Oh, really? My recording has stopped.

20:03Grant:Oh, no, you're good. I still see you. Okay.

20:07Corey:Let me hit something here.

20:11Grant:Oh, it looks like he just accidentally kicked himself off the stream. Well, he'll rejoin in just a moment. I'm curious to hear what he was talking about, but I just posted something in the chat just to recap what we covered. So we covered you go to your favorite website, and maybe you don't have a favorite website yet. So, you know, the easiest one to start with is chat.com. The one that I really like lately is Claude.ai. But you can't go wrong with either of those. Gemini and Grok are the third and the fourth probably best. And Gemini has higher intelligence than or is at least on par with the highest intelligence.

20:51Grant:But we find that it's not as consistent. And actually, the app is not as good. So that's why we would recommend Chat and Claude first. and then Gemini second, and then, yeah, for total beginners, for total beginners.

21:05Corey:Yeah.

21:05Grant:Did you mention Copilot? Yeah. There is Copilot as well. Yeah.

21:10Corey:The thing, the reason I often, and I know there are people who have strong opinions about Copilot and really dislike it, but the reason I recommend it sometimes to beginners is because for one subscription, you get access to ChatGPT's best models, Claude's best, Anthropik's best models. So you get ChatGPT and Claude in one account. You can switch between them. And there's a new feature they just dropped yesterday to where those models critique each other. No one else is doing that, at least not on a broad scale. It's a thing we've done in agents and such, but not a thing that has been like rolled out to the masses.

21:46Corey:And that's coming very, very soon. It's in Frontier now, and it'll be to the public by summer. The truth is, if you're a casual user, you're going to get a lot out of that, especially if you already save your files in 365 or something anyway. With that said, yeah, I'm ChatGPT Claude.

22:06Grant:Yeah. I mean, I use Gemini. We'll get into this. And Gemini. I use Gemini for specific use cases, but not for my daily driver, and I'll explain why in a minute.

22:15Corey:Yeah, Grant and I have opposing daily drivers is what you should know. And it's not for any real reason. The truth is most of the same features exist in both tools. Both of them have at any given moment the top two models in the space. But what I will say is that once you get a tool you're comfortable with, sometimes it's not worth changing every time there's a new model or a little tweak. I mean, if there's something significant you want, absolutely. but I highly recommend finding one you're just like the interface on, you're comfortable you can go and try out their free versions and see I just don't like how this works or I really like how this works and if that's the case, stick with it because these companies are putting out models every couple of months right now and if you're with either one of these, you're always going to have something that's right near state of the art, if not the state of the art so I would highly recommend either one of them or both if you just have the expendable money because um i use them both daily so um i just want to cover one thing so perplex

23:23Grant:someone asked about perplexity um the reason why i would maybe not subscribe to perplexity right now unless you're going to pay for their expensive version for perplexity computer which is a different product is because most of what perplexity does chat gbt does and gemini can do and Claude can kind of do. And we'll get into that more in part two. But if you like Perplexity and you find it really, really valuable for you, specifically for search use cases, totally fine to use it. But it's not, in my opinion, necessarily a replacement for a daily driver like Chat or Claude. If you're using it that way, I'd love to learn more about how you're using it.

24:05Grant:For me, it's more of a supplement too.

24:08Corey:I really love Perplexity's Discover feature where it caters news to me. Essentially, it writes news articles based on what it thinks I want to read the news about today. And a lot of times it's a lot more positive and interesting than the real news are.

24:22Grant:And yeah, Perplexity has a lot of really sophisticated financial use cases. Like I said, Perplexity Computer is like a totally different product. And that would be kind of like a replacement for OpenClaw. And that's really good. So I think if you're willing to spend the money for a Perplexity Computer, that's a completely different conversation. and that probably would be worth your time. Now, it's not to say that you can't do what Perplexity Computer does in ChatGPT or Claude. You just can't do it to the level that Perplexity Computer does it.

24:53Corey:Yes, exactly.

24:54Grant:But yeah, Perplexity for Research, Marcus Stone says, yeah, I mean, it's pretty good. I find that it can access probably the most websites besides ChatGPT. I think ChatGPT can access pretty much every website

25:05Corey:that you need to go to.

25:06Grant:um yeah but yeah so let's get into um are there any more absolute beginner things we want to cover

25:14Corey:um i i think hitting you know prompts how to use the interface the only thing i could have shown that i didn't and i'm hesitant to bring it back up because of my internet access is if you go to your settings in chat gpt there's a personalization option and it's a big menu it's not intimidating but it's like how do you want it to talk to it do you want it to you do you want it to be friendly nerdy pragmatic do you want it to be you know very professional you know you can customize all of that and you can tell it things about yourself and determine like how your interactions go and

25:44Grant:that was a really big unlock for me we can show it on mine why don't you walk me through how to do

25:50Corey:that okay go down to the bottom left where your information is there your settings click okay go

25:58Grant:up to set a personalization okay so you see this so what i did is i clicked on my profile in the bottom left corner that'll be your name or your email or something and i went to personalization

26:11Corey:and yeah and when you get over here like that top drop down for example is a really good one you know you've got all these different options you can have it communicate with you however you want i usually use quirky i believe it seems like there was a nerdy but i could be wrong you can adjust its tone in a variety of ways up and down a little bit if something's getting on your nerves or you've got something you really feel like you need from it to better unlock

26:38Grant:capabilities for you this is really cool it really is it's a great personalization feature and not

26:43Corey:nearly enough people use it in custom instructions like mine is told to do things like challenge my assumptions uh double check me if i've taken anything for granted as true that is not uh and a variety of things like that uh that are uh designed to get it to give me what i need out of my ai and what i need is a backstop to uh i rush i do a lot of things in a small amount of time every single day of my life and i need something that can be like slow down cory you for you've left the keys in the door much like claude did yesterday and uh yeah we'll talk about that maybe later. If not, I see there are a million questions in the chat here.

27:30Corey:Yeah, let's answer these. We can put all of these out and we can put them in a blog article after because it's pretty clear we're not going to get every single one of them today. But whatever we don't, we'll put in a blog article and share in the Neuron as well.

27:44Grant:I will say for the question about Markham, MVP, best way to use AI tools to integrate into a job search, we're going to cover something that will help you do that. And in fact, maybe we can use that as part of our scenario planning here. The one other thing I wanted to say, so I posted in the chat, okay, things we covered, go to your favorite platform. You might not have a favorite platform yet. So just try, maybe try chat.com, claw.ai, gemini.google.com or drop.com, and just ask a couple of questions to each of them and see which one you like the best. And then just go with that one. because right now the intelligence is pretty on par.

28:23Grant:I would say the first three are the better of the three at the moment. And the first two are the better of the two for all of the things we're going to cover next. Then you can iterate. So you ask a question. Then the important thing to remember is you're not going to get what you want the first time. So you iterate with it, which Corey showed us how to do. And then iterate on the question. And then you can search to find previous chats. And then the other thing that I didn't mention in that message is cory talked about this here which is the the model picker and he talked about why you would want this instant is for just everyday chats thinking is if you want it to use a higher degree of intelligence and you're okay waiting for an answer and pro this anywhere from five seconds to a minute or two yeah right it can take some time and you'll know that that's turned on right

29:15Corey:here yeah and you've got an extended thinking option right there as well is that that drop down

29:21Grant:oh yeah you can do standard and extend it if if what you need for it is to give you the highest intelligence possible then it's probably a good idea to just have extended thinking on all the time that's what i do when i use claude and then there's other tools here but we'll get into this slightly uh later um as they're relevant but just remember that if you have this plus sign here you You can use these tools and there's lots of different tools on here.

29:48Corey:All right. So now we're going to get into the second part of this.

29:51Grant:I'm going to stop sharing my screen for a second. So we were talking about whether or not if you're a total beginner, if you start by setting up a project or setting or learn how to prompt efficiently. Corey, what's your reason for learning basic prompting etiquette? And then I'll share my reasoning for setting up a project before you really get involved into anything.

30:14Corey:I'll tell you why I do. And the reason I say this is, you know, prompt engineering was a big subject a couple of years ago. It was largely mocked by the technical community and heavily adopted by non-technicals. It is still a valuable skill in interacting with AI, but it is not like a profession of its own. So something I see as what we thought was the end game early on was like, oh, the world's going to be full of prompt engineers. We're all going to be engineering prompts, and that's what it's going to be. Instead, I think of prompt engineering as AI 101. There are a ton of short classes. You can watch a 15-minute YouTube video and get a lot of ideas and different patterns that you can use to help you out.

31:02Corey:And what that does is it gives you a little toolbox to show up to the party with. So you know, like, ah, I want to do this. how can I get the best results to this question possible? You can absolutely just say some words and it'll probably get you close. But if you need that extra 3%, that extra 5 % that increases your accuracy or ensures what you want, the ability to understand how to carefully craft a question, how to ensure you're not doing things like contradicting yourself in a prompt, which is a big issue. Understanding how to just be clear. The truth is what you want is the AI to be clear.

31:42Corey:It's going to answer whatever you ask it. Most of the time when it answers it wrong, it's likely because someone asked it wrong. And just taking a little extra time there, watch a video. That's really all I would do. I would watch a video or two, take a little extra time when you're thinking about it, especially on hard problems. The truth is most of grant and i's day it gets no thought but as a as an absolute beginner i found it to be a major unlock for me personally and i highly recommend it

32:13Grant:all right now let me explain the logic behind projects and and what they are and and to do that i'll share my screen and i'll just kind of get into the explanation yeah yeah so give me one are we bouncing over to claude now yeah we're gonna we're gonna go to claude for this next part

32:29Corey:You might give the 20 second rundown of how everything's the same, but it looks a little different as far as like, you know, where you're at.

32:38Grant:Because I think that's you've got that under control here.

32:42Corey:Yeah. When you're in the desktop app, you're not in. Yeah.

32:49Grant:So we're going to talk about that. So another layer of confusion for these tools is that there's a web based version that you can access by going to chat.com. or claw.ai, whatever your preference is. And then there's also a web, or sorry, there's also a desktop version you can download. Of both. Yeah, of both of them. And I actually would recommend downloading the desktop version, especially if you want to be using this for any sort of serious work. And the reason I recommend that is because there's just more capabilities that you can do. So I'm going to show you a difference.

33:27Corey:I recommend it because then your AI isn't hiding in one of the 1200 tabs you have open, it's its own program.

33:32Grant:Yeah. So I'll give you an example here. So this is the web version of Claude, right? And you can see it's a chat window. It looks very similar to ChatTBT. It has a couple different items on the side panel here. But then it's, interestingly, it doesn't have some of the same features that the desktop version does. And the desktop version, without getting too complicated, has this tool called Cowork. which is where this lives. And then it has another tool called Code, which if you ever hear Claude Code, we write about that a lot in our newsletter. That's what this is referring to. And as you can see here, I was gonna ask Claude, is this safe to use an MCP server?

34:14Grant:But yeah, we'll get into that. That's not for today. We're not talking about coding. We're just talking about chat and how this works. So that's why I recommend the desktop. Now, ChatGPT also has a desktop app. I'm not logged into that right now, but it also has one called Codex. And I am logged into Codex. And we might do some demos if Corey's internet is behaving inside Codex later. But I'll show you the same thing I'm showing you in Claude. I will show you in Codex, which is sort of like ChatTPT's equivalent. So let's talk about this thing called projects. So I'm going to share my screen here.

