Subscribe
Share
Share
Embed
Most people use ChatGPT like a fancy Google search. In this episode, we show you what you might be missing.
Allen shares how ChatGPT helped him navigate a real medical situation and saved critical time in the process. Then we break down the tools most people scroll right past: Custom GPTs, Google Gems, Claude Projects/Artifacts, and more. We explore together and show you how to go from chatting with AI to actually building with it.
Key Topics
Allen shares how ChatGPT helped him navigate a real medical situation and saved critical time in the process. Then we break down the tools most people scroll right past: Custom GPTs, Google Gems, Claude Projects/Artifacts, and more. We explore together and show you how to go from chatting with AI to actually building with it.
Key Topics
- What are Custom GPTs and how to create your own
- Use case: Managing your HOA with a PDF knowledge base
- Connecting external data through actions and APIs
- Comparing tools across Gemini, Grok, Claude, and Google Gems
- Building simple AI apps: Recipes, games, data analysis
- How projects, datasets, and data organization actually work
Connect with us online here:
Donn
Allen
Creators and Guests
What is Autonomous?
Two founders, one a seasoned AI engineer, one a non-traditional builder who learned through AI, have unfiltered conversations about building products, using AI as a tool, and pushing limits in work and life. Part technical, part philosophical, part two friends being real.
The show is for ambitious people who want to do more than they think they’re capable of, whether that’s shipping their first product, going deeper technically, or just thinking bigger.
Alright Alan, welcome back man.
How you been dude?
Hey, what's going on?
How was LA?
It was good for folks who don't know I was in LA for some business stuff and had to go
over to Hollywood and just deal with all the craziness that's over in California.
How is the East Coast?
I heard you guys got hammered with some snow again.
We are finally out of the deep freeze on the East Coast.
It was a pretty miserable week and just running my heat like 24 seven couldn't breathe
through my nose at some points.
But yeah, it's it's been it's been a good week.
really diving deeper into the open claw.
I don't know what to call it anymore.
Claude bot, mult bot, open claw.
It is now a worldwide sensation, you know, in just the week that we haven't
recorded.
mean, look at what's happening.
going on with OpenClaw that we'll probably have to have a couple more episodes on it
because there's, I've sent these to you, a bunch of Twitter threads or long posts of
here's how you optimize it, here's how you save money using different models.
It's over the top, it's crazy.
Yeah, it really is crazy.
I mean, got some new models to play with too over the last week and a half.
since we recorded, got Opus four six, we got sonnet four six, just came out like two,
three days ago.
I mean, it was, it was five two codecs like three weeks ago.
And then they just like surprise dropped five three, right?
So it came out real quick.
Yeah.
going back and forth.
And I think that's one of the challenges a lot of people have is like, which model do I
use and everything?
But that kind of gets in little more of the advanced use when you start to hop around to
your different models.
But one of the reasons why we wanted to have today's show is because we wanted to help
people get started in just working with all the different types of AI models.
One of the questions I get a lot is like, hey, Don, how do I start with AI?
Like I use chat GPT and I might ask it questions and have it do some research for me.
kind of go back and forth with it,
what's kind of the next level.
You're working with agents and everything.
And so I think there's like this middle ground between, especially for people who are
non-technical, who want to go from using ChatGPT as a glorified search engine to actually
seeing it used for some real stuff.
So today we're going to talk about the custom GPTs projects inside of ChatGPT.
We'll talk about GeminiGems.
We're going to talk about some similar stuff in Clot and Grok and everything.
So that's what we'll be talking about.
I have used some of these custom GPTs
and so forth and Alan I know that you actually haven't worked that much with chat GPT and
you've actually done more work with Gemini is that correct?
Yeah.
mean, Gemini has been my daily driver for pretty much everything.
I'm mixing it up a bit now with some other models.
I'm using Grok.
I'm using Claude, but I haven't used too much of the chat GPT models.
Although I have been hearing from, you know, inside the developer circles that five three
Codex is an amazing model to code with.
Mm-hmm.
one of the ways that I had been using chat GPT though, is just to ask quick questions.
It's a super fast model.
Obviously it's one of the most intelligent models in the world.
and I have a, you know, I have a story, I guess I could, I could tell, happened the last
few weeks.
I have two kids, you know, they're constantly getting sick or, you know, something,
something bothering you.
What'd say?
They're pretty young, right?
Yeah, they're pretty young and, um, know, I'll listen, I'm I'm a first time data, even
though I have two kids are pretty young and I'm still learning like how to handle certain
situations.
So I will, you know, previous to this, you would just go ahead and Google something.
And, and now I'll go ahead and I'll ask chat GPT or I'll ask Gemini.
Um, and for whatever reason, uh, I've been using chat GPT to ask quick questions, to get
answers.
I just think I like the organization of the output.
It's quick.
It's been very accurate for those those quick type questions.
Anyway, about a week and a half ago, two weeks ago, my daughter is complaining of stomach
pain and she's telling me, daddy, my stomach hurts.
It's coming in waves.
She's going like 10, 15 minutes with like pain and then another 20 minutes or so.
No pain and starting to get worse throughout the day.
And I'm like, what is this?
That's kind of weird, you know?
And so I just
Asked chat GPT and I said, Hey, chat GPT, um, have a three year old daughter.
she's complaining about this, this and this.
And I'll tell you what, man.
it, it gave me a response that might've saved her from a much more serious outcome than
what had actually happened.
And it said to me, it said, Alan, you know, not, not to, cause any fear.
you know,
please stay calm, basically.
