September 16, 2025 • 40 mins
Can your AI tool actually do what it promises or is it just fancy fluff?
When it comes to running your business, choosing the right AI platform isn't just about features—it's about real-world performance under pressure. In this episode, I share my experience which AI platform is actually better for things like custom workflows, client-specific automation, data dashboards, deep research, or image generation, this episode is your ultimate field guide.
In this session, you'll discover:
- Why ChatGPT crushed Gemini in memory, personalization, and multi-step automations
- How Gemini unexpectedly won in dashboard building and deep market research
- Real use cases comparing Gems vs. Custom GPTs and the key limitation that still plagues Gemini
- A real-world test of building data-rich dashboards in Gemini vs. ChatGPT (with shocking results)
- How both tools struggled to pull business travel info across emails, proposals, and calendars and why that matters
- A step-by-step visual branding test using AI image generation and who came out on top
- The one killer feature ChatGPT offers that Gemini just doesn’t match (yet)
💼 Want to work smarter with AI? Follow Isar on LinkedIn: https://www.linkedin.com/in/isarmeitis
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.
GMT20250911-164740_Record (00:00):
Hello and welcome to the Leveraging AI
Podcast, the podcast that sharespractical and ethical ways to
leverage AI to improveefficiency, grow your business,
and advance your career.
This is Isar Metis, your host,and we've got a very interesting
episode for you today.
The reason it's interesting isit is based on my own personal
experience that is tied to aworkshop I have delivered two
weeks ago.
(00:20):
So two weeks ago, I was invitedto provide a several hour
workshop at a very large reallyknown tech company in San
Francisco.
The company that invited me usesGoogle as their office
infrastructure enhanced therequest was to focus most of the
workshop on Gemini as well asother Google AI tools, and
that's what I did.
(00:40):
But to do that, I had to take alot of the workshops that I've
been teaching to many othercompanies and customize it and
test it on the Geminienvironment, which gives me a
very unique insight to how it iscompared to other tools,
especially ChatGPT.
Now I've learned a lot in theprocess and I thought it's gonna
be highly valuable to you too,to learn from my experience from
(01:02):
what I've learned from trying toconvert use cases from Claude
and ChatGPT, mostly ChatGPT intothe Gemini universe.
And so this episode is not goingto be structured in a very
specific flow because these areunique specific use cases.
We're just gonna go use case byuse case and compare what the
results were and what were thepros and cons of each and every
one of these platforms.
There is a winner, which we'regonna get to in the end, but
(01:25):
let's get started with thecomparison.
(02:10):
The first thing that I want totalk about is a concept that is
critical for the success ofanything you do with ai, and
that is the concept of context.
These tools lack context, right?
They don't know you, they don'tknow your company, they don't
know your industry, they don'tknow your unique expertise, and
because of that, a lot of theanswers you're going to get are
generic.
So the more the tools.
(02:31):
Know about you, the moreinformation you provide them in
order to learn about youruniverse, the more specific, the
more accurate, the more relevantanswers that you're going to
get.
That is true for all thesetools, which means that if you
want to get accurate answers,you need to write very long
prompts every single time,unless you can teach the model
stuff about you that it willremember for the long run.
And that is a very bigdifference right now between
(02:53):
ChatGPT and.
All the other models, but inthis particular case, Gemini as
well.
So let's look at what I'mreferring to.
So what I did here is Ibasically asked ChatGPT, what do
you know about me?
A very simple andstraightforward question.
You can do that yourself to testit yourself.
You can try it with differentmodels.
But let's see what ChatGPT knowsabout me.
And this is just a questionthat's, there's nothing else in
(03:13):
this prompt.
And it says, here's a structuredoverview of what I know about
you based on our pastconversations and context.
You've shared background, formerF 16 pilot in the Israeli Air
Force.
Later on, instructor in the AirForce Academy transitioned to
business and technology with 15plus years as a CEO experience
in startups.
People post and largecorporations last minute travel,
hotel beds group,, moved to theUS about 20 years ago,
(03:36):
professional journey built andscaled multiple companies,
including taking one public andselling another.
Founded Multiplai in 2023,focused on AI education and
consulting.
Provides, and so on and soforth.
Then it has, content education,which are the things that I'm
focusing on right now.
It talks about this podcast, theLeveraging AI podcast, about our
newsletter, about the hangoutswe do every single Friday, and
about my collaboration with theCrummer School of Business,
(03:59):
which is the business school Idid my executive MBA at, and now
I've been collaborating withthem for the past year and a
half in delivering AI educationto their audience.
And then our other initiativethat I'm involved in, market and
audience, where it is, who amI'm catering to, personal notes
about myself, about my family,what I'm passionate about, what
my family does, and so on.
And some key takeaways.
All of that, it knows about me,which means when I'm asking it
(04:22):
anything, it uses that referenceinformation in order to provide
more relevant answers to me.
I've done the same exact thingwith Gemini and asked Gemini
what it knows about me.
So same question.
What do you know about me?
And the answer is, based on ourcurrent interaction, here's what
I know, just as important, whatI do not know.
