Learn more about Advance Course (Master the Art of End-to-End AI Automation): https://multiplai.ai/advance-course/
Learn more about AI Business Transformation Course: https://multiplai.ai/ai-course/
Most business leaders are dabbling with ChatGPT and Claude, but very few are leveraging the right tools in the right way. The confusion between Custom GPTs and Projects could be silently costing you time, money, and momentum.
In this solo deep-dive, host Isar Meitis breaks down the exact differences between Custom GPTs and Projects (in both ChatGPT and Claude), and shares real-life automations that have saved him and his clients hours per week — with no coding required.
From sales automation to social media hooks to instant proposal writing, you’ll learn what works, what doesn’t, and how to implement powerful AI tools that actually do the job for you.
In this session, you'll discover:
- How to choose between Custom GPTs and Projects for your business
- Real examples of AI automations that cut hours off daily operations
- A step-by-step walkthrough of building and deploying GPT automations
- The key differences between ChatGPT Projects and Claude Projects
- How to organize persistent memory, files, and workflows for smarter AI collaboration
- When you shouldn’t use Custom GPTs (and what to do instead)
- Why the newest update to Projects just made them a game-changer
About Leveraging AI
- The Ultimate AI Course for Business People: https://multiplai.ai/ai-course/
- YouTube Full Episodes: https://www.youtube.com/@Multiplai_AI/
- Connect with Isar Meitis: https://www.linkedin.com/in/isarmeitis/
- Join our Live Sessions, AI Hangouts and newsletter: https://services.multiplai.ai/events
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
GMT20251124-180014_Record (00:00):
Hello and welcome to the Leveraging AI
(00:02):
Podcast, the podcast that sharespractical, ethical ways to
leverage AI to improveefficiency, grow your business,
and advance your career.
This Isar Metis, your host, andwe have a really interesting
episode and.
I have a really interestingepisode for you today.
There is a relatively bigconfusion between different
tools that chat GPT provides toall of us basically right now.
(00:25):
And what I mean specifically isthe difference between custom
GPTs and projects.
Now, if you use one.
Or the other, or maybe a littlebit of both, or definitely if
you haven't used any of them,you'll find significant value in
this episode and you will learnhow to actually implement them
effectively.
You'll learn how to pick theright one for your project and
even how to.
(00:46):
Create specific automations thatyou can use in your business and
areas of expertise that you canuse in your business in order to
make the most out of chat GPT.
We're also going to touch alittle bit about the differences
between that and how to useprojects in Claude and some
minor differences between thisand that.
So if you want to understand howto use ChatGPT projects or
(01:07):
custom GPT or cloud projects inthe most effective way, how to
develop them correctly and whatthe hell are the differences
between them?
this episode is for you.
Before we dive into explainingall the differences and so on, I
want you to understand what arethe options, like what the hell
can you do with these tools?
And so I'm going to show youdifferent examples of
(01:27):
automations as well as differentother ways to use projects.
And then you at least have anunderstanding of what are they
good for, and then we can talkabout the differences and how to
pick the right ones and how toimplement them.
Let's start with an example ofhow you can use, in this
particular case, a custom GPT todo data analysis across several
(01:47):
different files.
What I'm going to show you rightnow is a really interesting
automation process that actuallytakes two separate files.
This process was built on a realprocess that I've done for a
client.
But using fake data and whatthis process does is actually
extremely helpful.
So they sell across multipledifferent channels.
They get an updated levels ofinventory from all these
(02:09):
different channels and everysingle day they have to combine
all that information and thencompare it to the sales data and
then based on the new levels ofinventory, they need to create a
new updated sales brief thatgoes to their salespeople.
So the salespeople know what tosell more, what to sell less,
what is inventory they ran outof, and they need to stop
selling whatsoever what theyhave a lot of, and they need to
(02:31):
run promotions and so on.
This process used to take themseveral different hours every
single day.
And then we've built onautomation routed using a simple
custom GPT.
So the inputs to this custom GPTare just two files.
One is yesterday's salesbriefing, and the other is
today's updated inventoryinformation after it was
(02:52):
compiled from all the differentsources.
And what you'll be able to seeis that this GPT now will run
four or five different steps,all its own, each and every one
of them.
Highly detailed by comparingcolumns and rows and
cross-referencing data from bothfiles.
And it is just going to read thefiles and then start doing.
So say, did step one and nowit's going into step two, and
(03:12):
it's gonna analyze step two.
And you can see in each stepexactly what it's going to be
doing.
So it's gonna pull the materialfrom one file compared to styles
in the other file.
And then considering all theroles matching to description,
it's doing a lot of highlydetailed work.
And now it's going to stepthree, each and every one of
these steps used to take humans.
Multiple minutes and sometimesthe combined effort would've
(03:34):
taken, as I mentioned, two tothree hours.
And now here we are and we'redone.
So you see now I have adownload, updated inventory, uh,
file that I can click well checkit first and then send it to my
sales team immediately.
So this took about.
Three minutes of me talking,meaning I could have done an
email or something if this wasmy actual work, and I've got the
file that previously used totake hours to do by
(03:56):
cross-referencing two differentfiles in two different formats
with different namings and soon, and yet it knows how to do
this effectively every singletime.
Let's look at a completelydifferent example.
In this example, I will show youa automation process that writes
hooks for my posts.
So how does that work?
Well, initially, my team hascollected multiple hooks from
(04:17):
many different sources on theinternet.
Ones that work and drove a lotof engagements.
And now all I have to do is dropa post into it and it knows how
to do that, and it knows how tofind relevant hooks that I can
use and recommend them to me,and I can pick the one that I
think makes the most amount ofsense.
So let's take an example.
Let's go to LinkedIn.
Let's pick anybody's post.
It doesn't really matter.
(04:38):
I'm picking out a post and I'mpicking it without the hook line
from the beginning, and then I'mgonna drop it into.
In this particular case, customGPT.
Then I'm gonna click go again,no instructions, no nothing.
And then the tool itself will gothrough multiple different
options, uh, of hooks and itwill find the ones that probably
will work best for this post.
(05:00):
So you can see now it gives me,uh, three different options
option.
Hook template number 14, storycategory, how I went from X to
Y, so this is the template, andthen the way it's suggesting to
use it is how I went from 12views on my first LinkedIn post
to speaking at a HubSpot eventalongside the CEO of a$2.1
billion company.
That is not bad.
(05:21):
Definitely a hook.
I will probably click to seewhat the rest of it hook.
And then it gives you anotherhook option.
Most people think X, butactually Y.
So this is the template.
And again, the recommendation ismost people think your first
version needs to be perfect, butactually your worst work is the
data that gets you to your best.
So I think the previous one isbetter, and then it give me
another one, and I can ask foras many more as I want, and then
(05:43):
I can pick the one I want touse.
Or at least get ideas and then Ican write my own, or I sometimes
combine some of them togetherinto one hook.
So this is, again, somethingthat just helps me be more
creative and follow templatesthat have been successful for
other people.
I'll show you one more examplethat I use all the time, and
that saves me hours.
Every single week.
I get approached regularly, soprobably several different calls
(06:05):
per week to provide AI workshopsto companies.
And these always work the sameway where I have an initial.
Call with usually a seniorperson, the CEO or the COO or
somebody like that, or somebodyfrom l and d that is asking me
about my services.
And sometimes it's follow upwith another call with somebody
more senior on another personthat can weigh in on how the
(06:26):
training should go, and then afew emails go back and forth.
And then they're asking for aproposal.
So either one call or several.
And then I need to write aproposal, and this used to take
me writing the proposals abouttwo hours per proposal.