But this is not something that could wait.
You're going to have to take your daughter to the hospital or try to get in touch with our
primary care doctor.
I'm like, OK, wow, it's like 1030 at night.
I don't know what to do.
I'm reading through what it's saying, and it's giving me all the reasons that I should
follow up immediately and the causes of what it could be.
And so I took her to the ER.
I was like, you know what?
this seems a little bit more serious than a regular tummy ache and um chat GPT is telling
me this, this and this and whatever.
go to the hospital, we'll get it checked out, all at peace of mind.
And at the hospital, I did some scans and they realized it wasn't gas pain.
I was kind of hoping they would just laugh us out at a hospital and say like, hey, your
three year old has just stomach pain.
um Go home.
Here's a bill.
It wasn't that man ended up being something very serious.
It's a condition called in a deception where the bowel actually backs up on itself.
And the longer your bowel stays backed up on itself, the more likely that that section of
bowel could actually become the crotic.
And in that case, what they do is they actually have to resect the bowel because it's at
high risk for being prefrontal rated and uh go septic and there's a lot of
You know, obviously negative outcomes of something like that.
What I want to say pretty clearly to people watching this is that like, this is not
medical advice, right?
You know that.
and, and, and at the act of telling me to go to the hospital was just saying, go seek
medical help from your provider.
So, so none of this is, is, is medical advice.
Our particular instance of me using this, setting it up with a prompt.
have a three year old here are the conditions.
What can I do resulted in me getting the output of this is why you should go to the
hospital.
here are the possible, things that she could be dealing with that I don't know about.
And I'll tell you it's, she ended up having to have surgery.
was, way more serious than I had imagined at that point in time.
and I am so thankful that I use that chat and that I, got that, got those results.
And it urged me to go so quickly because that time saved her from a much more complicated
and much more dangerous surgery.
Question for you though, when you had this chat with ChatGBT and it told you to go to the
hospital, did it give you a number of conditions that could be the situation and were any
of those conditions the ones that actually was?
Well, I can go ahead and I can share the screenshot or we'll post it here on the video
side of the podcast.
it like nailed the diagnosis.
Again, this is why I'm not trying to, I wasn't going to mention this because it's not
medical advice.
But it gave me a list of four conditions.
And the top condition that it said it could be likely in her age group of being like 18 to
I think it was like 48 months.
I'm not sure.
And it said in this deception in bold that it could be that and I had never even heard
that word.
Yeah.
When I was reading through it, I just kind of glance past that word.
I was like, I don't even know what that is.
Like, no way.
And yeah, after a couple hours in hospital, some imaging, lo and behold.
chat GPT kind of one shot to diagnosis.
you're It's not medical advice.
What these tools are, they just give you more information to help make a better informed
decision.
And I think that's really what your case was.
Like, you didn't think it was something super serious.
And I have two kids.
I've been in this situation countless times where a younger child is like, my tummy hurts.
It's like, OK, are you nervous?
Did you eat something that's not sitting well with you?
Maybe you're sick.
OK, let's wait it out.
Like, you probably have gas.
But.
there was very particular symptoms that you gave the model all the proper context,
everything that it needed to help give you that diagnosis and it ended up being correct.
And there's a chance that it could have not been correct and that's fine too, but at least
it urged you to make a decision that was beneficial for your daughter's health long-term.
sure and quickly it saved us a lot of time.
I think if we had gone to bed that night and waited to the morning and maybe took her to
our primary care doctor, I don't really want to think about what could have happened.
But um, yeah, you know, I'm fortunate also to have friends that are in the medical field.
You know, if you have friends who are doctors, I'm sure if I would have called one of them
or if you call one of your doctor friends, they probably would urge you to do the same
thing.
Not everyone is as fortunate to have those connections to somebody to call.
And for those reasons, I would use this as a first line of defense to ask those questions,
but make sure you're asking a very clear question and giving it the context so it can give
you the best answer.
That's what we talked about last week or two weeks ago now on the prompt episode.
the other thing about that experience with chat GPT is that after I had realized that it
was right and gave me the correct diagnosis and pushed me down the correct path, it had
built such trust with me that I used it throughout all the interactions I had with the
doctors.
There are a lot of points in the hospital where I wasn't really comfortable with their
knowledge about what was going on.
wasn't talking to doctors who were like familiar with the condition.
Uh, they were telling me, we want to do this non-surgical procedure.
it has a 60 to 70 % chance.
There is another type of the same procedure that has a higher percentage chance, but we
don't do it in this hospital.
And.
I was like, man, you know, like, I just want to make sure like, this is, this is my
daughter.
We're talking about here, you know, we're talking about like me and whatever.
Like this is my kid.
Like she needs to get the best possible medical care ever.
Um, and so I was asking chat GPT the entire time, like, Hey, this is what's going on in
the hospital.
This is what the doctor's telling me.
Like I'm uncomfortable about this.
Like, what should I say?
What are the questions I should be asking?
And
in preparing for them to come back to the room, I already have all my questions lined up
that I want to ask them.
dude, maybe another really important thing that came out of this whole transaction with
that chat was that I realized we weren't in the right hospital and that we needed to get
her to a hospital that was better suited to the condition that she had.
That's exactly what happened.
We pushed for a transfer to a different hospital.
We didn't want to continue down the same path of the same procedure that she had already
that was not working.
And we ended up in a place that was highly specialized for pediatrics.
We got one of the greatest doctors, amazing guy, was able to do her surgery.