34:53Grant:Shrink this a little bit. Okay. So what is a project? Projects are folders that you are basically like organizing folders. So in the same way that you would have a Google Drive where you organize all of your documents, this is a way of organizing all of your chats, but it's a lot more powerful. And I'll explain how that works. Corey, also, let me know if there's any questions in the chat related to this, and I can answer them because I can't see my screen.

35:19Corey:Yeah, I'll keep an eye on them while you're going, sir. Okay, great.

35:23Grant:So the reason why I think you should set up a project before you do any sort of work or chatting is because it's a really good way to keep this like really messy, like all chats log organized, first and foremost. but also you can set yourself up for success by using projects in a way that you can't if you just rely on a standalone chat. So let's set up a test project together. Everybody who has a problem that you're trying to solve, go ahead and ask those questions in the chat. What type of project would you like to set up for yourself in AI? What do you want to have the AI help you do? Give us a good example.

36:09Corey:Grant will show you how we do it right there. What is the co-pilot equivalent of a project? I should know this.

36:16Grant:I think they have projects. Yeah, they have projects. I think it's called projects. I was thinking it was called projects, but I could be wrong. It could be called folders, potentially. Are you pulling that up on your screen?

36:31Corey:Yeah, I am looking. Let's see. man the computer is not happy with me today we'll give you an answer to that we'll we'll we'll i'm not on like my best computer i'm on like yeah yeah i'll make sure you have an answer

36:50Grant:the chat gbt equivalent is also called projects so let me pull up the chat

36:57Grant:so projects is right here on the top uh in the sidebar corner so you just hit the little project thing and you can start a project so whether you are using this or this where'd it go disappeared hello here we go okay

37:19Grant:accidentally hit it okay so whether you're using projects in claude on the app or try gbt on the app or on the desktop or sorry on the web version that's how you do it so let's go ahead and make a new project uh corey did anyone give us any ideas for what type of project we want to create if not i'll use the job search example from before let's see uh our project's the same use quant excel job

37:49Corey:job search manager, find businesses, find people, customize contacts, interviews of career pivoters. We've had two mentioned job search. Another one mentioned lead gen. Yeah, I hit job search real quick. Let's do job search.

38:07Grant:So what we're going to do is we're going to organize all of our chats related to job searching into a personal project. So before we mentioned the five levels of AI efficiency. And I said that actually setting up a project is level one. So any tasks that you're trying to do, or any organizing structure to help you accomplish whatever you're trying to do with AI, I recommend making a project. Yes. And I have one that's like a general purpose project for everything we do at the Neuron. And basically what I do is I'm going to hit create project. And there's a couple different components here to keep in mind.

38:44Grant:And this is pretty much the same between Claude and chat, but we'll show the differences.

38:49Corey:If you're going to do it more than twice, it belongs in a project.

38:52Grant:Yeah. Yeah, I think that's right. And also, it's just helpful, like you can create a project and maybe I'll do this as an example too. I'll create a project that's just like grants a work zone. And I'm just going to say for organizing all my ideas and projects. Like this is probably not the most efficient way to do it. But But if you want just like one staging area to put all of your stuff in, you could do it that way as well. So let me explain the different components here. So first you have a chat window, which is just like if you went to chat.com and you started asking questions. But then you have some of this stuff over here.

39:31Grant:You have memory, you have instructions, and you have files. Why do you want to use any of this stuff? How is this helpful to you? So files allows you to either upload documents, add text content, add Google Drive or GitHub repos to your database so that anytime you start a chat in this project, it will be able to access that information and learn from it. instructions basically works the same way as having a set prompt that you apply to all chats in a project so basically let's go back to our let's go back to our job search manager so what are some instructions that we would want to give our job search manager one thing that we could do is we could give, we could upload our resume to the project.

40:29Grant:So anytime you start a new chat, this AI that you're talking to will know your resume and it will be able to reference it. And the way that I would do this is I would, you could upload it directly from your device or you could just add text and paste it in. And you could say like current resume April 2nd and then resume content. pasted here right then you can say help me update my resume in your project knowledge for this job description and then you can actually paste the job description in down below um if you want to be really uh really specific and really precise you can use what's called xml tags and this just says like help me update my resume in current resume April 2nd in your project knowledge for this job description and that the reason you would want to use that is because let's say you have multiple resumes uploaded in the file in the project knowledge over here.

41:46Grant:You want it to reference this specific one that's the most current. But maybe you don't want to delete all those other ones because maybe it's helpful to have them. You know, maybe you have different types of resumes. Maybe you create your resume a certain way for a certain type of job that you go out for and you create it differently for another type of job you go out for because different, let's say you have two sets of skills and maybe your, you know, some jobs and and previous positions were more relevant for that past, you know, for that, for this role, and, you know, other ones are more relevant, et cetera.

42:18Grant:So that's why you would use XML tags, if you want to direct its attention directly to a specific file. Does that make sense, Corey? Did I miss anything there?

42:28Corey:Yeah, that makes sense. That makes sense. Cool.

42:31Grant:Now, project instructions would be like, always use web search to look up the current date and time before drafting a cover letter.

42:48Corey:This is so different to how I use them in chat GPT.

42:52Grant:Okay, we got to get your take as well. But I would also say you are a resume builder and your job is to help me update my resume and write cover letters for specific jobs I'm going out for. Now, let's say you don't even know what to write here. One thing you could do is you could say, hey, I'm trying to set up a resume or job search manager project, and I need to write some custom instructions to get my agent to do things exactly the way I want. Can you help me write the best set of custom instructions possible based on Claude's current best practice documentation?

43:57Grant:Then what we can do here is we can have the AI actually look up and write the custom instructions for us. And what it's doing is it's searching the web and it's thinking and it's fetching the information from Claude's best practices. And it's going to now update them so that that applies whenever you have any chat or anything, any tasks that you need to do inside the project folder. Corey, how is this different from what you use it for?

44:26Corey:Well, it's not that different. You threw me off when you went in and started at always do it this way. I was like, wait, you're not giving it like a system prompt. I was like, oh, a system prompt would basically be a case because we haven't explained that before. And I'm not going to get into it too deep. But think of it as every AI runs on one core prompt. So like ChatGPT, Claude, they all have a single system prompt and a variety of others they connect to. But think of it as a single system prompt that tells it you are a helpful assistant. Your job is to provide users with the information they need.

45:03Corey:You're going to do this safely. You're not going to give them dangerous advice, that kind of stuff. That's what a system prompt does. And essentially, with a project, you're giving it its own little system prompt. And absolutely do what he said and have it write it for you.

45:16Grant:So that's great context. And that's something that I missed because I'm in this every day and I don't think about using the system prompt as explaining it. But that's basically what it is. when you're setting up custom instructions, it's a system prompt to help this project basically directly help you as much as possible. So one of the points about that that I would add is they can be as broad or as specific as you need. I used to have previous, you can see over here, I used to have a project folder that was just for cutting word counts. I had another one that was for helping me write the around the horn bullets and the intelligent insight bullets that we would write every day.

45:59Grant:I had one for writing subject lines, but which we're going to get to in level three, that's actually something that I am now replacing with skills, which we'll get to that. One thing I want to comment here, I don't believe ChatGPT does this, but Claude does. So it's asking me a series of follow-up questions. Do you see this here? So where will this agent run? This affects how we structure the instructions. So I'll say Claude projects.

46:23Corey:Oh, it does this too.

46:24Grant:ChatGPT does?

46:25Corey:Yeah, only on specific things. It doesn't do it on every query, but if you're asking for something like this, yes.

46:33Grant:Where should the agent primarily help you with? Finding and tracking jobs? Let's just say all of the above because I think we want this to be broad. What matters most in how it behaves? Takes action without asking? Ask before major steps? Follows a strict workflow? I would say asks before major steps is the really important one. um okay so i'm seeing some questions in the chat where are you answering those i'm answering a few

46:57Corey:uh do you want to take a minute and bring a couple up real fast yeah um can you bring up them on your side as the host i'm i'm looking at it from the wrong side yes i can i'm over here looking at them in youtube instead of looking at them in the public chat yeah yeah i do that too actually i'm like i'm like i'm over here clicking and they're not popping up on screen and i can't figure it out.

47:18Grant:Yeah, right. All right. And any and all questions related to projects, because I think projects are like really important to understand if you're going to be doing anything serious with AI inside the applications. So ask them all the way.

47:32Corey:Public chat is not coming through. I can't see it in here to bring them up. Interesting. It's very, very strange. I've never seen this happen before.

47:41Grant:Well, I'll go ahead and read them out loud.

47:43Corey:so okay yeah let's i have a couple i could read too if you want to if you want to just catch yeah

47:48Grant:why don't you why don't you read the questions and i'll respond and you can give me your answer

47:52Corey:as well okay so from i'm gonna miss up people's names so please please accept my apology in advance from karthik surrey does do projects retain context as we build out several script

48:05Grant:files for a coding project yes um now it's not perfect like for that specific use case you'd want to use cloud code um and you would want to use your local like repo for where you're writing all the code uh that's where you would want that's what you'd want to use for that but yes if you're if you wanted to do it in your chat window it could do that and basically the way that it does that is you can ask it hey what was the last conversation we had in this chat and it can actually like read the the um previous chats inside the project something someone pointed

48:45Corey:out here i want to call out real fast oh i'm sorry go ahead uh is that you know i've seen a couple questions about using it using projects as an archive as well like as where like you know you can do that and someone pointed out that you can also grab old chats and move them to your project.

49:03Grant:That's key.

49:04Corey:So that's, that's a baller move right there.

49:06Grant:So let's say you're not an absolute beginner and you actually have this just like massive log of all of these chats that you,

49:15Corey:it's just a hot mess of conversations about everything from, from how to boil an egg to how to start a car. Yeah.

49:25Grant:You can, you can do this option here in Claude, which is called change project and you can just select a different project and put it there um so that's one way to do it but that's going to be very manual um what would be nice is if they have a project organizer agent that they launch at some point that helps you just say like hey organize all my prompts into

49:44Corey:the my select button like the photos in your in your phone i want i want to select button where i can go okay bam bam bam bam bam like it's an email inbox i think chat gpt is working on something

49:57Grant:like that they have what's called the library and the library shows you all of your uploaded files yeah i believe it shows you all of the images that you've any image you've ever created any

50:08Corey:video you've ever created and every file you've ever uploaded uh i i within reason uh i yeah i don't know how far back it goes considering it's been around for a few years now and uh uh but it will do that and it's really cool to be able to go back to them i was i was really excited to see that added because it's the thing I've needed for a minute.

50:29Grant:It's actually, it's easier than finding things in my Google Drive. Yeah. So let's talk about this. So we asked it to create some instructions for us and it gave me this document here. And some people might wonder, what is an MD document? This has become the industry standard for the output that the AIs produce is they just produce basic what's called markdown files. And that's just a formatting. like for for example if I was going to open this in visual studio code you can see what it actually looks like it uses you know symbols to indicate when it's bolded or it uses xml tags which is why that's helpful because this is sort of what the agent reads when uh you send it a prompt and then it when it responds to you it produces content in this format as well but when you look at it it looks like this.

51:19Grant:And so in this case, it did use the XML tags because that's good best practices for setting up instructions. In this case, it says copy everything below this line into the custom instructions field. So we're going to do exactly that. I'm going to copy all of this and I'm going to go back to the project folder and I'm going to set this as the custom instructions and I'm going to save it. So now every new chat I start in this project will reference these instructions. Yep. Let me just fix this.