And now what I do is I actuallyhave a very simple process where
I take the transcriptions of thecalls from Fathom, which is a
tool I use that listens to allmy calls, re regard whether
(06:48):
they're on Zoom or Teams or uh,Google Meet.
And then I drop that into thiscustom GPT or project and it
writes the proposal for me.
So a process again that used totake me about two hours now it
takes the AI about five minutes,takes me another 10 minutes to
review it, correct what I needto correct, drop it into my
header and footer in a regulardocument and then send it over.
(07:10):
So a huge savings of time,especially that I'm doing two to
three of those every singleweek.
That gives you an idea of howmuch time I'm saving just with
that one automation.
So these are the things you cando with it when it comes to
automations, but actuallyprojects can be used for a lot
more than that.
So now let's take a look atprojects for a second and see
exactly how that looks like.
So where do projects live insideof ChatGPT and how exactly can
(07:35):
you use it?
Well, it lives on the left sidemenu, the top thing you all.
Every one of you has, it saysNew G.
It says New chat, search, chats,library, and so on.
And then underneath that youhave projects.
And projects can be used in twodifferent ways.
One is in similar ways to customgpt.
Again, we'll discuss exactlywhat that means, but the other
(07:55):
way is as a space to haverelevant context for a specific
task.
So what you can see here is youcan see that I'm looking right
now at a project that is calledSample Client Project.
The reason I have it is I don'tnecessarily wanna share what I'm
doing, these demos, how I'mactually working with each and
every one of my clients.
But how am I actually using itis really important and not
(08:17):
necessarily the details of aspecific client.
In general, all these AI toolsare thriving on context.
The more context they have, thebetter results you are going to
get.
Which means if you want to getgreat results when you're
talking to a specific client,you need to tell ChatGPT or
Claude or Gemini or gr.
Doesn't matter which tool youuse.
(08:38):
Everything you know about.
The client, the project, thepeople you're working with, the
specific proposal, theeverything that you want, which
just takes a lot of time andit's obviously not very
effective doing this.
Every single chat that you have,and this is exactly one of the
things projects are very goodat.
You can provide the chat.
With background information,context of the specific client
(08:59):
or on the specific project or onthe specific topic you are
working on regularly.
For me, as an example, I do AIcourses.
I have this podcast, I have aYouTube channel, so each and
every one of them has its own.
Project.
So what is in the project?
The project has a combination oftwo different things.
One is files or knowledge basein the language of custom gps.
(09:19):
So what could be these files?
Well, they could be everything.
If I just want to have a.
Project about a client that I'mworking with.
I will have something like this.
So I have information about theculture of this client as a
document.
How do I know that?
Well, first of all, I go totheir website and I see what
they write about themselves, butthen I take, uh.
Additional stuff that I knowabout them, and I add it to that
(09:40):
document and I upload that as adocument.
I have different proposals thatI've written for them, so you
can see a course proposal and aworkshop proposal.
You also have a consultationagreement that I have with them
in here, and people documentwhat's in the people document.
I will create a deep researchproject and I'll ask it to
research the top people that I'mworking in, the company and the
decision makers.
Then I will go to LinkedIn and Iwill pull some additional
(10:02):
information from there.
And then I have a segment inthis document about each
interview people that I'mworking with as well as hire
decision makers in the companyif they're not the people that
I'm talking to.
And I upload that as well.
And then about us.
So I'm literally taking theabout us page from their website
combined with deep researchabout the company, saving it as
a document and uploading that asa document about this client.
(10:24):
So.
What happens now is if I amopening a new conversation here,
different than a regularconversation in ChatGPT, this
conversation under projectsalready knows all of that
information.
And so instead of giving mevanilla answers, it knows
everything there is in here,including.
Projects, including proposals,including work we've done
(10:46):
before, including emails.
If they're important, I willinclude them in here about the
company and so on.
And hence, the answers that I'mgetting are specifically for
this company are related to thework we've done with them and so
on.
So extremely helpful.
But in addition to just thefiles, it also has instructions.
So if you go to the, theellipses, the three dots on the
top right corner in projects.
(11:08):
You can see that I have asegment.
Uh, I have a button called editinstructions.
If you click on editinstructions, you can see it has
instructions and you can seethat this one says Comprehensive
AI Strategy Development forclient.
And then I have differentobjectives and how to work with
it and what are the benefits andwhat are the pros and cons and
how I want it to respond and soon.
And so the way it is going toanswer is very, very different
(11:31):
than the way just a regular chatis going to answer because it's
custom tailored for this clientand for the kind of work I wanna
do with this client and the wayI want to communicate with GPT
about this client.
This is a complete game changercompared to using just a regular
conversation with chat GPT Now,because you can create these
(11:51):
smaller bubbles of context.
Each and every one of them canbe specific just to that topic,
which again, could be a client,could be a project, could be
something that you're working onregularly, et cetera.
So this is one way to useprojects.
But before we dive into theother way to use projects, I
wanna show you what our customGPT is, because that will help
you understand how you can useprojects as well.
(12:12):
So custom GPTs are these miniautomations that can do one
thing again and again, just likeI showed you before, either find
a hook or create proposals oranything else that is a
repetitive task in your company,whether they live.
They live down here on the leftmenu of chat, GPT, where it says
gpt, and underneath that it willsay Explore and then it will
show you a few of your gpt ifyou don't see all of them.
(12:33):
Then if you click on explore onthe top right corner of the
screen, you will see my gpt.
And when you click on that,you're gonna get a list of all
your gpt.
I have about a third of them ormaybe more so.
Let's look at the one thatgenerated the hooks for us
before.
So if I click on that, uh, if I,on the top left corner, it says
post hook generator, and it hasa little dropdown menu, and if I
(12:53):
click on that, I can click onedit and then we can see the
structure of how this works.
So in this particular setup,what you see is once we get into
a custom GPT, you can see it hasa name, which it just tells you
what it is.
Then it has a description whichtells you in more detail what it
is, and then it hasinstructions.
And in this case, theinstructions are really, really
short.
You're an expert copyright whospecializes in writing hooks for
(13:15):
LinkedIn post.
I want you to write three hooksfor any LinkedIn post that I
provide.
One, please refer to the hooksfrom the Resource 750 plus
Hooks.
Two.
Select three hooks, templatesthat work.
Three, please include the Hooktemplate and the number four,
use the templates from theresource two, craft a Suitable
Hooks.
That's it.
(13:35):
Really short, but reallypowerful.
So how does it know where topull the hooks from?
Well, down here it has what'scalled knowledge base.
And in the knowledge base youcan upload files in one of the
files.
Up.
Up I've uploaded has a documentthat has over 750 hooks with
examples on how to use them.
Now the other thing you have isyou have.
(13:56):
Conversation starters, what areconversation starters?
They add buttons on the customGPT.
These buttons can be used as oneof two things, either as to give
somebody an idea of what theyneed to do.
So in this particular case, yousee it says, input your LinkedIn
post.
So I know what to do.
Otherwise there's no way for meto know.
So if I'm gonna share this withsomebody, they may or may not
(14:16):
know what to do with this customGPD.
But the other thing that you canuse it for is you can create
several of these conversationstarters, and then in the
instructions refer to each andevery one of them separately,
which means you can allow theuser to.
Start with a different kind ofdata or a different kind of
source, or a different kind ofstarting point that based on the
(14:37):
button that they have clicked,which just makes it more
versatile.
So these are the capabilitiesinside a custom GPT.
And as you've seen.
All you have to do is uploadsomething and get the output,
because it already has theinstructions.
You don't need to tell it whatyou want it to do, and you will
do the instructions time andtime again.