They said it was gonna take like two hours took like 30 like 30 something minutes.
You know, he was he was just a incredible dude.
But again, like the experience of just just talking with that chat, arming myself with the
knowledge that, you know, I don't know how to ask these questions.
I don't know how to talk to doctors.
You know, I just would have been like, yeah, OK, all right, no problem.
Like, I trust you.
Like you're the doctor.
And it's I'm not pretending to be doctor.
You know what mean?
That's the thing.
Like I'm not talking to them and pretending that like I have more knowledge than them.
I'm just arming myself with the right questions to ask.
And yeah, I'm just so thankful that I was able to do that.
Yeah, that's the key is just getting yourself the additional information.
Like I don't come to be a doctor either, but there's times when I'll be discussing
something with a doctor and just come with the right level of questions to help expose or
perhaps even kind of tease out the information that I need, or maybe even it might help
them think differently and maybe expose a potential solution that we weren't thinking
about before.
Yeah, absolutely.
All right.
I'm glad that your daughter's feeling better and glad everything worked out.
It's actually quite an amazing story.
So thank you for sharing it.
Of course.
Yeah.
Happy to share it.
I hope it helps somebody.
if anybody ever from OpenAI sees this and here's my story, thank you so much to you and
your team.
I know they're doing amazing things on the life sciences and research side.
And I attribute all the knowledge and the model to that team who's working on that.
And I'll definitely continue to use it for similar questions.
And it's a great model.
yeah, thank you to the OpenAI team for that.
Yeah, I think that's a good segue into what we're talking about today.
And it kind of seeds it perfectly because it's all about context here.
And when people are moving from the basic, hey, I want to understand something a little
bit better and they're just chatting with chat GPT and then they kind of want to see what
the next level is.
And we start, they start hearing about agents and so forth.
my usual recommendation for them is, you should create a custom GPT.
And so that's kind of what this episode is about, is a custom GPTs, projects, gems, et
cetera.
And let's just hop right into that topic and I'll just kind of break it down.
Like what is a custom GPT?
A custom GPT is inside of chat GPT.
There's on the left hand side there, you can actually click explore GPTs and inside there,
there'll be a bunch of different ones.
Like there could be like a GPT to help you create comic book characters.
There's a GPT that help you write a story.
There's GPTs to help you with your homework.
There's all different kinds of ones, but you can create custom GPTs and people think like,
what would I create a custom GPT for?
Anytime you have any data that you want to particularly chat about perhaps on a regular
basis or you'd like to inquire about, that is a perfect opportunity for a custom GPT.
So what I'm actually going to do right here is I'm actually going to share my screen and
showcase an example of a GPT that I have.
Okay, so this is Chat GPT.
This is a GPT I've already created.
That is done by, on the left-hand side, you're gonna click Explore GPT.
I won't show that here.
And then on the top right, there'll be a Create.
And what I've done is I've created an HOA Assistant.
And so I'm gonna show you how I use it.
What this is, is it is a custom GPT that is customized to my current HOA, which is a
Homeowners Association.
There's a lot of rules that HOAs have to follow by.
Like sometimes, do you have to have your trash can in by a certain
in time or you can't build a fence or there's all these different rules, uh good or bad,
depends on how you look at it.
But sometimes you need to understand what they are.
And there's so many of these documents that you get from these HOAs, like the community
rules, the guidelines, the bylaws, you name it.
There's like all these PDFs and they get updated.
And if you're trying to cross reference what happened or even the historical progression
of all of this data, it's really difficult.
So what I did is I created a custom GPT and I dragged all those PDFs in and I'll show you
how to do that in a minute.
But first we'll say this is my HOA system.
I gave it little icon here of some cartoon person in front of a house and then I can ask
it questions and what it does is it will query all of the documents that I've uploaded to
it and provide me with the answer.
So now I can actually say what are the rules for parking in the community?
And so I could have typed this myself.
This is just a pre-canned example.
Now this is actually gonna go search all those PDFs and then return.
some data.
And so here we go.
I live in the parks in Carolina forest and it says here's all of our driveway parking
rules.
If you can have recreation of vehicles and boats, which you can have boats on the
properties here.
And then can ask it even further questions and continue to refine it.
Can I park on the street at night?
And this will give me more information here as well.
So you can't park on the street at night.
You have to have be parking inside of your regular parking spots here in my homeowners
association.
Now, if I wanted to say I got some new bylaws or I got some new PDFs that came from a
recent uh HOA meeting, I could edit my GPT by just clicking the GPT name and hit edit.
And this is the GPT editor.
You can see where I've uploaded all of my files inside of here.
And then I gave it instructions and it was basically a prompt.
That's all this is.
And I've provided all these information inside of here about what this GPT is for, a
description, a couple of conversation starters so it understands what it is.
And then I uploaded the files.
I allowed it to have access to web search and do image generation.
for example, there are certain images in my GPT that have like pictures of ponds or
certain things that maybe I wanted to have do a rendering because maybe I've noticed
something was broken.
And then I can say, hey, pond B has the
fountain is broken and then I could actually have it create an image or whatever off of
that.
If it's correct, that's a different story.
These things are always improving.
And this is where you can add new files and so forth.
Now, I will go back a little bit here and let me show you how we can create a custom GPT.
I'm gonna um cut this piece out, so right here.
All right, open this up.
More GPTs, close that, okay.
Okay, so now if we want to create a custom GPT, I have now went to the Explore GPTs on the
left-hand side.
And this is what you'll see.
This is where you can look at the other ones here.