What I know, our currentconversation, and then it says
what we talked about in thatparticular conversation, time
(04:43):
and location.
Based on the informationprovided to my system.
I know it is Thursday, August7th at 10:49 AM technical
information, very little.
And what it doesn't know, itdoesn't know my name, it doesn't
know my age.
It doesn't know where I live.
It doesn't know my emailaddress.
It doesn't know my browsinghistory.
It's a lot of other stuff itdoesn't know.
So basically it doesn't knowanything about me, which is good
(05:03):
from maybe a securityperspective.
It's bad from a contextperspective to.
Any conversations I'm going tohave.
So I actually take the time fora very long time now, investing
in teaching Chachi piti aboutme, about my business, about my
tone, about my customers, aboutthe proposals that I write,
about the services that Ideliver.
It knows a lot more than itactually shared in that one
summary, and because of that canprovide me much better
(05:26):
information.
The other thing that I use forcontext in chat g PT that does
not exist in the Gemini universeis I use projects.
Projects are on the left side,right under,, the custom GPT
segment just above your.
Chats.
And in there you can createthese folders.
And in these folders you willhave different conversations,
but each conversation can haveits own unique context universe.
(05:48):
So you can see in thisparticular case, this is a
sample client project.
It's not a real client, but Ihave one for each and every one
of my clients.
In there you will see that Ihave different kinds of
information.
I have proposals, for thatparticular client that I've sent
this client I have.
Different files that I'vecreated.
So the culture of the companybased on my experience and based
on their website, what proposalsI've sent them.
(06:08):
So you can see what that is.
The consulting agreement I havewith them, the people in the
company that I engage with andtheir profiles from LinkedIn,
the workshop proposal that Isent them, the about US PDF from
their website.
So more context about thatparticular client.
So then when I haveconversations in this folder, it
has a lot more context aboutthis particular which again is
not something you can do inGemini.
So that's on context, which isdefinitely a huge win on the
(06:33):
chat GPT side.
So let's move on to the next usecase.
The next use case is usingcustom GPTs versus using gems.
So the concept is very similar.
I'll explain in one sentencewhat it is.
So custom gpt and gems allow youto write instructions that you
wanna repeat time and timeagain.
So every time you have arepetitive thing you want to do
with ai, whether it's writing aproposal, analyzing a specific
(06:55):
document, comparing two Excelfiles, uh, creating content, et
cetera, et cetera, every one ofthose can become either a custom
GPT on the chat, GPT Universe orGEMS on the Gemini universe.
I recorded an entire episodeabout it.
And if you want to dive into thetopic of custom gpt, you can go
back to episode 1 75.
It is called Stop Wasting Time,automate repetitive Tasks with
(07:17):
Custom gpt so you can go backand listen more about that
topic.
But there is a similar conceptthat is more or less the same in
the Gemini universe called gems,where you can provide
information and so on.
So the way I create and the wayI teach how to create both
custom GPTs and gems is not bywriting the instructions on your
own, but rather by having aconversation with cha GPT or in
(07:37):
Gemini and then asking ChatGPTat the end of the conversation.
Once you explain to iteverything that you want, once
you ask it questions, once youwork inside the conversation.
On the actual output and youachieve the correct output, then
I would go and say, I would likeyou to now turn everything that
we've done, including the finalsuccessful outcome.
So ignore the stuff that didn'twork into highly detailed
instructions for a custom GPT ora gem, because it creates better
(08:00):
instructions than I can,especially after we've done the
entire exercise and we'regetting to the outcome that we
want.
I've done this in chat GPTmultiple times, and then I've
tried it in Gemini and itdoesn't work as good.
So let's look at an example fromeach and every one of those.
So in this particular case, thisis from a presentation I've done
for the Israeli parliament andI've started the prompt with you
(08:21):
are an expert data researcher inthe Israeli Parliament tset.
We are planning to use a customGPT to help us in the research
and summarization of data.
We would like to build a team ofseparate personalities that will
participate in this process, etcetera, et cetera, et cetera.
I'm writing in this prompt, andthen we're going back and forth
and it gives me suggestions ofwhat these people can be and
what they will do and what aregoing to be the responsibilities
and the roles of each and everyone of them, and what's gonna be
(08:42):
the research process and what'sgonna be the focus on different
things.
And then I ask it for a list ofresponsibilities for each and
every one of these people.
So it went ahead and did that.
So now I have a long list ofresponsibilities.
And then in the end, after goingthrough the entire process, I
said, please create a detailedinstructions for our custom GPT
that will create all thesepersonas.
The GPT should receive an inputthat would be a topic for the
Nset team would like toresearch.
(09:04):
The team should first discussthe topic debate.
And agree on the way forward, etcetera, et cetera, et cetera.
And he wrote these amazinginstructions.
So you can see there's systemdescription.
This is the first thing that itcreated.
Then it created a team overview.
It gave names to each and everyone of the people.
What's their role, what's theirstyle?
So they can have differentpersonalities that will support
the team.
Then a step-by-step workflow.
(09:24):
Step one, topic intake and groupdiscussion.
What are the actions that theyshould follow?