If you wanna dive deeper intohow these things work and how to
develop them.
Uh, there's a separate episodethat we've done about it.
(14:59):
So if you go back to episode 175, it was called Stop Wasting
Time, automate Repetitive Taskswith Custom gpt.
You'll get a little moreinformation that we're sharing
today, even though I'm gonna tryto share a lot of it with you
today as well, even thoughtoday's episode is more about to
compare that with projects andhow to choose the right solution
for you.
Now down here on the bottom, youhave two more things.
(15:21):
Inside the setup of custom gpt,you have the capabilities that
you wanted to include, and thatincludes web search canvas.
Image generation and codeinterpreter in data analysis and
just pick the ones you actuallyneed in the custom GPT versus
everything.
And then on the bottom, on thevery bottom, you have create new
action.
What are actions?
Actions are the ability for thecustom GPT to actually run code
(15:43):
and connect to other things inyour tech stack.
So.
That means if you have othertools or databases and you wanna
be able to either query thedatabases or connect through an
API to third party tools, youcan do that inside a custom GPT,
which makes it even morepowerful because now it can
query your CRM, your ERP, youremail marketing platform and so
(16:03):
on, and work with it in tandemto give you the results that you
want versus.
You having to export data, copy,paste it into the custom GPT and
so on.
Now, if you don't know how towrite code, then there's a
little button here that says Gethelp from action GPT.
That uses another.
Custom GPT that OpenAI createdto write code for you to put it
(16:24):
in here, for it to connect withthird party tools.
Most people, and in most cases,that is not necessary, but it is
a huge benefit if it's somethingthat you need or want to do.
So these are custom gpt now on avery high level, these two
things, projects and custom gpthave very similar concepts,
meaning they have a knowledgebase of files that you can
(16:44):
upload, and they haveinstructions that they know how
to follow in order to deliverspecific kind of outcomes.
So let's dive a little deeperinto what are the differences
between projects and custom gpt.
That will give you a hint onwhich one you need to pick for
different use cases, and I willgive you an idea.
As of right now, there's a veryclear winner that you should
(17:07):
pick probably 95% of the time.
So let's start with the purpose.
Is there a difference in purposebetween custom gpt and projects?
And the answer is, it depends.
Custom gpt are built to be arepetitive task creator.
Basically, if you have a.
Single task that you wannarepeat again and again and
again, then Custom GPT will dothat.
(17:29):
Projects can also do that, butin addition, they can be used as
a way to have conversationsabout a specific topic, meaning
it's an entire workspace thatcan use similar kind of
information for two differentgoals.
One is to run an automation thathas to run exactly the same way.
Or to have open-endedconversations about the same
(17:51):
background information.
Now, to be fair, in most cases Icreate two different projects
for these two different tasks.
One of them would be to createvery specific instructions that
will repeat a task again andagain and again and the other
type of project will be a.
And the other type of projectwill be just a data source for
open-ended conversations with acontext about this particular
(18:15):
topic.
But in general, custom GPTs area one trick pony.
They will repeat the task thatyou give them, and projects can
do that plus allow you to haveopen-ended conversations.
The next topic is knowledge baseor files that you can upload
that is very similar on bothtools, so both on custom GPT and
projects.
You can upload PDFs, worddocument, excels images, et
(18:35):
cetera, and that is very similaron both sides.
The file limit is.
Also very much the same.
There's slight differencesdepending on the level of
license that you are using.
But in general, on the pluslevel, which most people are
using, you get 20 files of five,112 megabytes.
Each with 20 megabytes maximumper images, and as far as the
(18:57):
number of files that you'regetting in the projects that
goes up if you have the higherlevel plans.
But usually the 20 files is morethan enough for what you need.
Now here is where it getsinteresting.
The next thing we're going tocompare is how can you share
them with others?
so custom GPS can be shared inseveral different.
The very basic is just you.
You are not sharing it, andyou're just using the custom GPT
(19:20):
that you created.
Option two, you can share itwith anyone with the link.
Option three, you can share itif you have the enterprise level
with other people from yourcompany.
And option four, you can shareit with what's called the GPT
store, which means anybody whohas access to chat, GPT can use
your custom GPT on projects.
Until not too long ago, youcould not share it at all.
(19:42):
Which was the biggest and moreor less only disadvantage of
using project that changed backin September for people with
enterprise level licenses, andit changed for everyone in
October.
As of last month, now you canshare projects just by giving
people the link and they canwork inside the same projects
(20:02):
with you, which is a hugebenefit, and again, was the only
real serious disadvantage ofprojects before.
So that was a big advantage ofcustom gpt that is no longer an
advantage.
And now let's talk about the twobiggest advantages of projects
over custom GPT.
Biggest benefit number one ispersistent memory.
So custom GPT is every time yourun the custom GPT, it runs and
(20:26):
it's done.
It doesn't know what happenedthe previous times you run the
same custom GPT.
However, projects, as Imentioned, were originally built
not to do automations, butactually they were built in
order to be a context space, abubble of context for a specific
topic, and they have their ownmemory feature.
So just like there is a memoryfeature for ChatGPT as a whole.
(20:49):
There is a memory feature justfor the project, so every new
chat within that space adds morecontext to the overall
knowledge, experience, andcapability to be more precise
with the answers of thatproject, which does not exist in
a custom GPT, which means it's ahuge benefit and a lot more
value.
Two projects versus custom gpsbecause it learns from every
(21:12):
conversation that you or otherpeople who use the same project
are having within the project.
So this is huge benefit numberone.
Benefit number two is theorganization of the chats.
If you use a custom GPT, everytime you use it, it creates a
new regular chat.
Inside of chat GPT.
I will show you an example soyou understand what I mean.
As an example, you can see herecustomized proposal as one of
(21:34):
the regular chats that I had.
It says Your chats and it wasdone.
With the custom GPT, meaningevery time I run the custom GPT,
it actually creates a completelyregular chat as part of the
very, very long list of hundredsof thousands of chats that I
have done with chat GPT in thelast three years.
However, if I create the sameproposal inside the proposal
(21:55):
generation project.
You can see that it shows upevery single time I ran it
inside just this proposalfolder, meaning it is a lot
easier for me to findconversations that was done,
that are relevant to a specifictopic that I am looking at right
now.
So.
Both from the perspective oflearning from every single chat,
(22:16):
as well as from the perspectiveof having it a lot better
organized in one place, runningautomations, and definitely the
other option of just having openconversations because that's
something you cannot even doinside a chat.
GPT is a added value of aproject.
Now let's talk about twoadvantages of custom gpt that do
not exist on the project side.
(22:37):
Number one, it is the ability tocreate code that will connect to
third party tools such as yourC-R-M-E-R-P.
Et cetera.
And as I mentioned, that existson the bottom of the custom GPT.
It requires writing code thatyou can use AI to help you
write, but it means you need tobe a little more technical.
And to be fair, I have about 30plus custom gps that I use and.
(22:58):
One or two of them actually isusing this capability to write
code.
That capability does not existinside of projects.
So if you need that, then customGPEs are your only option.
And then the second thing isconversation starters.
We talked about this, that youcan create these buttons.
Inside of a custom GPT that willexplain what to do, or that can
start at a different startingpoint.
(23:20):
Inside of the custom GPT.
This does not exist in projectsat least yet.
So what does that mean?
It means that for the vastmajority of use cases.
Projects are right now a betteroption than custom gpt.
Meaning if you have old customgpt, there's no reason to go and
convert them because there's nopoint.
(23:40):
However, if you want to createnew automations or new
workspaces for you and your teamto work on a specific topic,
either through an automation oras a context baseline for
open-ended conversations, justcreate projects.