51:51Corey:And you don't have to type that prompt over and over or go back to an old, this is what I used to do before projects is I would be like, when did I talk about this last? And then I go search through my conversations and I'd be like, there it is. And then I'd go to the bottom of that and be like, write a prompt that gets me the final result here. And you know, you kind of work through it and then it'll give you a prompt at the end. Or I could go grab the prompt I'd used previously because, you know, there was never a good way for restoring them frankly uh before this and yeah so i'm gonna see if i can get a sample sample

52:24Grant:resume somewhere that i can upload and we can test this out um you want a sample resume do you have one i mean i have i i have i have a picture of one let me see if i can transcribe like can you a generic one might be the best idea add my name grant harvey to it and uh and um i would say transcribe this picture exactly as it appears because the vision models with claude are still really weak at this point um yeah my name is harvey to it and fill in any other details that are missing so what you would do instead of this is you would upload your actual resume i need a cost break oh yeah that's fine you know what you could also do you could also just send it to uh your linkedin profile and say like hey can you build a resume based on this linkedin profile okay i'm gonna look at the chat make sure there's any other project questions related can you also set instructions later yes you can um can you use it to set reminders So we're going to cover that in level four.

53:36Grant:That's more of level four topic. Talk about Claude co-working. Yes, we can. I don't want to get too distracted with that because I want to keep this very simple. But everything that we've done with chat, you can do in co-work and it's a little bit more robust and easy to use.

53:56Corey:We can also add that if there's ever one of these that a large number of you really need a deep dive into, we can absolutely do a live sometime just on that. You know, we would love to know what you need to know so we can do this right. We might put a form together or something that we can share out.

54:14Grant:Yeah, so what we've been doing is in the newsletter every day, we have the AI skill of the day. And there's a request form in there. So if you ever have a specific request you want to know, if we don't get to it the following day, what we're trying to do is publish like a digest of them at the end of the month that covers all of the skills that people have requested. So for March, we did that where we covered, you know, maybe it wasn't like the specific thing you requested, but it was like multiple people requested the same type of skill. So we wrote about it in a generic way that everyone could learn from.

54:46Grant:That's what we did. So March, it's on the website, then you're on.ai.

54:50Corey:I'm going to drop one of these down here in the chat for you just so you can see it. By the way, while you're over here checking this thing out, please take a moment to subscribe to the newsletter. The Neuron.ai, more than 700 ,000 people every morning. We share stuff like this. We share the news. We share tips. We share our videos and interviews with any of the movers and shakers in AI that we can kidnap and get to come on our show.

55:20Grant:So one thing to keep in mind, so somebody mentioned co-work. So co-work actually, this whole project file structure here is meant to replicate how the file system works on your computer. What makes co-work different from that is that co-work can actually access the files on your actual computer. So you can give it access to wherever you have your resume already existing on your computer. and you could just say like, hey, go to my resume folder and reference that. And that's really powerful, but there are some security concerns with that. So that's a little bit more advanced than what we're talking about, but we'll cover it at the end of the session today.

56:06Grant:Cool. So somebody else has a question. Can I keep client info and projects to create profiles in order to, for example, have Claude find best fit colleges based on a client's GPA and swim times? yes you can now there is an ethical question about whether or not you upload someone else's personal uh identification like material so i would probably like generify that um or make sure that you're on the team or enterprise account to make sure that that account uh that information is like safe because otherwise claude does train on your chats unless you change your settings let's see where it's where it does that i think they all default to yes you know any yeah so you make sure you have to try someone mentioned privacy make sure the one of the very first things you do is go to settings uh uh help improve claude turn that off unless you don't care yeah unless you don't care but if you're going to be uploading client information to create custom profiles for them, definitely turn this off.

57:12Grant:Because you do not want to, you know, upload somebody else's personal information into the training data. That's like a big do not do, in my opinion. So that's what I would do there. Be careful with that. LLMs sometimes go all out to keep you happy and would output what things make you happy rather than objective facts. Yeah, for sure. That's a real problem. Can co-work access emails? Yes, it can with something that's called connectors which we didn't really have um we didn't really plan to cover today but we will cover it in abstract um any other questions about projects does this make sense um hey thanks marcus appreciate you uh gotta bounce into family obligations yes the live is always available on youtube uh if it if youtube accidentally deletes it like what happened with the dan shipper episode, we'll re-upload it.

58:05Grant:So don't worry. We have a safety backup recording.

58:09Corey:So we absolutely want this to stay available and be watchable. And when this is over, we'll go and pull this whole chat screen transcript and try to extract all the questions and make sure we go through and hit things we missed. Okay.

58:26Grant:So I want to show you an example of this before we move on projects, because this is really important. So I've uploaded basically a fake resume here but imagine i've uploaded my actual resume um i've set my project instructions so it knows what the goal of this project is and uh i am now going to say can you look up jobs that i could apply for based on my resume so we're gonna walk through this here we were just referred

58:53Corey:to as the ben and jerry of ai you should know wow that's really kind like that i like that that's high praise.

59:03Grant:Yeah, interesting. There's implications there, which I'd love to unpack another time. Before I dive into the search, a quick flag. Your resume still has placeholder contact info and an unconfirmed location. That won't affect our job search now, but we need to lock those in before you can actually apply anywhere. That's good feedback. Let's see. Ooh, the neuron is required reading for students in my AI class. Love it. You can just hover over chat in the side panel, hit the three dots, and then add to project. That's right. Would you suggest moving chat dbt instructions to project instructions? So actually, yes, I would.

59:38Grant:I actually don't recommend having custom instructions that apply to all of your chats, unless it's something that you absolutely need the AI to do for you every time. Because I think it will impact the results that you get. And one reason why I personally like Claude better, and Corey, correct me if this is wrong, but it does have project-specific memory. Yeah, GPT does as well.

1:00:07Corey:You can choose between project-specific or account-wide memory.

1:00:11Grant:Yeah, that's right. And I'll just very briefly, since I mentioned we would do this, show how you would set this up.

1:00:17Corey:I can't stress what a great time this is, because the truth is, all of these tools are great. Yeah. You know, just rest assured that they chase each other so fast with feature announcements you can't see straight. So if one of them has something you don't like, just give it a couple weeks.

1:00:33Grant:So here I'm replicating the same process on ChatGPT for anyone who has ChatGPT. So I just created my project folder, JobSearchManager. And you hit the arrows in the, sorry, the dot, dot, dots in the top right corner, Project Settings. That's known as an ellipsis, Grant. I just prefer dot, dot, dot. I prefer dot, dot, dot. So it does sound more fun. And then you can set instructions. And for some reason, you can't set project can access memories from outside chats and vice versa. This cannot be changed. So I think you have to set that when you start the project, right? Okay, so here's where you do it.

1:01:12Grant:So if you want to make a project that has memories solely related to that project, like let's say you're working on something and you don't want to reference all your personal chats, this is for work. You can set the memory on chat GPT to when you go to create new project, hit the little gear icon, go to instead of default project only because you can't change it once your project is set. I don't know why you would want to have these little icons here, but you could add them if you want. I guess those are just suggestions. Yeah. Yeah. But yeah.

1:01:53Corey:Ideas, yeah.

1:01:54Grant:And then you can upload. Where do you upload files? Right there. Okay, great. Yeah, so.

1:02:02Corey:Used to be called knowledge.

1:02:04Grant:Yeah, that's right. That's what was throwing me off. So you can upload files, same as we did before. You can use text input, same as we did before. You can upload your resume. You could connect it to your Google Drive or to Slack here. And the best part about this is you can actually share these projects with other people on your team. So if you have a project for, let's say, a project that you're working on at work, you can share this with them so that everybody can ask chats and have all the same context and all the same memory that you have, which is really, really helpful. I believe you can do this on cloud.

1:02:36Corey:Does Gemini have projects?

1:02:39Grant:No, Gemini has something called Gems. Let's go to that.

1:02:43Corey:Which are like their custom GPT's really.

1:02:45Grant:Yeah. So.

1:02:49Grant:I'm going to show this real fast. So Gemini has what's called the Gem Manager and this is more like custom GPTs. This would be more akin to what we're going to cover in a minute, level three called skills. and so essentially what you would do here is you would set up a new gem and so what i did is i went to gemini.google.com slash gems slash create and you can give it a name a description you can give it instructions you can give it knowledge that you can add uploaded from your drive you could import code or from notebook lm and uh and then give it instructions for how to respond and so So the way that that works, I'm actually going to show this, Corey, if you don't, or actually, let me ask permission.

1:03:38Grant:Corey, can I show the header image creator?

1:03:40Corey:Yeah.

1:03:41Grant:Yeah, show it. Okay. So every day when we make a header image for the neuron, I will put in something here. Like I'll either give it the full draft of what the newsletter was, or I'll just give instructions will say, right, make a header image for a new around the horn digest for today. And use the text around the horn digest everything that happened in AI and AI today Thursday April 2nd 2026 and I make sure I set it to pro so I get Nana Banana Pro

1:04:27Corey:you know what though Nana Banana Pro screws up my pictures I get typos I get text that is like all caps instead so you're saying fast is more reliable on this project I have found on this excuse me on this gem i have found fast to be more reliable i want to use pro i click it but it doesn't do as good with text for me uh yeah yeah not to say it it may not sometimes uh it's specifically text the rest is great yeah um i do mine just a little bit different all i do is i type in header colon and in parentheses put the copy i want to be there and then i say uh something like

1:05:10Grant:article text for graphical context oh okay real quick i saw somebody oh here it is so this is the image it created and that's because what i did is i gave it you know all of our our previous examples of what we've done i have very specific instructions for how it's supposed to look and uh there you go it created the header image so i'm actually going to use this one for today's around the horn digest so gonna go ahead and use that it's very cute and I like it all right so but that's so that's basically how it works and as Corey mentioned he uses fast because he thinks that that performs better for his needs and that's fine but the reason he can do that is because I've actually shared this with him and so you can share these gems if your team uses Gemini and you're in the Google suite you can share these gems with other team members and then other team members can use it so that's kind of equivalent to projects it's not as uh useful i would say because it doesn't it doesn't organize well actually let me make let me make sure i don't eat my words there um if i got a video explainer no it doesn't show it doesn't show all the previous video explainer chat so it's not as useful yet um but you know that's why we recommend claude or chat gpt as your daily driver and then you can use gemini gems for one-off use cases that gemini are just better is just better at um yeah cool one more one more note on this before we move to level three um is somebody asked where can you find the same like don't train on my data and chat gpt so So I think it's under settings.

1:06:55Corey:Data controls. Oh, data controls?

1:06:58Grant:Okay. Improve the model for everyone. Turn that off.

1:07:04Corey:It was off. Yeah. So it defaults to off. Interesting.

1:07:09Grant:Just make sure this is off, right? Just make sure that's off. And that way, it doesn't mean that OpenAI can't still read the information, but it means that OpenAI is preventing the information that you upload from going into the next training run of GPT-6.

1:07:30Corey:Yeah, and honestly, there's probably a lot of stuff that provides less value than it used to in the way that grabbing everyone's conversation does. Yeah. I would think, but I click the same buttons.

1:07:44Grant:There's a lot of really good feedback in the chat and people are adding their own tips, So definitely read through those folks who are listening. So here's the other thing we can collect.

1:07:53Corey:We could share your tips too.