There's all different kinds of them.
You can search around for different ones, humanizers, you know.
You can spend all kinds of time in here, a lot of really cool stuff in different
categories and they'll do image creation, programming, education, whatever.
But let's say you wanted to create that thing like I had, you want that H away GPT, or
maybe you've got some financial documents from your company that you want to really talk
to that data.
You'll create, create.
And inside of here, you can create the name.
This is the similar thing that you just saw that I had that was filled out before.
Give it name, whatever you want to call it.
It could be, and then give the description.
What does the GPT do?
Fill out as much context and information as you have here.
You can put some conversation starters in there, et cetera.
And then you can upload the files.
I believe the file limit is limited to 10 files.
The last time I checked, now this may have changed and potentially will change in the
future.
So if you're watching this months, years down the road, that might've changed.
but at the time of this recording, I believe it's still 10 files.
If you need more than 10 files, here's one tip I have for you.
Let's say you've got 15 PDFs.
Take a couple of those PDFs and then combine them together into one PDF and then upload
them.
I believe there's also a file limit size, but that will kind of allow you to kind of hack
your way around a little bit of it as long as the files aren't too big.
And then once you have set that up, you can actually start interviewing and playing with
your GPT by just kind of asking it questions to see if it gives you what you want.
and then you can kind of refine over here.
Click this icon will allow you to configure the actual icon that shows up for it and so
forth.
There's also kind of a creation tool here where you can actually just chat with chat GPT,
custom GPTs to have it automatically create a GPT for you.
So if this configure window is a little too intimidating, then you can come over here and
actually kind of walk through it.
So, I want to create this GPT that does X, Y and Z and I can click this little file here
and upload files or photos and then it will
help me create it and then I can refine it in this window here.
um So that's how you can create a custom GPT.
Once you click create, you will then be, and you have it, you'll be able to access it from
your custom GPTs on the side.
Again, you're go to explore GPTs and I'll show you where that's at again.
Let's go back here.
and can see my GPTs right here.
So you'll click my GPTs and then you'll be able to access all of the GPTs you have.
I have a number of them, so I'm not gonna show you all of them, but that's where you will
find them.
And then the last thing I wanna show you is if this is more of an advanced use case, if
you have external data that you would like to expose to your GPT, this is super useful.
You can actually create a new action and you can provide authentication details such as
API keys or log in with OAuth and maybe you know your API key.
does is it'll communicate with an external service.
And there's some examples here, like maybe you want to use some weather service.
This tells the custom GPT, this is what the data looks like.
And this is what the data is going to look like when it comes back.
And then it says the available actions and so forth.
And then you can start interacting with it over here.
And if you have a system, maybe your developers have created, exposed some data endpoint
that allows something to query it.
You can add these actions as long as it's accessible publicly.
And then your custom GPT
can actually start chatting with some external data as well.
how this happens behind the scenes is we won't get into the details of it.
It's probably an MCP server.
It's probably an API.
That's besides the point.
But if you do have external data sources that you would like to have access to and you can
authenticate it with them, take a look at the actions, go check out the documentation and
you can really dive deep there.
So hope that helps people understand some GPTs.
um
Alan, have you worked with the custom GPTs
Or has this kind of been your first dive into it, really?
I haven't worked too much with GPTs.
I've worked more with Google Gems, Claude the projects, which are very similar, and Grok
projects.
I really like the ability that you have to connect with an API to query external sources.
That's very cool.
to be honest, I'm wondering if I might just have been missing that feature on some of the
other platforms.
And as you're talking, I'm kind of digging around to see if it's available to me.
I will definitely do some more research after the show.
But I had been using that functionality to upload documents.
And I know on the Gemini and Grok side, I believe it's limited to about the same at 10
documents.
And I do the same thing, by the way, with kind of hacking around it to make a longer PDF
so I can query more data.
The web search is super helpful as well to get external information.
I've been using mostly Google gems to do exactly what you'd been doing.
I shared a couple prompts for some of my Google gem instructions uh on previous shows in
the show description.
there's a lot of other things going on inside Google as well.
I mean, I could show you something like a neat little trick that
been looking into over the last couple of weeks to actually create apps inside the Google
Gemini interface.
And I'll go ahead and share my screen.
So give me a second.
So before you show the Gemini gem stuff, I just wanna kind of bridge the gap for people
that aren't familiar.
most people are familiar with like chat GPT.
It's kind of like the Google of AI, right?
People are just familiar with it.
But there's Google gems, there's something for Grok I haven't even used.
I don't even know what it's called.
I guess what I'm trying to ask is chat GPT has these custom GPTs.
Is there a similar feature in Gemini?
Is that the gem and same in Grok?
Is that what that is?
Inside of Gemini, exactly what you just described is called gems.
And, to be honest with you, they don't do a great job of making the distinction between
these gem apps and like the gem projects, so to speak.
Uh,
You could see on my screen,
like here are my gems and I'll click into one and they turn
into basically conversations.
You know, I can just talk to it and there's some tools here.
You know, you can, do a bit more.
can create images, you know, depending on what it is that you want guided learning.
and you could change the model and you could talk to it with a microphone, but most of
your interaction with it is going to be here inside of this chat window.
Then you have your other gems.
And these are different, right?
So they're actually like apps inside of the web interface.
And there's a whole bunch of templates here.
And then the one that I was going to show was this one that I made.
it was based on a template of one that they had provided.
And I kind of just dabbled and like changed a couple things.
This is pretty much stock ah
this for?