There's really no reason.
To create custom GPEs right now,unless you wanna connect it with
code to something else.
(24:01):
Now, I want to give you a fewmore hints on how to create
these automations, whetheryou're creating them in
projects.
Or in a custom GPTI always startwith a regular chat.
I don't actually start trying tofigure out what needs to be the
instructions.
I bring it the data.
So every time I create anautomation, there's really three
components, right?
There's the input, the process,and the output.
(24:22):
So I need to have a clearunderstanding of what the input
is.
So I bring in the input, I cleanthe data in the input.
So the data needs to be verywell organized.
If it's in Excel, then it needsto be set up correctly without
any double headers, without anyspaces, without any merged cells
and stuff like that.
So I clean the data, I bring itinto the chat, and then in the
(24:43):
regular chat, not inside tryingto build an automation, I try to
get.
To define the process in orderto get the outcome, and I just
iterate.
I ask it, oh, I need to do thisand that, and then it gives me
an opposite.
Oh, it's not what I meant.
Let's try this.
Let's try that.
Until I get to the final output,the way I want it, once I get to
the required outcome in theformat I wanted, in the length I
wanted, with the details that Iwanted, I ask it to write
(25:06):
instructions.
For a repetitive custom GPT thatwill do the entire correct
process and avoid the mistakesthat we've done in the regular
chat.
It writes amazing instructions.
Way better than I can write, andmost likely better than you can
write on your own and it willcapture all the nuance.
You can ask it to ask youquestions if there's any open
things it's not sure about, andthen it will write the
(25:27):
instructions for you.
You just copy and paste theinstructions into the
instructions section, either inthe custom GPT or in the
project.
You also tell it what kind ofreference material or knowledge
base you will give it, and askit to include that or reference
for that in its instructions.
And then add those files intothe knowledge base inside of the
(25:47):
custom GPT or the files insideof the project.
So now you have everything youneed in order to run these
automations.
And you can also use it if youcreated a project, which again
would be my recommendation.
You can also have open-endedconversations based on the
information that is in the filesand in the instructions.
And as we mentioned, if it is aproject, it will also learn over
time and get better over time.
(26:08):
So now that you know thatprojects is probably the way to
go for most of your use cases,there's a very similar concept
in Claude also called projects.
So what are projects in Claudeand how they are different from
projects inside of a custom GPT?
Well, not by much actually,projects in Claude was there
before projects in custom gptand OpenAI, more or less copied
(26:33):
the concept of projects fromClaude.
This is why OpenAI currently hastwo different features that are
very, very similar becauseOpenAI started with custom gpt.
Then Claude created projects,which was broader and more
capable, and then OpenAI copiedit and provided projects inside
of Chat GPT, which was not myrecommendation until a month
(26:53):
ago, where now you can actuallyshare them with other users,
making them a lot more powerfuland relevant to most people than
custom GPTs.
But let's talk about what arethe differences between cloud
projects and chat GPT projects?
So I already showed you howChatGPT projects look like.
Let's go to Claude and see thatit's more or less identical.
So if I'm in Claude over here,there's like a little folder
(27:16):
icon on the left, and if I clickon that, it takes me to projects
and very similar, I can create anew project.
But if you click on a project,you will see that it has a
memory, that it remembers stuffabout you, it has instructions,
and it has files In thisparticular case.
Uh, you can see that it has, 10different files and it has a set
of instructions that tell itexactly how to work and same
(27:37):
kind of thing.
This is a proposal generator andall I have to do is drop in
whatever the it is looking for,and you will know how to run
this automation.
But I can also have open-endedconversations about this.
Topic, which is AV proposalgeneration in this particular
case, and it can give me ideason how to write better proposals
and what kind of proposal am Iwriting and maybe a new
(27:58):
different kind of structure, orhow should we approach a
specific client and so on and soforth.
Because it has all theinformation about previous
proposals that I have written.
But there are still somedifferences between cloud
projects and chat GPT projectsand let's review them quickly.
Claude in general is much betteras of right now, so Claude 4.5
(28:18):
compared to ChatGPT 5.1, Claudeis much better in formatting.
The documents that it'sgenerating and the Excels that
is generating is significantlybetter looking and well
organized than what ChatGPT doesright now, which is a huge
benefit, especially if you arealso using skills.
I'm not going to dive intoskills right now.
I'm gonna do a complete separateepisodes about skills.
(28:40):
But skills are these minicapabilities that you can create
inside of Claude that knows howto do very specific things.
As an example, I have created aClaude scale that creates
everything to my brand when Iask it to do.
So, it has my color coding, thetone that I like to use,
examples of how I writedifferent things.
(29:01):
It has my logo, it has my entirebrand guidelines and it knows
how to apply it very, very wellwhen it is creating new
documents.
So the combination of that withinstructions means I have to do
less work afterwards to formatit as I need.
The disadvantage of that and theadvantage of ChatGPT is that in
both cases, when you create itin ChatGPT, it creates it in
(29:22):
Canvas if you ask it, whichmeans it's an editable work
collaboration document togetherwith the ai, meaning, you can
create manual changes to theoutput from ChatGPT versus in
Claude it looks better, but youcannot make any edits.
Meaning if you wanna make edits,you have to copy this into a
Google Doc or a Word document inorder to make these changes.
(29:43):
And in ChatGPT, you can makeedits.
On your own, combine it with thework together with ai, that I
find a lot more helpful thanusing artifacts, which is the
Claude version.
So what does that mean?
It means that right now I amusing ChatGPT more than I am
using Claude because of thatparticular reason I have.
(30:04):
Multiple times taking the outputfrom ChatGPT, dropped it into
Claude to get the formattingdone much better.
And then I have less work in themanual editing afterwards when
it comes to just making it looknice.
So final summary of all of thatcustom gpt are really good to
create a specific automationthat just repeats itself again
(30:25):
and again and again.
But you can do the same thing ina project, whether a charge GPT
project or a Claude project,the.
Projects also allow you to havean open ended conversation about
the topic that it has in itsinstructions and knowledge base,
which provides another benefit.
It has memory that helps itlearn from one chat to the
other.
(30:45):
And it keeps all theconversations of that topic in a
single place that they're easierto find.
So overall projects right noware winning over custom gpt.
I will say one more thing thatI've noticed because I'm working
in both environments and whenyou're running projects in a lot
more cases and you, when youdrop in the input, it will
actually gonna ask youquestions, clarification
(31:06):
questions that it never actuallyasks me.
Custom gpt that is good and bad.
It is good because it isverifying information that it is
not sure about on how to executethe task.
It is bad because sometimes Ijust wanted to do the task, uh,
and.
And so that's another smalldifference, which is a nuance.
I think that in most cases, Iprefer when it's asking me these
(31:27):
questions, when it's running theproject, because it helps me
clarify things that the AI isnot certain about.
That is it for today.
If you haven't used thisfunctionality, this can change
your life.
Literally, every single processthat I have in my company is run
through either projects or thesecustom GPTs, and it saves me
hours and hours and hours everysingle week between myself and
(31:49):
my team.
And without it, I probably wouldnot be able to do all the things
that I'm doing right now.
So if you haven't created those.
Go and experiment with them.
You can try to do both.
See where you're getting betterresults.
In most cases, it'll be verysimilar if you follow the
process that I explained earlierin this episode, and I would
love to hear your feedback aboutthis after you tested out.
(32:09):
So look me up on LinkedIn, ISARMetis, and send me a message.
Say, Hey, I tried projects and Itried custom gpt and here's what
I found.