1:07:54Grant:Yeah, no, that's good stuff. I mean, it's stuff I would probably recommend. The one I read is something that I would also recommend. It just isn't part of this presentation. But as you can see here, so here's what I found. And more importantly, how I think about your search strategically. So it says where there's a market for you. And it gives me some links. Now, what I would do instead is I would update my instructions and I would say formatting rules. I'll add down here whenever giving me or whenever I request job postings or looking for jobs, list the direct URLs to the direct job postings with the job description in a bullet point list so I can easily click to apply.

1:08:50Grant:And I'm just going to save that because this is great. like okay cool it's giving me this but what i want is just a url that i can click go check out and go yeah um yeah so that's kind of part of level three so let me ask it can you read the current instructions about job job postings and try that again let's see if it updated in real time if not I'll just start another chat. Yeah.

1:09:26Grant:There you go. It updated. You're right. Let me get you the direct links to actual job postings you can click and apply to. So this is key. So whenever you get, whenever it comes back with a format of what you didn't want, give it, uh, give it updated instructions for how you actually want it to, um, respond to you. Yeah. A hundred percent. Yeah. Um, Maybe this would be a good time to just cover anything we haven't covered about prompting for level two. Since I'm realizing if we're considering this level one, then we should probably talk about prompting. Yeah.

1:10:00Corey:Yeah. Yeah. You know, there are a number of things I would say about prompting. If I was going to give a couple simple rules.

1:10:10Corey:The hack that's going to make it all not matter is going to come at the end of this. But I would say, number one, think clearly. remember not to write the way you speak out loud you know in the end this is a computer your language needs to be clear for it to understand what you do we speak with a lot of analogies and and other goofy phrases that wouldn't make sense to a computer that thinks more literally so i think it's important to watch out for things like that in your language i wish i had a good example I don't have one handy. I can pull one out from the blog.

1:10:46Grant:So this is all based on, this presentation is based on this article we wrote this week, the five-level AI proficiency stack. So level one is setting up your projects, which we just covered, and custom instructions. There's great links to this. I'll share this. Which is also prompting. Yep. Yeah, it is prompting. Yeah. So to your point, Corey, maybe level two, level one is still prompting. But here's some good examples. So once your project is set up, prompting gets much simpler because half the context is already loaded. You don't need to say, I'm a marketing director at a fintech startup every time the project already knows because it has your current resume and your instructions that tells it everything about you.

1:11:30Grant:For the prompts themselves, the simple formula is persona task context format. Honestly, cut this out. You don't need this. If you have custom instructions, you don't need to give it a persona. You just need to give it task, context, and format. So analyze the sales data.

1:11:48Corey:Unless for some reason you want it to have one. You know, I mean, if you, you know, specifically want that one to be a... Yeah.

1:11:56Grant:So analyze the sales data and identify the top three trends. In this case, you know, if we have a product set up, you could say, analyze the sales data in quarter three sales data.

1:12:11Corey:And then that's referencing here. Yeah.

1:12:13Grant:Something like that. Context. This is the Q1 2026 data for our SaaS product. We're preparing for a board meeting. I would even be more specific than that. I would be like, I have a board meeting in three hours and I need to create a full spreadsheet. analyzing all these data across XYZ. Yeah. Use case. I don't know. Whatever. Across XYZ. And then format presented as a one-page executive summary with bullet points. So in this case, I would say the format needs to be a spreadsheet formatted exactly like the example in example.

1:13:03Corey:spreadsheet template.

1:13:05Grant:And then, you know, upload a previous example of how your reports look so that it actually knows what you're trying to produce and it produces it the exact same way.

1:13:12Corey:I still have such a tendency to just copy and paste it and drop it in there and say, do it like this. Follow this format and paste. I do so much of that. And then just yesterday, we were working on data about all of our sites because we're the Neuron Zone by Technology advice we have 60 or 100 websites i don't remember they have a bunch and uh we were going through a list of some of them and trying to make sure that we had like who's the audience here on there and so what i did is i grabbed one that was really good and i went in and i i dropped it at the bottom of my chat window and then i went and emptied the form went up here and said here's the blank form here's a good one and uh and i said go check this site fill out that form and uh and it you know does the thing uh i think there's a way this is this is a team account so maybe you

1:14:08Grant:can't do it on a team account um or maybe you can there's there should be a way either in security or data controls where you can download the data like download all your previous chats cory do you know where that is on chat gpt because i saw a question about this

1:14:26Corey:Oh, where you would download your archive. It'd be in your settings.

1:14:30Grant:Yeah, I did this recently.

1:14:33Corey:It might even be in data controls.

1:14:36Grant:That's what I was thinking, but I don't think we have access to that on it.

1:14:40Corey:Oh, well. Archived. You have none archived.

1:14:45Grant:I don't think that, because that just, like...

1:14:47Corey:I knew there was a way to do it. Yeah, I don't know if it's possible on the team account.

1:14:55Grant:or perhaps ChatTBT changed it because a lot of people are switching to Cloud.

1:14:59Corey:Yeah, we are on the business account too. That's the other thing. So ours is a little different than yours will look.

1:15:06Grant:But yeah, for prompting, you can get more specific with it if you want, but really the main thing you need to internalize is this. You need to give it a task. And I would probably also add a goal, but that can be added to your custom instructions for the project you're working on. like my goal is you know i am making a i am a newsletter writer and my goal is every day i need to make a newsletter covering the top ai news that has happened in that day um that's the sole purpose of this whole project right uh yeah task context and format this could also be um you know presented as a bullet point list like any constraints that you have on the project and what it needs to look like do not forget that um yeah it makes a big difference in the quality of

1:15:54Corey:what you get back and like cory said sorry cory no you're good and then like cory said uh caught

1:16:03Grant:like context it can be something you just add and upload you know right here you can even paste the nice thing about claude is like for example let me just copy and in fact you can do this i give everyone permission to do this um you can go to you can go to our website copy all of this in copy and paste it into a project that you start and you can say like help me set up my ai setup exactly like this project is set up or exactly like this article says um and you can just do that

1:16:39Corey:yeah and as you get a little more in the weeds like i'm doing things with like that's one of my things with open claw that made me fall in love with it so much is that i'm taking like if I see a research paper I like and it's something that can be and I'll drop it in my open claw and say hey is this interesting what what how could this benefit what we're doing and it'll come back with we could do this we could do that you want me to implement that and like we've had there was one night open AI dropped a research paper right now I can't even remember for the life of me what it was on but like I was at a Mexican restaurant and I saw the paper and I was like well that sounds cool so I dropped it in telegram to my open claw at home and uh and And it's like, do you want me to do that?

1:17:19Corey:And before I finish dinner, that research paper that just dropped what it was suggesting was implemented on my home computer. Yeah.

1:17:26Grant:It's pretty sick. All right. So Deb in the chat said, can Claude create concepts in mind maps? What I said is make one. In this case, it has access to a plugin that I gave it. And we can talk about that when we talk about skills. And what it's going to do is it's going to create a mind map concept map based on my resume. So I've never read this resume. I'm interested to see what happens. In fact, while it's doing that, let me go ahead and see. So here's the recommendation it gave me based on our blog post. You're inside a cloud project right now with custom instructions, reference documents, and a structured workflow.

1:18:12Grant:That's level one. and then it gives me advice on what to do from there, which is cool. Now, how did I get Claude to do these to use Excalibur? Well, what I did was I went to this tab called Customize. And on Customize, you can connect connectors, which we weren't really covering this.

1:18:39Corey:By the way, did you notice Claude changed him to connectors recently?

1:18:42Grant:Is that not funny? I think it's been connectors for a while. Quad invented MCP. Yeah.

1:18:47Corey:Shared it with the world. ChatGPT used it, called it connectors. And now, Quad has changed theirs over two connectors. It's all still MCP. And it's actually the industry standard now. At least for the moment. But that's another story.

1:19:06Grant:Now, the way you would do it, if you wanted to add your own, is you would come over here to connectors. So, customize connectors. and then you click this little plus button and you hit browse connectors. And this will show you all of the publicly available connectors that you can use. And if you see an app that you already use, like let's say you use Hugging Fate. Well, that's a very specific one. Let's say you use Figma because you're a designer or you use monday.com because that's your project management tool. You can just hit a little plus sign here and that will add it to your account. And what that means is it'll have you log in and once you log in, Claude can always use the publicly available functions that Figma or monday.com give you access to, which as you can see here, Beehive, which is our newsletter platform, just updated their own tool here.

1:19:57Corey:I have my Notion connected to ChatGPT and Claude. That's cool. I can, from either tool, ding the Notion board that runs the website I built.

1:20:09Grant:Yeah, that's awesome. Love it. I thought so.

1:20:11Corey:I was like, this is great. I can have this one manage this part. I can have this one manage this part.

1:20:15Grant:But yeah, so I had the Excalibur plugged in, and so now with that, it can create a little mind map here, and I can go in and I can edit it. Now, I think if you just are chatting with Claude on the website, I think it can also create concept mind maps.

1:20:31Corey:You could probably tell it use only black text. You could tell it, you know, I want to make sure that these are all in circles and not rectangles. you could there's a i'm certain it would take the instructions and just go do it yeah because

1:20:46Grant:basically when you have this connector set up it can use the app that um that basically anything that the app allows uh over the connector it can it can use that function and the best way to set these up if you have your own apps that you're trying to connect uh through a connector is to To give what we talked about with Dan Shipper last week is agent native engineering. Yeah. And what that means is treating the agent as a first-class citizen who could use the app in the same way a human can. So any functions that I could use, any buttons I can click, any tools I can add, the agent can also use.

1:21:23Grant:So if this was set up in that exact way, then the agent should be able to do anything you can do when you click the edit tab, which means you can just ask it, make all the colors is blank and it'll do that. That's just a very brief overview of connectors. Corey, anything else on prompting?

1:21:43Corey:No, I think we're good, but I think we jumped over skills, didn't we?

1:21:47Grant:Well, let's get into that.

1:21:49Corey:Grant and I aren't good with linear thinking. You should know that.

1:21:53Grant:That's why we created this run of show document for ourselves to try and keep us on track here. I have looked at it today, Grant.

1:22:00Corey:You'll be proud.

1:22:02Grant:Thank you. Thank you.

1:22:03Corey:I opened it and I've even tried to follow along.

1:22:07Grant:So here's what we've covered, right? The simplest formula. We did a live example. We didn't really talk about this, but context engineering is just exactly what I said. So in order for the AI to help you, let's go back to our, in order for the AI to help you with your job search manager and to actually save you time and not have you manually writing cover letters for every job, it needs to understand your resume. It needs connections. Maybe it needs to be connected to your LinkedIn if that's possible. Maybe it needs to be connected to like other services that you use like Gmail or things of that nature so that you can pull in all of the context that needs to actually give you a good answer that's not generic, that's not just like, you know, what Google would give you that actually helps you accomplish the task and reduces hallucinations because it knows everything about you.

1:23:03Grant:If I was just to go to a random chat and say, write me a resume, it'll write me a resume for some person it's never met.

1:23:10Corey:A good prompt engineering course, by the way, I know we mentioned those a little bit earlier, will handle context engineering as well because it's, you know, getting the right context is a vital part of that.

1:23:21Grant:Yeah.

1:23:22Corey:A short course can be a handy thing. Speaking of which. Yeah. Yeah. Stay tuned. We might have a full announcement in a few weeks for you.