This allows you to take photos of your leftovers in your fridge and then what does it do?
Generates recipes with it?
you have two user input steps, right?
You're going to ask the user to put in a photo.
Then you're going to say, all right, I'm vegan.
Or I only eat meat on Wednesdays or whatever it's going to be.
And so you insert your prompt, and then it's going to generate recipes based on the images
that you just took.
of your fridge.
It's pretty cool.
And so this is the flow and then it feeds it into generate recipes.
There's a prompt here.
It's saying analyze the leftover photo, which is the picture we sent.
And then my preferences for food and the recipe preferences right here.
And then it's telling it, all right, here's the output structure for exactly what I'm
looking at.
And then it's going to feed into generating a dish image.
So it's basically like creating a recipe book based on all of the
information that we fed in in the initial first two steps.
It's using Nano Banana.
So you can actually select the model here, which is amazing.
There's three flash, 2.5 flash, and you got Nano Banana, NanoBanana Pro, which is the
advanced image generation app.
Audio LM.
So I guess if you, I don't know, maybe you want to have like a voice that kind of actually
tells you.
what it is instead of just showing you.
That's kind of cool.
I'm sorry, I'm just playing around here.
like in every one of these steps, if you click on one of the other steps, like the green
one, you have, I guess, each the blue ones you have there, each one of those can have
their own different models.
So I might decide the first step needs to use banana banana and the next step might need
to use Gemini 2.5 or whatever, right?
Or 3.5 or whatever the model is.
so the reason why you would change your model here,
from like, let's just say using a Gemini 2.5 flash versus a three pros, you have your low
reasoning models, which are faster, um that will produce better responses, but they don't
think as deeply about them.
And there's pros and cons because not every step in a process requires this like deep
thinking and reasoning.
And then you have your deeper reasoning models
But yeah, so your low reasoning models are going to be uh lower usage, higher out, like
faster output.
And then your deeper, your deeper reasoning models are going to be higher So, you know,
when you're, when you're structuring this and it's kind of cool that they gave you this
option to do it, you can say, this step I don't really need it to think too hard.
This one, really want the best images.
You know, I'm to use this in a cookbook or I'm going to display it on my website or
wherever it's going to be.
maybe instead of using nano banana, I'll use nano banana pro, gives you better visuals,
but it uses more tokens.
So you could, have that flexibility along each step of the way to, to define that out.
and it's pretty cool.
So like I can, I, can walk through.
exactly how this works, right?
This is a simple template from the Gemini site that I copied over and made it my own.
is just running on Gemini, right?
Like you're not hosting this anyway, right?
Because like I'm asking the question as an actual, I'm interested because I've never
worked with Gemini like this.
I didn't even know it was a feature till you mentioned to me this morning.
And as far as I know, this isn't in chat GPT, but I could be wrong.
Yeah.
Yeah.
I have, I have no idea.
I think you could probably do something similar with artifacts and Claude.
We'll take a look at that in a bit.
All right.
Uh, I just wanted to grab a quick picture of a fridge.
didn't want to use my own, uh, Don and I took a couple of seconds to browse the web, but
my God, it's, it's even harder to find a picture of a normal fridge online.
So, uh, okay.
So let's just say, um, you got leftover chef.
We're gonna hit start and it's looking for me to upload the picture of the food.
All right.
So here's, here's the, uh, picture of the fridge that we found.
All right.
There it is, okay.
and hit this.
All right.
So tell us your dietary needs and tastes.
uh Open to really everything.
I'm really interested to see how this works.
like lemons.
Uh I'm gonna throw it a curve ball.
There were definitely some lemons in the picture.
Okay, so I'm gonna hit submit.
All right, so it's gonna generate some recipes and image descriptions.
uh
idea this even created this was even here.
This is crazy.
This is almost like an 8n, isn't it?
Kind of.
It is.
So this section of it, especially here, setting up how it works with the triggers and the
input.
That's pretty cool.
Here we go.
This is actually really cool.
This is fun to watch to see how it's gonna come out.
I got to show you what I made with grok.
The other night, Jess texts me and she says, I need you to marinate the chicken.
And I'm looking through the fridge and I don't know what to use because we don't have like
any garlic or whatever.
So I took all the stuff that I thought I could make a marinade.
I just lined it up on the counter.
And it gave me a full recipe of how to make this marinade for this chicken.
And it was amazing.
We call it grok sauce now and uh
We'll put the recipe down in the description, but grok sauce is pretty good.
It's like a salty, sweet, tangy barbecue sauce.
And yeah, it was a great marinade.
It was pretty good.
So anyway, we got the picture from my fridge that I found online.
Again, not my fridge.
And here you go.
I mean, it just created a really nice page.
um The leftover chef.
Yeah.
Yeah.
I saw sausage in there or something.
Maybe it ham, guess, whatever that is.
Yeah, deli meat, ham, turkey.
Okay.
All right, so we can go ahead and we can download this orange and lime.
I did not use my lemon.
It listened to me.
I don't like it.
And so it probably is not included anywhere in these recipes.
But this is just a really cool thing that you can do.
I let's just say you're having
friends over at your house and you're making a special dish or you're really into cooking
or you just want to put something together nice and you don't want to run out and shop.
mean, you could make menus with this.
could hand it to somebody.
Anyway, this is an example of one very basic thing.
you could do this all kinds of stuff.
You could probably even have it create that.
I've seen the apps where you can take a picture of Legos and it'll give you an
instructions of how to build different Lego things.
I bet you could even create something like that.