I would really appreciate that.
If you are enjoying thispodcast, please.
Give us a review on ApplePodcast or Spotify, and while
you're opening your phone rightnow to do that and thank you for
doing this.
Click the share button and sharethe podcast with a few people
(32:30):
that can benefit from thispodcast.
I'm sure you know a few and Iwould really appreciate it.
They would probably appreciateit as well, so everybody wins.
Keep on experimenting with ai,keep on sharing with the world
what you are learning, and havean amazing rest of your week.
Hello and welcome to the Leveraging AI
(00:02):
Podcast, the podcast that sharespractical, ethical ways to
leverage AI to improveefficiency, grow your business,
and advance your career.
This Isar Metis, your host, andwe have a really interesting
episode and.
I have a really interestingepisode for you today.
There is a relatively bigconfusion between different
tools that chat GPT provides toall of us basically right now.
(00:25):
And what I mean specifically isthe difference between custom
GPTs and projects.
Now, if you use one.
Or the other, or maybe a littlebit of both, or definitely if
you haven't used any of them,you'll find significant value in
this episode and you will learnhow to actually implement them
effectively.
You'll learn how to pick theright one for your project and
even how to.
(00:46):
Create specific automations thatyou can use in your business and
areas of expertise that you canuse in your business in order to
make the most out of chat GPT.
We're also going to touch alittle bit about the differences
between that and how to useprojects in Claude and some
minor differences between thisand that.
So if you want to understand howto use ChatGPT projects or
(01:07):
custom GPT or cloud projects inthe most effective way, how to
develop them correctly and whatthe hell are the differences
between them?
this episode is for you.
Before we dive into explainingall the differences and so on, I
want you to understand what arethe options, like what the hell
can you do with these tools?
And so I'm going to show youdifferent examples of
(01:27):
automations as well as differentother ways to use projects.
And then you at least have anunderstanding of what are they
good for, and then we can talkabout the differences and how to
pick the right ones and how toimplement them.
Let's start with an example ofhow you can use, in this
particular case, a custom GPT todo data analysis across several
(01:47):
different files.
What I'm going to show you rightnow is a really interesting
automation process that actuallytakes two separate files.
This process was built on a realprocess that I've done for a
client.
But using fake data and whatthis process does is actually
extremely helpful.
So they sell across multipledifferent channels.
They get an updated levels ofinventory from all these
(02:09):
different channels and everysingle day they have to combine
all that information and thencompare it to the sales data and
then based on the new levels ofinventory, they need to create a
new updated sales brief thatgoes to their salespeople.
So the salespeople know what tosell more, what to sell less,
what is inventory they ran outof, and they need to stop
selling whatsoever what theyhave a lot of, and they need to
(02:31):
run promotions and so on.
This process used to take themseveral different hours every
single day.
And then we've built onautomation routed using a simple
custom GPT.
So the inputs to this custom GPTare just two files.
One is yesterday's salesbriefing, and the other is
today's updated inventoryinformation after it was
(02:52):
compiled from all the differentsources.
And what you'll be able to seeis that this GPT now will run
four or five different steps,all its own, each and every one
of them.
Highly detailed by comparingcolumns and rows and
cross-referencing data from bothfiles.
And it is just going to read thefiles and then start doing.
So say, did step one and nowit's going into step two, and
(03:12):
it's gonna analyze step two.
And you can see in each stepexactly what it's going to be
doing.
So it's gonna pull the materialfrom one file compared to styles
in the other file.
And then considering all theroles matching to description,
it's doing a lot of highlydetailed work.
And now it's going to stepthree, each and every one of
these steps used to take humans.
Multiple minutes and sometimesthe combined effort would've
(03:34):
taken, as I mentioned, two tothree hours.
And now here we are and we'redone.
So you see now I have adownload, updated inventory, uh,
file that I can click well checkit first and then send it to my
sales team immediately.
So this took about.
Three minutes of me talking,meaning I could have done an
email or something if this wasmy actual work, and I've got the
file that previously used totake hours to do by
(03:56):
cross-referencing two differentfiles in two different formats
with different namings and soon, and yet it knows how to do
this effectively every singletime.
Let's look at a completelydifferent example.
In this example, I will show youa automation process that writes
hooks for my posts.
So how does that work?
Well, initially, my team hascollected multiple hooks from
(04:17):
many different sources on theinternet.
Ones that work and drove a lotof engagements.
And now all I have to do is dropa post into it and it knows how
to do that, and it knows how tofind relevant hooks that I can
use and recommend them to me,and I can pick the one that I
think makes the most amount ofsense.
So let's take an example.
Let's go to LinkedIn.
Let's pick anybody's post.
It doesn't really matter.
(04:38):
I'm picking out a post and I'mpicking it without the hook line
from the beginning, and then I'mgonna drop it into.
In this particular case, customGPT.
Then I'm gonna click go again,no instructions, no nothing.
And then the tool itself will gothrough multiple different
options, uh, of hooks and itwill find the ones that probably
will work best for this post.
(05:00):
So you can see now it gives me,uh, three different options
option.
Hook template number 14, storycategory, how I went from X to
Y, so this is the template, andthen the way it's suggesting to
use it is how I went from 12views on my first LinkedIn post
to speaking at a HubSpot eventalongside the CEO of a$2.1
billion company.
That is not bad.
(05:21):
Definitely a hook.
I will probably click to seewhat the rest of it hook.
And then it gives you anotherhook option.
Most people think X, butactually Y.
So this is the template.
And again, the recommendation ismost people think your first
version needs to be perfect, butactually your worst work is the
data that gets you to your best.
So I think the previous one isbetter, and then it give me
another one, and I can ask foras many more as I want, and then
(05:43):
I can pick the one I want touse.
Or at least get ideas and then Ican write my own, or I sometimes
combine some of them togetherinto one hook.
So this is, again, somethingthat just helps me be more
creative and follow templatesthat have been successful for
other people.
I'll show you one more examplethat I use all the time, and
that saves me hours.
Every single week.
I get approached regularly, soprobably several different calls
(06:05):
per week to provide AI workshopsto companies.
And these always work the sameway where I have an initial.
Call with usually a seniorperson, the CEO or the COO or
somebody like that, or somebodyfrom l and d that is asking me
about my services.
And sometimes it's follow upwith another call with somebody
more senior on another personthat can weigh in on how the
(06:26):
training should go, and then afew emails go back and forth.
And then they're asking for aproposal.
So either one call or several.
And then I need to write aproposal, and this used to take
me writing the proposals abouttwo hours per proposal.
And now what I do is I actuallyhave a very simple process where
I take the transcriptions of thecalls from Fathom, which is a
tool I use that listens to allmy calls, re regard whether
(06:48):
they're on Zoom or Teams or uh,Google Meet.
And then I drop that into thiscustom GPT or project and it
writes the proposal for me.
So a process again that used totake me about two hours now it
takes the AI about five minutes,takes me another 10 minutes to
review it, correct what I needto correct, drop it into my
header and footer in a regulardocument and then send it over.
(07:10):
So a huge savings of time,especially that I'm doing two to
three of those every singleweek.
That gives you an idea of howmuch time I'm saving just with
that one automation.
So these are the things you cando with it when it comes to
automations, but actuallyprojects can be used for a lot
more than that.
So now let's take a look atprojects for a second and see
exactly how that looks like.
So where do projects live insideof ChatGPT and how exactly can
(07:35):
you use it?
Well, it lives on the left sidemenu, the top thing you all.
Every one of you has, it saysNew G.
It says New chat, search, chats,library, and so on.
And then underneath that youhave projects.
And projects can be used in twodifferent ways.