1:23:33Grant:I like that subtle point. That's really all I can say. I'll say now, so what I'm doing here is I'm iterating, right? So I have this set up the way I want. And what I'm going to say is now let's turn this into a skill I can call at any time I want to make a mind map in my preferred format. So this is the unlock. this is level three this is uh the thing that most people it's relatively new so most people haven't wrapped their heads around this yet um and it's the thing that's going to save you so much time so if project saves you a lot of time because it's organizing all of your structures and um organizing all of your chats and helps you so that you can do recurring tasks on a recurring basis with all the same information uh then skills are basically automating the actual prompting process So you don't even have to prompt.

1:24:29Corey:And your tool calls as well as any like reference docs. Yeah. You know, it's it's that's what kind of makes it a step past a project, I think, is that. Yeah. That it's not just your prompt. It's your prompt. It's the tools you want it to use. It's the they're really handy. They're not available in standard chat GPT yet. They are available in ChatGPT business accounts and in codecs. They are available in all of Claude's tools. I don't believe Gemini has skills at all yet, but I shouldn't say that. I wonder about this.

1:25:09Grant:I wonder if I was just in Gemini, and let's try this, if I could say, can I call my header image creator?

1:25:18Corey:By the way, if I'm honest, what I would do anywhere that doesn't have them, is I would connect your Google Drive to it and start a folder where it saves skill files for you and it would do it. It could reference them as you tell it to and it would really negate whether they offer them at all, I think.

1:25:35Grant:That's a really good tip. So what you're saying is like have a shared skill file folder and you can connect it to Google Drive and then that way Gemini can reference it. Yeah, it's connected to Google Drive. Claude and ChatQPT. Yeah, that's a really good tip.

1:25:50Corey:The only catch would be it doesn't do, you'd have to keep it from converting the MD files over, but that's not a big deal. You could absolutely do it that way. It's a little backwards, but I love these. Grant and I probably combine for 100 of them, if I had to guess, give or take. Maybe more. I haven't counted. I haven't counted. Grant will have more than me.

1:26:16Grant:That's funny. So when Linda said, I upload your newsletter to Claude whenever you have new info about him, generally he will thank me or say he didn't know. I love that. I think that's great. I think the best way to read our newsletter, because there's so much information in it, is probably to just upload it to the AI and be like, how could this help me?

1:26:35Corey:Yeah. Deb, I'd like to send your question here. So Deb Shiano asked, so whenever you access the skill, it will perform that specified skill with whatever you get of it to access. Yes. Yeah.

1:26:48Grant:Yeah. I'm trying to, I'm trying to get the cover letter customizer into a good point. Cause I think this will be a better skill to test.

1:26:55Corey:The truth is they're also relatively good at recognizing I need to call that skill, but it, yeah. Grant and I both still tell it to call the skill because two times out of 10, they don't, if you just assume it will. Uh, but, uh, Normally, I just say something like, hey, call my blog article skill or whatever it is, and it knows. So what's funny is, so basically, my chat is being very, like, reasonable with what's requesting from me, what it needs from me before it can actually do this.

1:27:31Grant:I hear you, Grant.

1:27:33Corey:That's my favorite response.

1:27:35Grant:Yeah. Basically, I was like, just make it up. This is for a demo. But it was asking for specific information it needed for me so it didn't hallucinate, which I actually really respect. I'm glad that it did that. Same.

1:27:50Corey:That's a thing I didn't do a year ago. Stop and ask questions.

1:27:55Grant:Yeah. It's so important. So what I'm about to do is I'm about to have it reverse engineer this chat where I wrestle with it. I try to get it to give me exactly what I want. What I want is I want it to give me the exact jobs that I am qualified for. And I wanted to write custom cover letters for each of those jobs. And I wanted to do that every time I, you know, start my day or whatever. So we're on step three of that process, which, or actually we're on step two of that process, which is I'm iterating with it. I'm asking questions. I'm, when it doesn't give me exactly what I want, I force it to give me what I want by giving it more context about what I'm looking for.

1:28:35Grant:in this case I said just make it up but you know you would want to tell it give it the information it needs to be able to do the job the way you're asking it to do any questions about this so far I know we've barely touched on skills we're going to explain it a little bit more in depth yeah the truth is it's not

1:28:53Corey:once you understand the other things skills aren't too hard to jump yeah but the unlock is big I think that's the real key with them is that And I would apply that same rule. If this is something I'm going to do more than twice, it needs a skill or a project or something that makes it quicker, easier. Make sure you've got the result you want and then have it build the skill.

1:29:20Grant:So while we're waiting for this to build out, someone asked in the chat a couple of minutes ago, how do you get individual ChatGPT chats into a project? Let's see if we can figure it out.

1:29:36Grant:I have extended thinking on. Okay. Hi, I'm here. Okay, great. So now I have a chat. Move to project. So it's the exact same workflow. So you hit the dot, dot, dot. You hit move to project. And then you can add it to job search manager. And voila, it's in there. Cool. All right. Let's see if our... Okay, great. So as you can see, so it gave me a lot of stuff here. So what am I looking at? So I've got a cover letter for each role, right? For each role. And as you can see, each of them are customized, taking a minute to load. But for example, I could open it in my text editor. And I can then go in and edit it myself.

1:30:26Grant:And then what I can do is I can tweak it. I can edit it. and then I could re-upload it and say, okay, use this version that I've edited. But here I have a workflow where I've wrestled with it and I said, you know, this is what I want you to do. Can you look up jobs that I can apply for based on my resume? Can you create custom cover letters for those jobs? Can you give me the URLs? And so now what I'm going to say is I'm going to say, now can you use your skill creator skill to turn this conversation into a skill that I can call anytime I need to look up jobs and create custom cover letters for those jobs.

1:31:11Grant:Be careful when doing that as it's easy to drop a chat into the wrong project. Good call, Lexi. Appreciate that. Linda says, oh, Claude says your newsletter is well-respected in the AI community. I hope so that's nice sometimes I realize like not everyone in the ag community reads us but a lot of people do but it's funny when we come across people who don't it's like oh hi

1:31:36Corey:I was at NVIDIA GTC a couple weeks ago and it's the first time Grant and I always joke about we go to these trade shows and it's funny how we just never run into anybody that's like oh we're with the neuron and they're all like huh but In VideoGTC, we found the neuron readers.

1:31:54Grant:Yeah. It was like, oh, here's where you are.

1:31:58Corey:It felt like every time we turned a corner that week, somebody would come around. It was really cool. It was nice to meet some people.

1:32:05Grant:Yeah. So what's happening here is ChatGPT actually has, sorry, not ChatGPT, Claude, has what's called a skill creator. And if you were to go to customize, and instead of choosing connectors, you choose skills, Now you can see a series of skills that we have here, but you can hit plus, you can hit browse skills. These are actually skills that other people have created that they've made public. So perhaps you don't know what type of skill you want to create. Well, you could come here and there's some that are already for you. This one's a really good one.

1:32:41Corey:I'm always a little careful with other people. There's also skills.so.

1:32:44Grant:So this is from Anthropic. So these ones are safe.

1:32:47Corey:Okay, these are Anthropic built. Okay.

1:32:49Grant:Yeah, exactly. And what I wanted to reference was this one, which is the skill creator. So what it's doing here is it's actually calling this skill creator that creates other skills. And as you can see, a skill is a series of folders and marked on files. Yeah, but it also has code that it can call to do stuff for you. So you can totally analyze this and reverse engineer it and make your own version. But as you can see, the skill is like a series of Python files and Markdown files and HTML that helps it create other skills. And you can look at the same thing here for how all of these others are set up.

1:33:36So let's go back.

1:33:38Grant:So that's what's happening under the hood when you're calling the skill creator skill. So let's go back to our current conversation. Actually, let's go to job search manager. Okay, so let's hit the bottom. All right, so now what it's done is it says, here's your package skill. To install it, download the skill file. Actually, what Claude does, which I really, really like, is they have this what's called save skill. So I don't have to download this. I just can hit save and it will copy it to my skills. So now, as you can see here, job search cover letters. Boom. There it is. Now let's go back to our project folder.

1:34:17Grant:So back over here.

1:34:20Corey:Yeah, Robert, there will be a, this'll be archived on YouTube forever and absolutely available to watch. We'll have some follow-up materials as well. I just wanted to respond because I saw you look like you might have to go.

1:34:32Grant:Go look up the top jobs that I can apply for today. And let's actually see if it does, it calls the skill.

1:34:48Grant:Let's see.

1:34:49Corey:Grant, how are skills different from an agent?

1:34:53Grant:That's a really good question. you can think of skill as like giving an agent everything that it needs in order to accomplish a task so yeah the skill is more like the organizing principle and all of the files and all of the tools that the agent needs to go out and accomplish a job yeah did it actually call my skill yeah read the job search skill now let me search for jobs matching their profile that's great so now it knows exactly what I want and let's see how well it adheres to what I'm looking for one other thing to comment about this so let's say it doesn't call it right away you can now go and change your instructions to say whenever I ask for job what did I have that before that I edited format and structure well you can even say at the top right core behavior decision making protocol okay it's got this good stuff here yeah I would just throw something in here or I could even copy all of this into the chat and I'll say like hey can you update my instructions for this project to use the newly created job search skill whenever I ask anything job search related and then I'll just paste that in and then it can then go and give me the updated file so I don't even have to mess around with that I can just copy the new version and replace it so let's go back to the other chat over here cool so all right so what it did is it found the direct job postings.

1:36:44Grant:These are new ones that we haven't seen before. Pre-filtered search pages. That's nice. Lot of openings. Now, what it didn't do is it didn't immediately start writing the cover letter for me. So what I would do is I would go back to this. And oh, wait.

1:37:06Corey:I was in actually I was in the wrong chat.

1:37:11Grant:Okay, let's update this skill so that it always automatically makes a cover letter for every job that it finds.

1:37:26Grant:Now this will then update the skill and I can show you how updating skills works. While that's working, let me show you how to do this on ChatGPT. So if you're using ChatGPT, what you can do is you can go to skills. So click on your profile, bottom left corner, go to skills, and they have a skill creator skill. But what you can do is you can do new skill, create with chat, you can create with an editor, or you can upload it from your computer. So if you follow Corey's advice, what you can do is you can create a central skill repository for yourself, like on Google Drive, drive.google. and then you can upload all of the skills that you've created in Claude or ChatGPT and upload it there.

1:38:11Grant:And then you can just transfer them to chat like that. Oh, or use your connector or you can use a connector. That's right. Yeah. Yeah, that's right. You could just go to, you could go to job search manager. And if you have uploaded your drive to ChatGPT or sorry, if you've uploaded your skills to your drive, you could say use the google drive uh connector and pull my job search skill um to this chat or something like that yeah that'd be a way to do that or reference i would just say reference it even it wouldn't even have to bring it in probably yeah and reference that's my theory at least uh i very much just made that up while we were talking but as you can see like there's you can also share skills across our company which is a huge unlock so as you can see right now our company doesn't have skills but we probably should work on that um yeah i've not i've no created skills yet on here but i've installed the skill creator skill i have and uh i think it's just that we don't have any

1:39:15Corey:company created ones yeah yeah we need to fix yeah yeah well you're gonna do a whole uh presentation

1:39:22Grant:on that in a couple weeks so that'll be good actually it's a good reason to do some where do find corey where do you find new skills like doesn't chat just have a skills library that

1:39:31Corey:you can yeah uh i don't know if there's one a library or not i just tell it to create them

1:39:36Grant:and it does yeah but you would do the exact same process so like let's say you have the exact same chat and chat tpt you can say um use your skill creator skill um to create a skill for this um chat um yeah i don't even tell it that much what do you say just like make i just say i just say make this skill yeah and so here's what i did uh on claude what i did is i had it update the job search cover letter skill and then it gave me a new skill here so what i'm going to do is i'm going to hit save skill and then it gives me the option to upload and replace it so i don't have to go in i don't have to edit all those files i don't have to do any of that it just automatically has updated the skill for me in the skill manager, which is great.