Okay, here's my Legos for my kid and I have no more instructions.
So what can I build with this?
Or I'm just trying to think of other use cases.
You could do all kinds of stuff, maybe adding...
hey, upload your financial documents or your blood work and let me give you a graph of
over time of something, you Kind of crazy.
That's cool.
Yeah, so this is a Gemini Gem though, right?
This is a Gemini gem and let's see, expanded view open.
I should have did this from the start.
Okay.
This looks way better, but it gave me three recipes based on what it saw on the fridge.
Uh, and I could share it to you as well, you know, so do I want to publish it or want to
keep it private?
So if you want to use this, I can send it over to you.
Uh, again, Opal is the advanced, editor for the process.
And it's where you get a little bit.
deeper, granular control over what it is that it's outputting.
That's awesome.
So these are the gems.
Have you worked at all inside?
mean, do these things exist in Claude or in Grok or what?
Yeah, so right now we're looking at artifacts in Claude.
There's a bunch of templates that you could choose from for artifacts.
So it lists apps and websites, documents and templates, games, productivity tools,
creative projects, quizzes, surveys, or start and build something completely from scratch.
I clicked games.
just wanted to.
I just wanted to see what it is that I could build as far as games.
And I put in a pretty simple prompt.
I just said, I'd like a connect four style game for two players.
Each player takes turn adding pieces to the grid.
And it went to work.
Everyone listening who is building with Claude knows that it does a great job.
And I gave it a pretty basic problem.
Nothing advanced, but
Within two to three minutes, it spit out a working Connect4 game.
And so that's what we're looking at on my screen right now.
It's a web interface.
It's playable.
I click the position that I want to go in as player one.
Now I'm playing as player two.
I'll click here.
Here.
I'm just playing the game for everybody else just listening to me.
my god.
so as you're clicking, it's just adding those things in there.
Can you do me a favor on this?
All right.
just trying to, I'm just trying to wrap this one up, but, here we go.
Okay.
Connect for players.
wins.
that code option up at the top?
There's like a view option and a code option at the top of the window there.
I'm wondering what that looks like.
I'm just curious.
Like what is, so it's a React, it's for the programmers in listening or watching.
This is a React app that it built, which is pretty cool.
Okay, cool.
It's all the source code of how it built it.
So very interesting.
Cool.
Yeah.
So if you're somebody who's just using the LLM to just ask questions or work on projects,
that that's fine.
But there's a lot of different features.
If you click around inside of these apps, if you're looking to get started building
something, this, didn't take me very long to do.
It's kind of fun.
It doesn't have to go into it and like have a clear idea of what it is that you want to
build.
Just go and play around and, talk to the model.
I mean, that's what I do sometimes.
I'll just sit there and, you know, just riff and say like, Hey, I had this kind of weird
idea.
Like,
How do we build this?
And then if you can build it inside the app, something simple right here, it's a great way
to learn before jumping into something a little bit more advanced, like a Cursor or
something.
What do you think?
done a lot of work over in Grok, too.
What does Grok have to offer of these similar things?
it depends on the person.
And the reason why I ask is, is have a friend of mine whose mother uses Grok for
everything, for whatever reason.
I don't know why, but she just loves Grok as her main model.
Does Grok have something like this as well?
Like projects or custom GPTs or something like that?
Yeah.
So Grok has the projects, and has something else that's really cool to call tasks.
And I like using tasks because I can basically set up a prompt, to do an action, whatever
it is that I wanted to do.
If I wanted to go search stock information every single morning, if I wanted to just send
me the weather for the day or whatever it's going to be, and I can, I could schedule it as
like a daily task, weekly task, give it the time.
And then I wake up.
And it'll either email it to me or just post it in the app, um, and it's excellent.
And the reason why I like using grok to build on top of is because it has live, data from
X.
So it's constantly pulling in fresh data.
So on top of all the information that it has,
Um, it can go and it can search on X.
can use web search.
lot of these, a lot of these, LLMs have access to tools like web search, which is really
important, but the amount of data that is coming through on X in real time is insane.
it's one of the largest communication platforms in the world.
And so, uh, as somebody who really just wants up to the minute data for like applications,
it's hard.
Yeah, it's really hard to beat.
You know, I wouldn't, it's, it's a major edge to the grok model.
and so that's why I'm using it for tasks because when I wake up in the morning, I want to
know what happened overnight, up to the minute, you know what I mean?
And it's been super effective and, accurate, which I love, you know, that a lot of models
will go and search sometimes and get results that are just like outdated articles and send
me something and.
You know, it's cool.
didn't know it, but I mean, like, it could just already be old news.
don't Yeah, so you mentioned the...
the tasks like this, that is super useful.
In fact, I learned about this recently, probably about three or four weeks ago from
someone in a local community.
I was helping host like a tech event recently and we were just having a chat about AI and
she had mentioned like, I use chat GPT tasks.
So same thing that X has.
And I asked her what she does.
Like, I have it connected to my email.
And then what she would do is she would say, hey, every more, like every Monday morning, I
want you to go through all of my emails from last week.
Give me a synopsis of what has happened and then kind of give me a summary so I can be
caught up as soon as I show up to work Monday morning or I can read it during my commute.
And I was like, that's genius.
And so she'll have that.
And then I'm thinking about it.
I'm like, wait, I could you could do this with Grok too.
I just plug it in my phone into my car and I could when I'm driving to the office or
whatever,
or commuting, I could actually have it read it to me like I'm driving to work or whatever.
say, this is amazing.
So the tasks are available in chat GPT-2.