One is in similar ways to customgpt.
Again, we'll discuss exactlywhat that means, but the other
(07:55):
way is as a space to haverelevant context for a specific
task.
So what you can see here is youcan see that I'm looking right
now at a project that is calledSample Client Project.
The reason I have it is I don'tnecessarily wanna share what I'm
doing, these demos, how I'mactually working with each and
every one of my clients.
But how am I actually using itis really important and not
(08:17):
necessarily the details of aspecific client.
In general, all these AI toolsare thriving on context.
The more context they have, thebetter results you are going to
get.
Which means if you want to getgreat results when you're
talking to a specific client,you need to tell ChatGPT or
Claude or Gemini or gr.
Doesn't matter which tool youuse.
(08:38):
Everything you know about.
The client, the project, thepeople you're working with, the
specific proposal, theeverything that you want, which
just takes a lot of time andit's obviously not very
effective doing this.
Every single chat that you have,and this is exactly one of the
things projects are very goodat.
You can provide the chat.
With background information,context of the specific client
(08:59):
or on the specific project or onthe specific topic you are
working on regularly.
For me, as an example, I do AIcourses.
I have this podcast, I have aYouTube channel, so each and
every one of them has its own.
Project.
So what is in the project?
The project has a combination oftwo different things.
One is files or knowledge basein the language of custom gps.
(09:19):
So what could be these files?
Well, they could be everything.
If I just want to have a.
Project about a client that I'mworking with.
I will have something like this.
So I have information about theculture of this client as a
document.
How do I know that?
Well, first of all, I go totheir website and I see what
they write about themselves, butthen I take, uh.
Additional stuff that I knowabout them, and I add it to that
(09:40):
document and I upload that as adocument.
I have different proposals thatI've written for them, so you
can see a course proposal and aworkshop proposal.
You also have a consultationagreement that I have with them
in here, and people documentwhat's in the people document.
I will create a deep researchproject and I'll ask it to
research the top people that I'mworking in, the company and the
decision makers.
Then I will go to LinkedIn and Iwill pull some additional
(10:02):
information from there.
And then I have a segment inthis document about each
interview people that I'mworking with as well as hire
decision makers in the companyif they're not the people that
I'm talking to.
And I upload that as well.
And then about us.
So I'm literally taking theabout us page from their website
combined with deep researchabout the company, saving it as
a document and uploading that asa document about this client.
(10:24):
So.
What happens now is if I amopening a new conversation here,
different than a regularconversation in ChatGPT, this
conversation under projectsalready knows all of that
information.
And so instead of giving mevanilla answers, it knows
everything there is in here,including.
Projects, including proposals,including work we've done
(10:46):
before, including emails.
If they're important, I willinclude them in here about the
company and so on.
And hence, the answers that I'mgetting are specifically for
this company are related to thework we've done with them and so
on.
So extremely helpful.
But in addition to just thefiles, it also has instructions.
So if you go to the, theellipses, the three dots on the
top right corner in projects.
(11:08):
You can see that I have asegment.
Uh, I have a button called editinstructions.
If you click on editinstructions, you can see it has
instructions and you can seethat this one says Comprehensive
AI Strategy Development forclient.
And then I have differentobjectives and how to work with
it and what are the benefits andwhat are the pros and cons and
how I want it to respond and soon.
And so the way it is going toanswer is very, very different
(11:31):
than the way just a regular chatis going to answer because it's
custom tailored for this clientand for the kind of work I wanna
do with this client and the wayI want to communicate with GPT
about this client.
This is a complete game changercompared to using just a regular
conversation with chat GPT Now,because you can create these
(11:51):
smaller bubbles of context.
Each and every one of them canbe specific just to that topic,
which again, could be a client,could be a project, could be
something that you're working onregularly, et cetera.
So this is one way to useprojects.
But before we dive into theother way to use projects, I
wanna show you what our customGPT is, because that will help
you understand how you can useprojects as well.
(12:12):
So custom GPTs are these miniautomations that can do one
thing again and again, just likeI showed you before, either find
a hook or create proposals oranything else that is a
repetitive task in your company,whether they live.
They live down here on the leftmenu of chat, GPT, where it says
gpt, and underneath that it willsay Explore and then it will
show you a few of your gpt ifyou don't see all of them.
(12:33):
Then if you click on explore onthe top right corner of the
screen, you will see my gpt.
And when you click on that,you're gonna get a list of all
your gpt.
I have about a third of them ormaybe more so.
Let's look at the one thatgenerated the hooks for us
before.
So if I click on that, uh, if I,on the top left corner, it says
post hook generator, and it hasa little dropdown menu, and if I
(12:53):
click on that, I can click onedit and then we can see the
structure of how this works.
So in this particular setup,what you see is once we get into
a custom GPT, you can see it hasa name, which it just tells you
what it is.
Then it has a description whichtells you in more detail what it
is, and then it hasinstructions.
And in this case, theinstructions are really, really
short.
You're an expert copyright whospecializes in writing hooks for
(13:15):
LinkedIn post.
I want you to write three hooksfor any LinkedIn post that I
provide.
One, please refer to the hooksfrom the Resource 750 plus
Hooks.
Two.
Select three hooks, templatesthat work.
Three, please include the Hooktemplate and the number four,
use the templates from theresource two, craft a Suitable
Hooks.
That's it.
(13:35):
Really short, but reallypowerful.
So how does it know where topull the hooks from?
Well, down here it has what'scalled knowledge base.
And in the knowledge base youcan upload files in one of the
files.
Up.
Up I've uploaded has a documentthat has over 750 hooks with
examples on how to use them.
Now the other thing you have isyou have.
(13:56):
Conversation starters, what areconversation starters?
They add buttons on the customGPT.
These buttons can be used as oneof two things, either as to give
somebody an idea of what theyneed to do.
So in this particular case, yousee it says, input your LinkedIn
post.
So I know what to do.
Otherwise there's no way for meto know.
So if I'm gonna share this withsomebody, they may or may not
(14:16):
know what to do with this customGPD.
But the other thing that you canuse it for is you can create
several of these conversationstarters, and then in the
instructions refer to each andevery one of them separately,
which means you can allow theuser to.
Start with a different kind ofdata or a different kind of
source, or a different kind ofstarting point that based on the
(14:37):
button that they have clicked,which just makes it more
versatile.
So these are the capabilitiesinside a custom GPT.
And as you've seen.
All you have to do is uploadsomething and get the output,
because it already has theinstructions.
You don't need to tell it whatyou want it to do, and you will
do the instructions time andtime again.
If you wanna dive deeper intohow these things work and how to
develop them.
Uh, there's a separate episodethat we've done about it.
(14:59):
So if you go back to episode 175, it was called Stop Wasting
Time, automate Repetitive Taskswith Custom gpt.
You'll get a little moreinformation that we're sharing
today, even though I'm gonna tryto share a lot of it with you
today as well, even thoughtoday's episode is more about to
compare that with projects andhow to choose the right solution
for you.
Now down here on the bottom, youhave two more things.
(15:21):
Inside the setup of custom gpt,you have the capabilities that
you wanted to include, and thatincludes web search canvas.
Image generation and codeinterpreter in data analysis and
just pick the ones you actuallyneed in the custom GPT versus
everything.
And then on the bottom, on thevery bottom, you have create new
action.
What are actions?
Actions are the ability for thecustom GPT to actually run code
(15:43):
and connect to other things inyour tech stack.
So.
That means if you have othertools or databases and you wanna
be able to either query thedatabases or connect through an
API to third party tools, youcan do that inside a custom GPT,
which makes it even morepowerful because now it can
query your CRM, your ERP, youremail marketing platform and so
(16:03):
on, and work with it in tandemto give you the results that you
want versus.