1:40:28Corey:Grant, do we want to take a couple and run through these last couple of things now that we're about an hour and 40 minutes deep?

1:40:35Grant:Yeah, that was supposed to be like five minutes.

1:40:39Corey:Honestly, automation's ought to be its own thing.

1:40:43Grant:Well, let me just show you scheduled tasks because I think that's important. So now that we have our job search skill, I'm going to cover level four, which is... All right. So level four of this is automation. So once you've got skills you can call anytime, now you can schedule them for recurring tasks. So you can do this in Clawd or in Cowork. And Corey, can you do this on the ChatGPT app?

1:41:17Corey:Are they in Claude's website, Grant?

1:41:21Grant:Where are the skills or the automations?

1:41:24Corey:They're not in the, I don't think. Yeah, automations. Are automations in Claude's web UI?

1:41:30Grant:This is the web version in ADS. So I think if you're on ChatGPT's web app, you might be able to do it, but you can definitely do it on what's called Codex. And Codex is ChatGPT's coding app. but don't be intimidated by that. This is a great place where you can do scheduled automations. I use it to write.

1:41:54Corey:I use it to manage my banking. I use it to give me cooking recommendations. I use it for about everything now.

1:42:03Grant:Yeah, and full disclosure.

1:42:04Corey:And it looks like you're moving toward this as a super app anyway, so at some point in the interview, we all will be.

1:42:10Grant:Yeah, so eventually the ChatGPT core app will have the exact same features that this has. which will be great um so but yeah the the difference is um instead of scheduled tasks chat gpt calls them automations so now i'm going to go back to the claude app here um and i think actually for this one you might need to be in co-work yes okay so this is why you need the chat

1:42:35Corey:or the claude app the desktop app yeah that's why i was asking i thought it may be that neither of have it in the web UI.

1:42:43Grant:Yeah, let's see.

1:42:44Corey:Yet.

1:42:47Grant:Let's see. Job. I don't know how to get my job search manager over here.

1:42:58Corey:Is it not?

1:42:59Grant:Oh, you know what I can do? I think I can do this. So let's go to projects, job search manager, start a new task in Cowork. And then I give it permission. Yeah, so this is the flow. So now you've got your project set up and you've got a skill that you can use inside your project. What you do is on the desktop app, you start a new task in Cowork, you give it permission, create a task, and then you say, hey, every day at 6 a.m., I want you to run the job search skill and find all the new jobs that have been posted today.

1:43:40Corey:And you can give it restrictions like it needs to be above$80 ,000 in the greater San Francisco area. It needs to offer a comprehensive health insurance package if you wanted. And it can filter that down for you so you don't get a lot of trash and you get jobs that you care about.

1:44:03Grant:And then, does it give me permission to work in the project? No, so I'm just going to go with it like this. so what happens in co-work now is co-work knows the project because i've said to start a new task um it's going to get this set up for me it's going to check my resume and then the relevant skills so the scheduled task knows exactly what to search for and it will put this together yeah and it'll be very cool and it won't take long no it's pretty nice it also gives you on co-work

1:44:39Corey:it gives you like a little progress report that's your real phone number no it's not

1:44:45Grant:yeah don't call this this isn't me whoever this person is i apologize for doxing you

1:44:50Corey:it's all right if anyone has the number three three three three three three three three three

1:44:55Grant:they're already that's funny so now it's going to create a scheduled task and it's going to give me the option here. It's going to say, search for new software engineering jobs in the greater San Francisco area with health insurance, generate cover letters, schedule. So it's 6 a.m. every day. But let me do it. Let me update it. So I'm like, hey, can you actually run that every 15 minutes? And it will change it.

1:45:23Corey:I want to answer a question real quick here from sorry if I mispronounced that I'm certain I did you're probably not finding them in chat GPT because you may not be on a business account they were just rolled out in business accounts a month ago and they should probably be rolling out broader if it was today I wouldn't be shocked but it could be a month I don't know but I gather that they're coming very soon but if you download their Codex desktop app it's right there really easy to use and all of this. So the desktop X has it. The web UI does not yet.

1:46:01Grant:Yeah. So what I would do is don't be intimidated by codex. Go to, Corey, could you drop the codex link, the download link in the app? Sure. And I would do it that way. And it has pretty much the exact same thing. You can start with a project. You know, in this case, it wants you to use a project that's actually on your folder that, or sorry, a folder that's on your desktop. But you can just do a new folder test.

1:46:26Corey:Yeah.

1:46:27Grant:I don't know if you can see my screen because it pulled up my file manager. But then you can do it here and you can say, like, let's create a skill to track job postings related to my resume. And let's see if it lets me do it.

1:46:51Grant:okay so i'm going to scaffold over usable codex skill for job posting tracking starting by reading local skill creation guidance and that'll create yeah so perfect it knows exactly what it needs to do it's reading the skill creator skill and it works the exact same way and anyone can do this i think i think it's for the codex app is free right yeah yeah yeah so you don't need to have

1:47:11Corey:It'll connect to your normal chat GPT account. Yeah. That's the other thing. So if you have a chat GPT plus subscription for$20 a month, it'll connect to that. And it's just going, man.

1:47:23Grant:It's just writing stuff. It's doing its thing. Yeah.

1:47:27Corey:So I legit have codecs building AI models for me right now. Yeah. Crazy. It's, it's simple. It's easy to understand. I'm not a software engineer. I always do, but,

1:47:39Grant:But I did stay at a Holiday Inn Express last night. If you remember the old commercials.

1:47:44Corey:No, but that's hilarious. That was it. We need a doctor. And a guy comes running over and they go, are you a doctor? And he says, no, but I did stay at a Holiday Inn Express last night. Hilarious. Yeah, they were amazing. I just dated myself, but someone will get it.

1:48:00Grant:So let me just show people. So remember, if you remember a couple of minutes ago. And oh, that's nice. Notification.

1:48:07Corey:Yeah.

1:48:07Grant:Click that away for now. So I said, hey, every day at 6 a.m. I want to run a job search skill to find all of the new jobs. And then I asked that, hey, can you actually make it run every 15 minutes? So now I can go to scheduled on the left-hand bar here, right under search. And I can see my daily SF job search. It says it's active. It says next run is at 12.03 p.m. But if I actually want to just run it right now, boom, I can hit run now. and now you can see the scheduled task and it's running the search yeah so uh someone asked i'm not finding skills on chat tv you already answered that i know which project without specifying does it know which project the only reason it knew that is because i went to the project um and opened it from there and i said start a new task in co-work so it already knows to use this project folders context to start the task inside co-work yeah um and that's how it knew and then so now i go back to co-work and we can see my scheduled task is running here so it's got the skill md it's running the commands it's searching the web it's finding me my job postings okay and we'll see it outputs yeah do uh do we want to hold off on agents and do like an

1:49:34Corey:agents 101 well let let me just hate to halfway do agents let me give a brief overview everything

1:49:42Grant:that we're doing here we're interfacing with an agent the agent i've given the agent a goal the agent is going off and it's uh you know following the instructions that i've set for it but it's making its own decisions here it's reasoning it's um you know it can't access certain things so it's searching uh again it's it's it's doing all of the decision making so in order for me to make an agent uh to accomplish what i want you can do level one through four and just using Claude or the Codex, you're working with an agent. You don't need to use any sort of complicated agent builder. You don't need to use any N8N.

1:50:21Grant:You don't need to use the agent kits. This is technically working with an agent.

1:50:26Corey:Yeah, you're right. Technically, these automations were exactly what every company was billing as AI agents two years ago, except they were much more complex to set up and build. Grant and I talked about it a number of times along the way that it's really soon. You're just going to say, hey, build me an agent that does this. And now that's what we have. We have something I'm willing to call agentic finally.

1:50:51Grant:I mean, I'm using the co-work agent, let's say, to create a scheduled automation. But I've essentially created a daily job search. Yeah.

1:51:01Corey:But an agent lives within that automation that is pre-qualifying those for you. Yeah. So your agent created an automation that contains an agent.

1:51:11Grant:yeah exactly exactly and this is why like the cloud code leak right that happened yesterday was kind of a big deal because uh anthropic puts a lot of effort into um creating these agents and creating all of the ui and the and the user experience elements that make this uh work really well um and now you know everybody is like basically shared that and is learning from how

1:51:36Corey:There were some serious improvements this morning. I was watching a few things, and they were like, oh, do you want to not worry about memory anymore? Do you want to know? I've seen it now duplicated into Python, duplicated into Rust. I've seen Claude Codex, where it's blue, and everything that's orange is blue, and it runs on the Codex models. I've seen it. Wow. Yeah. Not to mention the other companies Who are reversing I say companies It's probably more I say that They're all looking Because if anybody leaks anything You're going to But Yeah It's really interesting But Organizing ideas Their two competitors Are both open source Yeah

1:52:20Grant:That's true

1:52:20Corey:You know Codex and Gemini Are both open source Well at least their CLI is And this was Coworks This was Cloud Code's CLI Not the app And they didn't leak like the codex model weights we're specifically talking about the harness around cloud code that

1:52:38Grant:makes cloud code good yeah exactly so so if you want to do all this with like a coding agent right you would just do it all in code yeah uh instead that's how you do it yeah but it has pretty much

1:52:52Corey:for those of you discussing m dashes earlier uh at least in codex and i haven't tested it in Claude code, but I'll bet it's true there too. There are no M dashes. Because as Grant would say, say it, Grant.

1:53:07Grant:Oh, it's not going to tolerate that. Code bases don't need M dashes. They're not going to tolerate that.

1:53:13Corey:No. No, they're not. Which means somewhere along the line, it's been figured out how to not have M dashes. I'm still angry about the M dash. It was like, as a journalist at small town newspapers, and working my way up through my career to bigger publications. You know, the M-Dash was, I mean, I made M-Dashes every single day in every single article I ever read. Speaking of them. Speaking of M-Dashes, hello there.

1:53:42Grant:Yeah, they're here on the screen. But yeah, so this is a scheduled automation. It went ahead. It created cover letters for all of the jobs that it found, for jobs for Chime, for Salesforce, for Flex, Prelim, Cargo, Gusto.

1:53:58Corey:Um, this is the, uh, the job seekers revenge on applicant tracking software.

1:54:04Grant:Yeah. And it's, I mean, now look, this is very basic. It just created the cover letters. You still have to go in and apply, but you know, in theory, you know, you get creative and you can figure out how to make the agent actually apply for you. Um, but look, this runs every day, um, at, uh, every 15 minutes. So I can be applying to, I can be creating new cover letters for new jobs that are legit jobs from the internet every 15 minutes. And actually, you can do these scheduled tasks every minute if you wanted to. Now, you're going to run out of tokens when you do that. As you can see, I'm on the max plan here, so that does make a difference.