You just have to just go in and ask it.
I believe there's a setting somewhere.
You're about to Google it.
The other thing that I wanted to say is that that up-to-date information from XAI is
critical when you want to get up-to-date information from your query with an agent.
So this is one of those situations where I will
you
I use different models for different things.
My general purpose model for asking general questions, health related questions, just, you
know, psychology stuff.
That's chat.
Jpt for coding related things and engineering related things.
I am in Claude, like hardcore into Claude.
Sometimes I'll dive into grok, but for like real up to date live information, I
immediately go to grok.
I'm like, Hey, this is happening in the news.
Or can you give me a synopsis of this five things that have happened in last two hours
from, you know, whatever thing
happening I'm seeing and it'll give me accurate data.
So good call out on the live data stuff.
Yeah, man.
It's like a gear shifter, you know?
That's exactly how I'm using the models as well.
They all have their strengths and the GROK model is insane for agents, you know,
especially at the pace that, you know, news is coming in.
It's really hard to keep up with it and GROK data is, incredible.
keep
think Gemini and Google, what we're going to see is it might be like a secret sleeper in
this area because it seems like they've taken a long time to really get going and the
Gemini stuff was okay at the beginning, but now I'm seeing it come out with more and more
stuff.
Like the thing you just demonstrated with the Gems is like amazing, because now I can
chain these things together.
has, Google has access to my Gmail, has access to my calendar, has access to my Google
Drive.
these connectors are not great yet, they're getting better and I can only see them getting
better and it goes back to something I said before is that this is the worst it's ever
going to get and that statement always holds true with AI here.
So I'm gonna be, I'm really interested to see where Gemini goes but I have a feeling that
Gemini is gonna come out to be like a major player for integrated data if that's where
you're keeping all of your information.
Yeah, the Google ecosystem is vast.
the more I dig around, the more I realize just how fast it actually is, to be honest.
You can do everything inside of Google.
And the model is so intelligent.
the reason I still use Gemini every single day is because it's one of the most intelligent
models I've ever used.
If I'm having a hard time working through a problem,
Yeah.
help me work through to a solution.
And it doesn't take too long to do it.
I see a lot of people saying that they're using Gemini 3 for coding as well.
I haven't used it inside of cursor yet.
but I think I should give it a shot.
mean, for a long time, I'd just been copy pasting the code that Gemini had used over to
terminal or cursor.
it also has like a massive context window.
I think it has a million tokens.
So it's a really efficient model.
Speaking to that exact thing, that's I was gonna add in here is that I've used it in
cursor or I've maybe had the context window limitation on maybe sonnet or opus or
whatever.
And I don't wanna compact or summarize the conversation because you're gonna lose some
details in that.
So what I will do sometimes is I will switch the model, like if I'm right towards the end
of my task, I'll switch the model to like a Gemini model that has a huge context window.
And then I finished it with Gemini because it had extra context and everything there and
it did a really good job.
I do also think that now we're seeing these new larger context windows also from Anthropic
because just this week they dropped the Sonnet 4.6 and Opus 4.6, which I think both have 1
million token context window options as well too.
So brand new things coming out should be interesting.
Yeah.
Yeah.
dude, I started using the, um, Grok four one fast reasoning model on my open claw bot.
yeah?
Yeah, it's a 2 million token context window, by the way.
And it's super cheap to run.
But, you know, back to our earlier discussion about like low reasoning or a fast reasoning
Yeah, so you can really just feel the difference because I had originally built the open
call bot using four or five and then I upgraded it to four, six opus, but the API costs
are insane.
And yeah, I can't keep up with it.
It's like, want to do so much more, but I literally can't afford to do it.
And again, just for uh people who are listening we're talking about using
API keys to get access to like bulk amounts of tokens.
You can use it to access your subscription tokens as well.
But those models, every time your agent does something, it burns through tokens.
And so when you're using like a big model like four, six Opus burning through a lot of
tokens, and it's very expensive.
So you could spend $60 in a day on
Not a lot, not a lot, to be honest.
I mean, you can get a lot done, but I mean, it just, it just goes pretty quick.
So the four one, uh, low reasoning model has been efficient just for like doing quick
searches and giving tasks.
Um, and, and, and the costs have been pretty low, but I can, again, I can see the
difference in what the model is actually able to do.
Like I asked it to create some skills for itself and they were all broken.
Like nothing worked.
So I had to go into cursor and build it myself And then it worked.
But it just feels like there has to, I know.
I haven't been able to figure out the balance of like what model is correct.
I just, again, I liked having the agent get the ability through the XAI API to search X
and I need to have that for now.
So we'll see where that goes.
interesting.
So I wanted to just wrap up here with just one last topic that I get asked, and I'm sure
you've probably been asked too.
And I'm gonna just go back to chat GPT, but it's similar in other tools as well.
Chat GPT has a concept of custom GPTs and it has a concept of projects.
And kind of like what is the difference between these two?
And it's the real simple thing is a project is just a grouping of your related chats that
you put together.
So if you're having, anytime maybe you've got a product that you're working on,
you can have that inside of a project.
So you can group all those chats together inside of that project and get some more context
in there.
Custom GPTs are where you're to have it work with external data or your PDFs or images or
whatever, and you want to chat with that data as well.
So if you want to group things together and keep everything organized, you can put that
inside of a project.
You can have multiple projects.
If you want to chat with data, you want to create a custom GPT.
And of course, you can have different types of GPTs.
I have one for my HOA.
I have one for my kids' school stuff.