You having to export data, copy,paste it into the custom GPT and
so on.
Now, if you don't know how towrite code, then there's a
little button here that says Gethelp from action GPT.
That uses another.
Custom GPT that OpenAI createdto write code for you to put it
(16:24):
in here, for it to connect withthird party tools.
Most people, and in most cases,that is not necessary, but it is
a huge benefit if it's somethingthat you need or want to do.
So these are custom gpt now on avery high level, these two
things, projects and custom gpthave very similar concepts,
meaning they have a knowledgebase of files that you can
(16:44):
upload, and they haveinstructions that they know how
to follow in order to deliverspecific kind of outcomes.
So let's dive a little deeperinto what are the differences
between projects and custom gpt.
That will give you a hint onwhich one you need to pick for
different use cases, and I willgive you an idea.
As of right now, there's a veryclear winner that you should
(17:07):
pick probably 95% of the time.
So let's start with the purpose.
Is there a difference in purposebetween custom gpt and projects?
And the answer is, it depends.
Custom gpt are built to be arepetitive task creator.
Basically, if you have a.
Single task that you wannarepeat again and again and
again, then Custom GPT will dothat.
(17:29):
Projects can also do that, butin addition, they can be used as
a way to have conversationsabout a specific topic, meaning
it's an entire workspace thatcan use similar kind of
information for two differentgoals.
One is to run an automation thathas to run exactly the same way.
Or to have open-endedconversations about the same
(17:51):
background information.
Now, to be fair, in most cases Icreate two different projects
for these two different tasks.
One of them would be to createvery specific instructions that
will repeat a task again andagain and again and the other
type of project will be a.
And the other type of projectwill be just a data source for
open-ended conversations with acontext about this particular
(18:15):
topic.
But in general, custom GPTs area one trick pony.
They will repeat the task thatyou give them, and projects can
do that plus allow you to haveopen-ended conversations.
The next topic is knowledge baseor files that you can upload
that is very similar on bothtools, so both on custom GPT and
projects.
You can upload PDFs, worddocument, excels images, et
(18:35):
cetera, and that is very similaron both sides.
The file limit is.
Also very much the same.
There's slight differencesdepending on the level of
license that you are using.
But in general, on the pluslevel, which most people are
using, you get 20 files of five,112 megabytes.
Each with 20 megabytes maximumper images, and as far as the
(18:57):
number of files that you'regetting in the projects that
goes up if you have the higherlevel plans.
But usually the 20 files is morethan enough for what you need.
Now here is where it getsinteresting.
The next thing we're going tocompare is how can you share
them with others?
so custom GPS can be shared inseveral different.
The very basic is just you.
You are not sharing it, andyou're just using the custom GPT
(19:20):
that you created.
Option two, you can share itwith anyone with the link.
Option three, you can share itif you have the enterprise level
with other people from yourcompany.
And option four, you can shareit with what's called the GPT
store, which means anybody whohas access to chat, GPT can use
your custom GPT on projects.
Until not too long ago, youcould not share it at all.
(19:42):
Which was the biggest and moreor less only disadvantage of
using project that changed backin September for people with
enterprise level licenses, andit changed for everyone in
October.
As of last month, now you canshare projects just by giving
people the link and they canwork inside the same projects
(20:02):
with you, which is a hugebenefit, and again, was the only
real serious disadvantage ofprojects before.
So that was a big advantage ofcustom gpt that is no longer an
advantage.
And now let's talk about the twobiggest advantages of projects
over custom GPT.
Biggest benefit number one ispersistent memory.
So custom GPT is every time yourun the custom GPT, it runs and
(20:26):
it's done.
It doesn't know what happenedthe previous times you run the
same custom GPT.
However, projects, as Imentioned, were originally built
not to do automations, butactually they were built in
order to be a context space, abubble of context for a specific
topic, and they have their ownmemory feature.
So just like there is a memoryfeature for ChatGPT as a whole.
(20:49):
There is a memory feature justfor the project, so every new
chat within that space adds morecontext to the overall
knowledge, experience, andcapability to be more precise
with the answers of thatproject, which does not exist in
a custom GPT, which means it's ahuge benefit and a lot more
value.
Two projects versus custom gpsbecause it learns from every
(21:12):
conversation that you or otherpeople who use the same project
are having within the project.
So this is huge benefit numberone.
Benefit number two is theorganization of the chats.
If you use a custom GPT, everytime you use it, it creates a
new regular chat.
Inside of chat GPT.
I will show you an example soyou understand what I mean.
As an example, you can see herecustomized proposal as one of
(21:34):
the regular chats that I had.
It says Your chats and it wasdone.
With the custom GPT, meaningevery time I run the custom GPT,
it actually creates a completelyregular chat as part of the
very, very long list of hundredsof thousands of chats that I
have done with chat GPT in thelast three years.
However, if I create the sameproposal inside the proposal
(21:55):
generation project.
You can see that it shows upevery single time I ran it
inside just this proposalfolder, meaning it is a lot
easier for me to findconversations that was done,
that are relevant to a specifictopic that I am looking at right
now.
So.
Both from the perspective oflearning from every single chat,
(22:16):
as well as from the perspectiveof having it a lot better
organized in one place, runningautomations, and definitely the
other option of just having openconversations because that's
something you cannot even doinside a chat.
GPT is a added value of aproject.
Now let's talk about twoadvantages of custom gpt that do
not exist on the project side.
(22:37):
Number one, it is the ability tocreate code that will connect to
third party tools such as yourC-R-M-E-R-P.
Et cetera.
And as I mentioned, that existson the bottom of the custom GPT.
It requires writing code thatyou can use AI to help you
write, but it means you need tobe a little more technical.
And to be fair, I have about 30plus custom gps that I use and.
(22:58):
One or two of them actually isusing this capability to write
code.
That capability does not existinside of projects.
So if you need that, then customGPEs are your only option.
And then the second thing isconversation starters.
We talked about this, that youcan create these buttons.
Inside of a custom GPT that willexplain what to do, or that can
start at a different startingpoint.
(23:20):
Inside of the custom GPT.
This does not exist in projectsat least yet.
So what does that mean?
It means that for the vastmajority of use cases.
Projects are right now a betteroption than custom gpt.
Meaning if you have old customgpt, there's no reason to go and
convert them because there's nopoint.
(23:40):
However, if you want to createnew automations or new
workspaces for you and your teamto work on a specific topic,
either through an automation oras a context baseline for
open-ended conversations, justcreate projects.
There's really no reason.
To create custom GPEs right now,unless you wanna connect it with
code to something else.
(24:01):
Now, I want to give you a fewmore hints on how to create
these automations, whetheryou're creating them in
projects.
Or in a custom GPTI always startwith a regular chat.
I don't actually start trying tofigure out what needs to be the
instructions.
I bring it the data.
So every time I create anautomation, there's really three
components, right?
There's the input, the process,and the output.
(24:22):
So I need to have a clearunderstanding of what the input
is.
So I bring in the input, I cleanthe data in the input.
So the data needs to be verywell organized.
If it's in Excel, then it needsto be set up correctly without
any double headers, without anyspaces, without any merged cells
and stuff like that.
So I clean the data, I bring itinto the chat, and then in the
(24:43):
regular chat, not inside tryingto build an automation, I try to
get.
To define the process in orderto get the outcome, and I just
iterate.
I ask it, oh, I need to do thisand that, and then it gives me
an opposite.
Oh, it's not what I meant.
Let's try this.