1:54:41Grant:But yeah, you can create scheduled automations for anything you can create skills for and projects for. If you're on the$20 plan, yeah, you may well hit some rate limits with some of what you've been shown today.

1:54:56Corey:if you start using it extensively. But they have a mid-grade plan for$100, right, Grant?

1:55:04Grant:Yes, and I think people have done the analysis to say the$100 plan for Claude is probably the best value that you get versus the$200 plan. But on the$200 plan, I rarely have usage limits. It's very, very rare that I ever get stopped from my job.

1:55:26Corey:Contrast on the$20 plan, I only have usage. Yeah.

1:55:31Grant:Yeah. If you want to do anything serious with AI, you probably don't want to be on the$20 Claude plan. You want to be on the$100 minimum.

1:55:40Corey:You can buy with ChatGBT's$20 plan a little longer. Like their rate limits are higher. They've kind of strategically bought a lot of...

1:55:50Grant:They've raised more money, frankly. They've raised a lot more money so they can burn it more.

1:55:54Corey:Yeah, exactly. So, you know, I hit them less there. Well, actually, I rarely hit them, but this month is hard to judge on because it was double rate limit month and they've been resetting them every third day.

1:56:10Grant:Hey, one thing I just did real quick. I turned this off, so it's going to pause. But what will happen is if you have a scheduled automation, it will still show up here and it will just say skipped. So, for example.

1:56:22Corey:If one's missed or something. Yeah.

1:56:24Grant:So I paused it. you probably don't want to delete it if you're going to use it again i would just turn off

1:56:28Corey:repeats yeah and then you know you can just keep that that makes sense um somebody had one question

1:56:34Grant:here too yeah i think you're gonna flag the same thing i was gonna flag go ahead you prefer cod cloud code or codex i mean wouldn't you say we would answer differently yeah yeah we would

1:56:44Corey:absolutely i i say we'd answer differently i don't i don't believe that either one of us dislikes either of them well i actually really like codex but i use claude code as well uh and if i had to guess grants the opposite yeah i mean uh so this this brings us to part

1:57:03Grant:two of the the guide and i'll pull that up uh and part two which we schedule down so this is kind of everything we've talked about the only thing we didn't talk about was like building an agent um you know there's different ways you can you know if you want to get into coding and engineering you can do that. But I think for absolute beginners, like working with Codex or Claude Cowork is like the two easiest ways to create custom agents for yourself. So I'd probably go with that. Now we did another stream about using agents inside of Copilot. So some people who were on here were wondering like, well, what do you do if you can only use Copilot?

1:57:42Grant:And Copilot has what's called Copilot Studio. And that's where you do a lot of the more complicated agent building stuff and so you should check out that stream i think it was renamed to copilot agents

1:57:53Corey:uh oh did they rename it maybe i can't remember i could okay we'll figure that out if you could uh

1:58:00Grant:if you could drop the link to that at some point that'd be great yeah i'm gonna try to find that

1:58:04Corey:now we got a couple important questions that i think yeah really good really good wrap-ups let's do it um nice things nice things claude co-work is only available on mac not on windows uh no well Claude Cowork is available on Windows now. Yeah. As of a couple weeks ago? Three weeks ago? It hadn't been long. Maybe a month. Also a wild thing. Next question, if you can only afford$20 a month, what do you recommend for the best quality value? Claude, Gemini, or ChatGPT?

1:58:35Grant:Oh, without question, Corey, you answer.

1:58:38Corey:Yeah, it's ChatGPT. Yeah. You're going to get the most bang for your buck out of ChatGPT. A thing ChatGPT has done really well that, in my opinion, and the others have not, is ensure that good quality stuff is accessible to about anyone. That's like a core part of their mission, and they've really kind of stuck to that. Like even the free version. The free version is very much, it's weaker. It doesn't have all the things, but it is there. You can use it for no money. And that means that even without money, you have access to some level of artificial intelligence tool. And I think that's really important.

1:59:13Corey:But their$20 plan is a great plan. you really get a lot like you get nearly all of the tools uh i i never hit rate limits in my daily chat gpt use and uh if you're new to it i promise you you're not going to use it more

1:59:30Grant:than me in the short term future yeah so cory hasn't gotten uh rate limited on it you probably won't either yeah um but but yeah i would say like you know with the 20 plan if you really want to do these like automations and really speed up your work that you're trying to do um use the codex app don't be intimidated by it i know we've done a lot of streams recently where we're talking about like you know everyone can code now don't even worry about the code like like just worry about they've got this nice project folder set up you can create custom automations you can do plugins which you know is what connectors earlier so you're just going to tell it how you want to use it and have it walk you through it yeah and you can literally do that yeah so i i think it's and they know that this is a better app than ChatGPT and they're trying to combine ChatGPT

2:00:17Corey:and their browser Atlas and Codex into one app.

2:00:20Grant:So in a couple weeks, maybe there won't even be one single app. Sorry, maybe there won't even be a difference between Codex and ChatGPT. I don't know how they're going to do it.

2:00:31Corey:Yeah, I don't think there will be pretty soon. I think it'll work a lot like the way that Quads does.

2:00:36Grant:I see a couple other questions. Maybe what we're about to run through, we'll cover these. but if not, we'll cover them at the end. So here's part two. So step one, pick your daily driver. We've kind of talked about this a little bit. We recommend, based on everything we've showed you here today, either doing ChatGPT or Codex or Claude, right? So test both of them. You know, money is no object. Test both of them. See which one you prefer. If you're cache constrained, just stick with GPT. But then once you've selected those two, pick your platform. So for us, the Cloud desktop app, that's what we've been showing a lot today.

2:01:16Grant:If you're using Cloud, if you're using OpenAI, use the Codex app because it's really great for everything that we just did and can do everything that we just did.

2:01:25Corey:And has features you're not going to get to in the web UI, in the website.

2:01:30Grant:Yeah. And someone said, Leno, like they'd rather use Cloud ethics wise. Like that's been a topic lately um just be careful with the 20 tier because you you know you could run out of um usage pretty quickly because claude is you know they've they've been struggling with staying up because they've gotten so popular lately so you just gotta keep keep in mind of that if you if your work relies on it i'd probably try and get your work to pay for the hundred dollar tier um if they're open to it because like it's just you don't want to be you know on deadline you know have like you know five minutes before you need to turn something in and then it's like your usage

2:02:06Corey:is up for today it halted mid mid coding task on me uh it's it's those rate limits have burned me a few times and uh i've just reached the point where what i care about more than anything is i care that it works when i need to work and uh you know the models are relatively indifferent

2:02:28Grant:um and now if you do like gemini oh yeah like gemini or grok better we just don't find the platforms are as easy to use as what we just showed you today but they're both perfectly fine like we showed you like you could use gems um you can schedule gems with opal which we haven't shown you yet but we can pull that up in in a minute um you're not even going to notice a big quality difference yeah i mean someone asked earlier about where gemini lands it's technically on par

2:02:57Corey:intelligence wise with these two that we showed it's just i think the app they just need to retool that app it's just not not what it needs to be yeah um and they have a big uh conference in a

2:03:09Grant:couple weeks that cory's gonna go to so probably they'll announce something there too yeah something

2:03:13Corey:come out of that i'll be at google cloud next i know i'm meeting with their head of startups and i'm not sure who else yet but uh we'll have whatever's coming we'll we'll have it to you as fast as we know it.

2:03:26Grant:Yeah. Corey, you might, you feel free to quibble with any of this because I wrote this. You didn't, you didn't give your seal of approval here, but here's our current ranking of which of the major apps is best for what, right? So chat GPT basic app is good for one-off web searches and image generation.

2:03:43Corey:I would do image gen and intensive research or strategy.

2:03:48Grant:Intensive research and strategy. Okay. That's fair.

2:03:50Corey:I've done a lot of, of, of really in-depth, like go away for 40 minutes while it does research stuff and come back and been pretty floored. It's also really good at math, apparently, because it's dropped. I think it is now I went and updated I have a story that keeps a running log of all of the unsolved mathematics problems that have been solved and most of those have been. By GPT five to or GPT five for and it's it's large there were three more this week already certified they're confirmed. And yeah, I updated that. I think I'm tracking 18 total now, 12 of which are full solves, a couple of which are like good contributions to mathematics.

2:04:39Corey:But maybe it turns out that some of that was already solved and they didn't know. And a number that are also specifically credible but not verified, like ones that there's a good chance this might turn out to be true, and we're watching it.

2:04:54Grant:I know Gemini's Ultra version, which is like their$200 or$300 plan.

2:05:01Corey:You know, it's really intelligent as well. Yeah.

2:05:03Grant:I haven't seen as many like Gemini has solved like this XYZ, like unsolved math problem. So if they are solving it, they need...

2:05:11Corey:Those have been pretty specifically the GPT-5 Pro models.

2:05:14Grant:Yeah.

2:05:16Corey:Or is it Alpha Evolve? I think Alpha Evolve's got one.

2:05:20Grant:Okay, cool.

2:05:21Corey:I think I know that there are a couple that have passed through Aristotle which is owned by Harmonic we're going to probably have an interview with them at some point in the near future Karina Hong at Axiom Math has contributed to some but it's happening there it's happening in Sciences it's happening in so anytime anybody tells you well it can't solve unsolved problems they're just unequivocally wrong now it's absolutely a thing that's happening and it's there was a really good uh interview with

2:05:57Grant:terence tau where he was talking about this where he basically said like right now like imagine like the unknown right when we're when we're trying to solve unsolved problems uh it's sort of like we're hiking in the dark and there's all these different cliffs that we're trying to climb and some of them are small and some of them are really tall and we don't know the difference because we're hiking in the dark. And the AI is kind of like a jumping machine. And the jumping machine can sometimes get to the small cliffs. But we just don't know how tall the cliffs are. And so we don't know. And the jumping machine has a limit.

2:06:30Grant:It can only jump so far so far. But eventually, you know.

2:06:34Corey:But it's going to jump better over time.

2:06:36Grant:Yeah, exactly. All right, let me run through these real quick. So Gemini is best for accessing YouTube links. It can watch and summarize full videos, something no other model can do because obviously Google owns YouTube. Creating images. Go ahead.

2:06:52Corey:Not anymore.

2:06:53Grant:What do you mean by that?

2:06:54Corey:I mean, Google still owns YouTube. Wait, no. ChantGPT will analyze video, but I don't think it can do a YouTube link.

2:07:00Grant:If you need your AI to understand a YouTube video, you have to use Gemini. Like, you just have to use.

2:07:07Corey:Yeah.

2:07:07Grant:And that's the same for Grok, which I can cover right now. if you wanted to read anything that like people are writing on x just send grok there like like that's really the best thing that it's for is just reading those links um and on the free account it can read about 20 links per chat um about eight chats so you can read about 160 links a day with

2:07:29Corey:grok um for free is that if you're on a free x plan that's free yeah you don't have to pay you

2:07:36Grant:on if you don't want to. So then accessing YouTube, creating images. We showed you earlier, Gemini, you know, with our header image. We did that by, you know, giving it sample images that we really liked and trained it with a gem. It's using a model called Nano Banana, and then building automations with Opal, which if people still hang out after a couple of minutes, we can show that. And also deep research. It started with, it started deep research as Logan Kilpatrick told us. I find that sometimes the output from the other models is better, but it gives you a lot of links because it's using Google search.