There's multiple different use cases you can have for it.
And I believe Claude has projects and so do other ones.
if you, whatever tool you're using, just go explore those options.
If it's gems, if it's GPTs or projects and check them out.
You know, this is at scale.
This is exactly what like larger enterprise companies are doing now.
You have these massive companies like Databricks and Snowflake and you have these siloed
data sources and every all these companies right now are in the midst of transitioning all
of the data that they have and like legacy servers and Slack channels, whatever.
And they're taking all this unstructured data and they're putting it into one source where
they can query it in natural language.
You know, at scale, this is exactly what you're able to do.
Right.
So you're taking your HOA documents or you're taking your medical records and you're
building a chat.
That's personalized with your data and you're able to talk to it.
It's an incredible thing to be able to do, especially, you know, we, we don't read through
everything the same way that, that a computer is going to read through it and, and, and,
and detail.
And the pattern recognition is, is a major advantage across, you know, if you're, if
you're looking at a history of two or three
years of documents, you'd be surprised if you were able to mine out of that, you know,
just by asking simple questions.
Yeah, I think I told you this, maybe it was you, I don't know.
But I had taken my, I get my blood run every blood work run every three or four months
just for my own purpose.
I like to see the trends and see how I'm doing health and health.
And what I did is I was playing with Notebook LM, which is a different tool, which
basically allows you to interact with your data, create little mini podcasts, which I
wrote a blog post about and I'll put it in the show notes so you can see how I use it.
But in short, I was able to upload my PDF of my recent blood work
to Notebook LM, which is basically, it's from Google, it's basically like a chat.
And then what I told it to do is, hey, create a little mini podcast on my blood work.
I didn't give it any information whatsoever about my doctor, anything that I do that I'm
in fitness a lot.
I give it zero context.
I just gave it this and said, hey, give me a detailed analysis of my blood work.
And it generated a 15 minute podcast.
And I will tell you what, it is the best,
description and reading of my blood work I have ever received in my life better than any
doctor I've received out there.
And again, this is not medical advice.
I'm familiar with the blood work because I've studied it.
I know what all the markers mean, but to have it explained to me in this manner was so eye
opening because for example, on my blood work, there was an elevated thing on there that
was concerning.
It's like, this is concerning.
And what it was able to determine is like, oh, they also had a additional test that really
tests this,
It's a kidney level test to check their kidney levels.
And the original calculation was like, this is alarming, but the specialty test that I
also always get.
Reported back saying hey actually this person has really healthy kidneys and what they
were able to decipher from this Along with the doctors that I used but the name of the
doctor like hey most likely this person is very physically fit because this is a sports
doctor who ordered this test and it's like so most likely they probably work out a lot
they have high muscle mass so therefore they have elevated creatinine levels which means
that why that's why the first one would be off all that to say The tools that are out
there that allow you to interact with this stuff
completely change the game from your story about your daughter's condition to helping have
a positive outcome to your HOA, to blood work, to whatever you're working with.
Explore the tools, play with them, and just see the information that you can get out.
But you'll find you have amazing results, I think, in any one of these tools that you try.
Hey, Don, did you know that in Notebook LM during that podcast, you could jump in as a
co-host and actually talk to the two people that are going back and forth?
I had no idea you could do this.
Yeah, is.
Notebook LM is probably the most overlooked in the mainstream tool that that exists right
now that you can go and use.
And if you're a student, especially like if I had this in college.
Oh, my God, it's it's just such an incredible resource.
But to Don's point, if you're someone who likes listening to podcasts, I hope you are.
Come back.
a throw us a thumbs up.
then you're going to really like this tool because it'll take whatever documents, whatever
resource data resource that you want, throw it into the engine and it'll spit out a
podcast and you can jump in and talk with the host and ask them questions.
They'll bring you in and be like, Hey, we got somebody coming into the call today.
You know, it's, it's, it's actually, yeah, it is nuts.
It's really nuts.
Um, so go ahead and check out notebook Ella and we can drop a link down in the description
for them.
Uh, Google.
Yeah.
Back to the original point about Google just having this insane ecosystem of tools like
that, just still discovering every single day.
Yeah, that's one that everyone needs to know about for sure.
okay, so to this point we've covered prompts.
Now we've covered, where to start building maybe your first apps, working with larger
projects, talking to your data.
and hopefully by this point, anyone who's listening, who hadn't been using, LLMs or was
interested in learning how to use them has a little bit of a foundation now to kind of go
and get started.
Um, and again, I mean, if you have more questions about how to actually get started, you
could just talk to the LLMs, tell it what you want to do.
It'll literally walk you through the process.
Don and I can go and create a tutorial, probably be outdated in like a week and a half at
the pace that everything's going.
so in, in reality, um, I think our job is to kind of create the awareness and, uh,
hopefully push some of you to go and play with these tools and learn and push yourselves.
Um, it's kind of a, exactly what we set out to do with the podcast.
if you have any questions specifically about something we spoke about in the show, please
drop a comment down below.
or, shoot us a message on X, uh, or where else are you done?
You're on X you're on Instagram.
LinkedIn, any social media we have it all linked inside of the show notes inside of the
podcast.
So wherever you want to find us or wherever you spend your time, us up there.
Perfect, yeah, thanks for joining up today.
We're gonna get into some much deeper topics, AI agents, how to build them, open-claw,
maybe a live build session with that.
And we just wanna kind of set up the foundation that we can bring everybody along.
let's keep building.
Thanks again, everybody.
See you soon.