Let's try that.
Until I get to the final output,the way I want it, once I get to
the required outcome in theformat I wanted, in the length I
wanted, with the details that Iwanted, I ask it to write
(25:06):
instructions.
For a repetitive custom GPT thatwill do the entire correct
process and avoid the mistakesthat we've done in the regular
chat.
It writes amazing instructions.
Way better than I can write, andmost likely better than you can
write on your own and it willcapture all the nuance.
You can ask it to ask youquestions if there's any open
things it's not sure about, andthen it will write the
(25:27):
instructions for you.
You just copy and paste theinstructions into the
instructions section, either inthe custom GPT or in the
project.
You also tell it what kind ofreference material or knowledge
base you will give it, and askit to include that or reference
for that in its instructions.
And then add those files intothe knowledge base inside of the
(25:47):
custom GPT or the files insideof the project.
So now you have everything youneed in order to run these
automations.
And you can also use it if youcreated a project, which again
would be my recommendation.
You can also have open-endedconversations based on the
information that is in the filesand in the instructions.
And as we mentioned, if it is aproject, it will also learn over
time and get better over time.
(26:08):
So now that you know thatprojects is probably the way to
go for most of your use cases,there's a very similar concept
in Claude also called projects.
So what are projects in Claudeand how they are different from
projects inside of a custom GPT?
Well, not by much actually,projects in Claude was there
before projects in custom gptand OpenAI, more or less copied
(26:33):
the concept of projects fromClaude.
This is why OpenAI currently hastwo different features that are
very, very similar becauseOpenAI started with custom gpt.
Then Claude created projects,which was broader and more
capable, and then OpenAI copiedit and provided projects inside
of Chat GPT, which was not myrecommendation until a month
(26:53):
ago, where now you can actuallyshare them with other users,
making them a lot more powerfuland relevant to most people than
custom GPTs.
But let's talk about what arethe differences between cloud
projects and chat GPT projects?
So I already showed you howChatGPT projects look like.
Let's go to Claude and see thatit's more or less identical.
So if I'm in Claude over here,there's like a little folder
(27:16):
icon on the left, and if I clickon that, it takes me to projects
and very similar, I can create anew project.
But if you click on a project,you will see that it has a
memory, that it remembers stuffabout you, it has instructions,
and it has files In thisparticular case.
Uh, you can see that it has, 10different files and it has a set
of instructions that tell itexactly how to work and same
(27:37):
kind of thing.
This is a proposal generator andall I have to do is drop in
whatever the it is looking for,and you will know how to run
this automation.
But I can also have open-endedconversations about this.
Topic, which is AV proposalgeneration in this particular
case, and it can give me ideason how to write better proposals
and what kind of proposal am Iwriting and maybe a new
(27:58):
different kind of structure, orhow should we approach a
specific client and so on and soforth.
Because it has all theinformation about previous
proposals that I have written.
But there are still somedifferences between cloud
projects and chat GPT projectsand let's review them quickly.
Claude in general is much betteras of right now, so Claude 4.5
(28:18):
compared to ChatGPT 5.1, Claudeis much better in formatting.
The documents that it'sgenerating and the Excels that
is generating is significantlybetter looking and well
organized than what ChatGPT doesright now, which is a huge
benefit, especially if you arealso using skills.
I'm not going to dive intoskills right now.
I'm gonna do a complete separateepisodes about skills.
(28:40):
But skills are these minicapabilities that you can create
inside of Claude that knows howto do very specific things.
As an example, I have created aClaude scale that creates
everything to my brand when Iask it to do.
So, it has my color coding, thetone that I like to use,
examples of how I writedifferent things.
(29:01):
It has my logo, it has my entirebrand guidelines and it knows
how to apply it very, very wellwhen it is creating new
documents.
So the combination of that withinstructions means I have to do
less work afterwards to formatit as I need.
The disadvantage of that and theadvantage of ChatGPT is that in
both cases, when you create itin ChatGPT, it creates it in
(29:22):
Canvas if you ask it, whichmeans it's an editable work
collaboration document togetherwith the ai, meaning, you can
create manual changes to theoutput from ChatGPT versus in
Claude it looks better, but youcannot make any edits.
Meaning if you wanna make edits,you have to copy this into a
Google Doc or a Word document inorder to make these changes.
(29:43):
And in ChatGPT, you can makeedits.
On your own, combine it with thework together with ai, that I
find a lot more helpful thanusing artifacts, which is the
Claude version.
So what does that mean?
It means that right now I amusing ChatGPT more than I am
using Claude because of thatparticular reason I have.
(30:04):
Multiple times taking the outputfrom ChatGPT, dropped it into
Claude to get the formattingdone much better.
And then I have less work in themanual editing afterwards when
it comes to just making it looknice.
So final summary of all of thatcustom gpt are really good to
create a specific automationthat just repeats itself again
(30:25):
and again and again.
But you can do the same thing ina project, whether a charge GPT
project or a Claude project,the.
Projects also allow you to havean open ended conversation about
the topic that it has in itsinstructions and knowledge base,
which provides another benefit.
It has memory that helps itlearn from one chat to the
other.
(30:45):
And it keeps all theconversations of that topic in a
single place that they're easierto find.
So overall projects right noware winning over custom gpt.
I will say one more thing thatI've noticed because I'm working
in both environments and whenyou're running projects in a lot
more cases and you, when youdrop in the input, it will
actually gonna ask youquestions, clarification
(31:06):
questions that it never actuallyasks me.
Custom gpt that is good and bad.
It is good because it isverifying information that it is
not sure about on how to executethe task.
It is bad because sometimes Ijust wanted to do the task, uh,
and.
And so that's another smalldifference, which is a nuance.
I think that in most cases, Iprefer when it's asking me these
(31:27):
questions, when it's running theproject, because it helps me
clarify things that the AI isnot certain about.
That is it for today.
If you haven't used thisfunctionality, this can change
your life.
Literally, every single processthat I have in my company is run
through either projects or thesecustom GPTs, and it saves me
hours and hours and hours everysingle week between myself and
(31:49):
my team.
And without it, I probably wouldnot be able to do all the things
that I'm doing right now.
So if you haven't created those.
Go and experiment with them.
You can try to do both.
See where you're getting betterresults.
In most cases, it'll be verysimilar if you follow the
process that I explained earlierin this episode, and I would
love to hear your feedback aboutthis after you tested out.
(32:09):
So look me up on LinkedIn, ISARMetis, and send me a message.
Say, Hey, I tried projects and Itried custom gpt and here's what
I found.
I would really appreciate that.
If you are enjoying thispodcast, please.
Give us a review on ApplePodcast or Spotify, and while
you're opening your phone rightnow to do that and thank you for
doing this.
Click the share button and sharethe podcast with a few people
(32:30):
that can benefit from thispodcast.
I'm sure you know a few and Iwould really appreciate it.
They would probably appreciateit as well, so everybody wins.
Keep on experimenting with ai,keep on sharing with the world
what you are learning, and havean amazing rest of your week.
Popular Podcasts
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
The Brothers Ortiz
The Brothers Ortiz is the story of two brothers–both successful, but in very different ways. Gabe Ortiz becomes a third-highest ranking officer in all of Texas while his younger brother Larry climbs the ranks in Puro Tango Blast, a notorious Texas Prison gang. Gabe doesn’t know all the details of his brother’s nefarious dealings, and he’s made a point not to ask, to protect their relationship. But when Larry is murdered during a home invasion in a rented beach house, Gabe has no choice but to look into what happened that night. To solve Larry’s murder, Gabe, and the whole Ortiz family, must ask each other tough questions.