2:08:17Grant:So the link quality is good. The report text itself is slop or can be sloppy. So use it for the sources and then rewrite it if you need it. And then Claude is really good for deep research for niche links. This is because Claude hasn't signed all the same deals that OpenAI and Gemini have, and so it can't read as many websites. But if you need a really niche topic, deep research is really, really good for that because it can pull diamonds in the rough websites. Working with files on your computer via co-work, scheduling recurring automations, and writing. I said, I think Claude is the strongest writer of the bunch, though Corey would disagree.

2:09:01Corey:Corey disagrees. Corey would say Claude used to be the strongest writer, but hasn't been since 3.5.

2:09:07Grant:Yeah, that's fair.

2:09:09Corey:It's a good writer. I don't mean it's not a good writer. I just mean it's verbose. It talks a lot. Everything it does is really long.

2:09:17Grant:That's fair. Yeah. Claude's skill system, I said, was the most mature. Although we showed you today with Codex, you can basically do the same thing with skills. So I think ChatTBT will quickly catch up there with the main app. So just hang out a couple weeks if that's your daily driver. And then, yeah, Codex does pretty much the exact same thing with scheduled automations. I think it's better at organizing your coding work, like with the project format. Like Cloud Code does not have that.

2:09:46Corey:The thing I keep hearing is use Codex if you want it right and you're not rushing. if you're willing to wait a little longer on it, supposedly it's pretty spot on. And from everything I've built with it, it seems really great. Where Claude will move a little faster. It might, I'm told, might break some things along the way. But that's why I always say, my recommendation is always, if you are an engineer and you're starting Claude Code, if you're not an engineer and you want to learn building software with AI, I would start with Codex. The idea being it's probably a little safer for a beginner.

2:10:30Yeah.

2:10:31Grant:Now, we also listed on this blog post, we're not going to go through these today, but a list of professional services. We'll update this on a rolling basis, but if you need AI for a very specific thing, these are all of the best services that we've found for those things. There's more on here that we haven't covered. so like I said we need to update this but these were all the ones I could think of go check out the article

2:10:58Corey:yeah I'll show the link again yeah we'll drop the link in there make sure everybody's got it

2:11:03Grant:so if like let's say you don't want to try to set up skills and projects for a specific thing you just want AI for language learning for example right like speak we talk to them we recommend this it's great you know for organizing your work Corey mentioned Notion Notion works really

2:11:20Corey:well with AI yeah it communicates really well it can be really simple and clear uh i'm a big fan of it

2:11:28Grant:um obsidian is a good one for local file management as an alternative go ahead cool cool

2:11:33Corey:i was just gonna say a couple of things we wanted to make sure and mention today number one if you are here oh never mind i don't want to get into that one let's say this uh if you're here and you run a business you'd love to advertise with the neuron we'd obviously

2:11:50Grant:love to have you please contact mindy metz on our team she's got some crazy deals going all the time

2:11:55Corey:both for the podcast for the newsletter for the website for all of the things we do she can absolutely hook you up also if you haven't yet please take a moment to subscribe to the channel like the video it really helps us a lot helps ensure youtube's pushing us out there and people are actually watching when we come do these things so uh we're really really really grateful to every one of you who showed up today uh yeah a lot it was uh the numbers got high enough that grant and i were both like oh wow uh yeah it was kind of shocking actually because we had a couple

2:12:25Grant:streams lately that have been like not as high as this yeah it was amazing like oh okay there's a

2:12:29Corey:demand for beginner content yeah and that's really good to know and uh we'll have some cool stuff to share in the nearest future on beginner stuff as well that i think a lot of you will really love but it's not ready for showtime yet. And yeah, check out the newsletter and everything else. The Neuron.ai, we'd love to have you as a reader and a member of the community. We're carefully trying to build out here. But I think that's it for today. Is that it for today, Grant?

2:12:58Grant:Yeah, I'll just kind of cover one last thing here, which is you might've heard of OpenClaw. This is not for beginners, but if you do want an AI that's sort of like the operating system for your whole life, this is a whole other can of worms to check out but definitely try to use the codex and clawed first before you get into this stuff. Yeah, that stuff there is

2:13:22Corey:another step down the ladder. It's another step down the ladder. That's more like level

2:13:27Grant:five. And then there's also tools to run your own AI on your computer. We've done many streams about this

2:13:33Corey:but you can check those out all in the article.

2:13:37Grant:I don't really know where my riverside is anymore oh okay over here let's stop streaming my screen uh did we do the whole run of

2:13:46Corey:show did we cover everything i think we covered everything i mean maybe not as as perfect in an order as we'd like and a little bit of it was a was a 15 minute speed round there but that's okay

2:13:56Grant:i think i think we got it um oh we didn't cover opal can i just show this real fast before we close out you sure can if you want yeah i can i do want to um because i mentioned it and there

2:14:07Corey:There will be other streams, though. It's important to remember that. And don't forget to reach out to us if you have things that you're like, I don't get this. I don't think other people get it. If that's the case, we want to know about it, because we're always glad to either come teach it or go learn and then teach it. And, you know, come out and share with you guys the best we can, what we find and what we learn.

2:14:30Grant:All right. So I'm just going to show this real, real fast for folks who are like me and want to close the loop. so but if you if if you're not using Gemini you can you can peace out this is not relevant for you but if you do use Gemini then this would be helpful all right so let me go in and share so one of the cool things about Gemini gems is that you can connect them to Opal and Opal is like a little agent manager tool that Gemini has created and so we actually had a whole video going into this which we'll plug with the article that goes along with this but

2:15:08Corey:I'll actually look it up YouTube yeah so I'm gonna paste this in the chat here Cool.

2:15:26Grant:So that's the video. Too many. Oh, I'm showing the wrong freaking side. That's why. Whoops. Okay. So this is Opal. There we go. So what I did is I went to opal.google.com and then I connected it with my Google account. And as you can see here, what you can do is you can describe what you want to build and it will create a workflow for you. So, for example, I could say, help me create an automation. I can schedule to review my YouTube videos every day, which, you know, is something that people could need help with. And Gemini is a good use case for that. so opal will um their agent will start putting together a workflow for me um but what you can do is you can also connect this to your gem so if you have a gem that you've created like let's say we've created the header image gem you could create a scheduled automation that says like hey every day at um 5 p.m run the header image creator on the article draft in this google folder and it would then like you can set it up in a way that would do that for you yep um okay so click to try this opal sure let's start it enter the url of your youtube channel do a shameless plug here the neuron.ai oh wait the neuron youtube that's a website yeah there we go the channel is at is it the neuron ai neuron ai why do i know it through here and i don't even really know what it's doing what i would do is i would actually like review this and see is this even what i want each of these little cards and if you want to know what all these cards mean you should watch the

2:17:18Corey:the youtube video we just shared in the chat yeah um where we actually talked to megan lee who's the

2:17:24Grant:product manager of opal and she walks us through the whole thing it's really cool yeah specify the criteria of the video for VVU. What channels are, or what's the latest video on this channel? Okay, so now it's going to, it's taking my criteria, it's taking the YouTube link, it's searching the Neuron AI YouTube channel. Let's see what it comes up with.

2:17:55Corey:Awesome. Hey, thank you to everybody who's saying thanks and heading out in the chat. Really appreciate everyone who's joined us today. It's been a lot of fun. Is there a free version of this, Grant?

2:18:06Grant:This is free.

2:18:09Corey:Except the models in it aren't, right?

2:18:12Grant:You know what? I don't know, actually, if you have to have like a workspace. Or your own APIs. Yeah. But I think there's a certain number of API calls you get per day. And what that just means is like that's how many times you can ask the AI questions. It's like when you're on a subscription or a free account, they give you a certain number of limits.

2:18:32Corey:Yeah.

2:18:33Grant:But yeah, this is through our company account. So I assume that we have higher limits than the average person. But as long as you have a Google account, I think you can use this.

2:18:44Corey:Cool. Cool.

2:18:45Grant:Yeah. It's generating a whole report for me. That's interesting. You can also look in here. That could be a while. Yeah. You can see what the agent's doing. now it's at the point where it's generating the report nice you can change the theme I guess which is cool yeah it's generating HTML that's interesting I don't know why oh it says render the review report content along with the YouTube channel and review criteria into a well formatted HTML webpage cool there's a lot you can do with this tool honestly they need to somehow bake this functionality into the regular Google app so that it works as easily.

2:19:28Grant:But yeah, this is a daily YouTube review. Channel Neuron continues to solidify its position as a premier destination for high-frequency AI news updates. Hey, hey, hey. The most recent video titled OpenA's New Model is here, and it's incredible. That is... That is slop. I don't think we've ever made a video called that. Yeah, that's slop.

2:19:50Corey:Maybe there's a short from the social team we don't know about?

2:19:53Grant:Oh, maybe?

2:19:55Corey:Maybe? Let's see. I'm going to give you the benefit of the doubt, Gemini.

2:20:00Grant:Shorts? No. No, we completely made that up. See, this is what I mean by it being... And to be fair, we didn't give it the best instructions, so let's give it the...

2:20:11Corey:This is why I'm so leery of Gemini models that are attached to products.

2:20:15Grant:Yeah, because we don't know what model this is running under the hood. I don't know if we can change the output. it um but you can prompt it away but it's just like like this is a lot of extra steps you have to do versus like codex and claude are a little bit more sophisticated in terms of the harness that's controlling it um and it just kind of does it for you um so that's why we recommend those as our daily driver but there's a ton of functionality here if you're willing to like get in here um uh and mess around with it and get it to the point where you're really happy with it by the way if

2:20:48Corey:If you're checking out cool tools that maybe we don't know about, we would love to hear about them because we're always looking for, you know, the next great thing and tinkering with new things and new models and new platforms and new agents. And if you know anything, we may not shoot us an email anytime. We'd love to hear from you. We need to get an email account that is specifically for reader communication. we don't have that but that's something we need to set up right away I think because it's as it is it comes in through a mighty busy account and it's easy to overlook them so but yeah that's it for today please like it subscribe call Mindy if you are interested in advertising on our show podcast newsletter any of the things I hope you love this

2:21:37Grant:check out the robot newsletter

2:21:38Corey:there's a robot newsletter coming and more news you still want to stay tuned for not gonna go there yet but trust me it's cool and on that note thanks we really appreciate everybody who hung in there i can't believe this many people hung in here from you know two and a half hours uh respect mad respect and really really appreciate it it's been a lot of fun and uh we will do this again soon uh we will not have a thursday night next week by the way we might have a random evening live there's a good chance i'll be live on monday morning though for a new show we're starting called AI News and whatever.

2:22:15Corey:AI News Rumors, and we're going to change this last half of it to be goofy every time, so we'd love to have you.

2:22:21Grant:I'm going to plug one more link as well, Corey. If you have questions that we didn't answer, fill it out here. Oh, good call. Did you make a form? This form is the one that we have. Let me just remove the parameters from it. Okay, let me make sure this works. Yeah, so this is our skill of the day request forum. So this is where we take user requests for our AI skill of the day section for the newsletter. So if you have a specific thing you want to learn, like fill it up, fill that out. And like I said at the beginning, if we don't cover it, you know, tomorrow, we'll try to make sure that we cover those all by the end of the month and put them in a digest for you.

2:23:04Corey:Yeah, there's always a day we're slow and need something quick. So if you send something, you bet it's not going to go to waste. yeah exactly thank you everyone really appreciate it and we'll see you back here next time

2:23:17Grant:farewell humans

2:23:21Corey:oops now i gotta look around to go hit the stop button