Step two, data collection.
Step three.
Domain analysis, what eachpeople, which, and it even has a
breakdown of who's leading whataspect of the process based on
the personas and the roles andthe actions that they need to
take and the responses that theyneed to provide.
Step four, political and publiclens.
Step five, quantitativeanalysis.
(09:46):
Step six, synthesis anddrafting.
Step seven, risk and ethic.
Review, and then who is the leadon that and the actions for all
of that Step eight final review.
And then there's a communicationformat on how they should
communicate with one another.
And then what is the outputstructure of the report that
they should generate andadditional capabilities.
So, this is an incredible set ofinstructions that, again, there
is no way I could have writtenthis as accurate and as
(10:07):
detailed.
And then I did a similar processin Gemini asking it to create
instructions for a specificbusiness process, not exactly
the same.
And when I did that, it actuallygave me.
Instructions that were okay.
They weren't horrible, but theydefinitely weren't great.
And it was weird because it waskind of like relating it in
third person, and I'll explainwhat I mean.
So my prompt was, I would likeyou to create instructions for a
gem that I can run inside ofGoogle Slides that will ask the
(10:30):
user this set of questions andtopic, blah, blah, blah, blah,
blah.
And then it created a set ofinstructions that should be able
to work, but it invented thingsthat just don't exist.
So as an example, it said thisGem will open in the sidebar in
Google Slides.
It will guide the user through aseries of five steps, asking
questions and so on and soforth.
But the reality is it cannotopen on the sidebar.
So I wrote back, I said, I don'tbelieve the sidebar can have a
(10:53):
next button.
I believe it is a simple chatinterface.
Am I wrong?
but then it said you are rightto question the specifics as the
interface isn't free foreigncanvas.
However, it's more powerful thana simple interface, and you can
absolutely have a next button,which by the way is not true.
At least I wasn't able to see ittrue, but it gave me
instructions that just don'twork.
(11:13):
So, instead of just giving meinstructions for a gem, it's
referring to what needs to bethe instructions in the gem,
which is not horrible.
And in many cases it actuallyworked okay, but it's definitely
not as good as the solution thatI got from chat GPT.
Then I tested something else.
I tested a very complexmulti-step custom GPT that I've
created a while back for aclient, and I was trying to
(11:35):
recreate it with the same exactinstructions on Gemini.
So let's look at that.
So before we dive into this,let's talk about what this
multi-step process does, andit's actually gonna blow your
mind just by itself.
Because what it does is it takesa inventory file that is
completely fake.
I made it up.
All the numbers are made up, butit's an inventory file from
multiple sources showinginventory of different SKUs in
(11:57):
different locations for thecompany.
And it is using it to create asales brief for the salespeople
to explain to them whatinventory we have, what
inventory was were out of, inwhich regions of the world and
so on.
So this is something that manycompanies struggle with, right?
So the ability to takeinformation and compare the
inventory from yesterday toinventory from today, the
(12:19):
inventory from last week to theinventory from this week, and
based on that update salesinformation that they can be
sent to salespeople.
This is a process that manycompanies do completely manually
today, and I've developed acustom GPT that does the entire
process, but it's a five stepprocess, so all I have to do
right now is upload the oldinventory file, the new
inventory file, and the oldsales file to the custom GPT,
(12:41):
and then it follows all thedifferent steps.
It does step one, step two, stepthree, step four compares
different things, uh, where itis, what are the sizes, what are
the SKUs, it's the same packagetype, uh, and so on and so forth
to verify all that information.
And then when it's done withthat step, it goes to step
three, and then it removes allthe roles that.
Do not have the relevant rows inthe other file anymore.
So basically we have it in thesales files, but they don't
(13:02):
exist in the inventory fileanymore.
So it compares three differentfiles back and forth.
It updates all the quantity andthe numbers and the SKUs and
everything to provide me a newupdated file, and then it
provides it to me.
As a CSV file that I candownload to either upload to
whatever CRM system that I'mrunning or to email to my team.
(13:23):
Again, the CSV file can then beconverted to anything that I
want.
So this is an extremely powerfulmulti-step process that I have
tested consistently multipletimes and provides.
Accurate and amazing resultsevery time that I run it.
You can imagine similarprocesses across multiple
aspects of businesses where youhave multiple data sources that
you need to align, compare, andprovide some kind of an output
(13:44):
based on all of them, and itworks flawlessly inside of
ChatGPT.
However, when I try to run it onGemini with the same exact
instructions it literally getsstuck.
It doesn't do anything.
It says, of course, I willprocess the files, uh, according
to the steps provided.
Please provide a moment while Iperform the data comparison,
filtering, merging, andupdating.
I'll provide you with a finalupdated file for download
(14:06):
shortly.
That sounds very promising.
Only nothing happens, and so.
I tried several different ways.
I changed the instructions alittle bit.
Again, the instructions werecopied and pasted, uh, and yet
it does not work.
The inputs are the same exactfiles that I gave ChatGPT.
So in this particular case,again, a huge win for ChatGPT
compared to Gemini when it comesto actually running a multi-step
complex process, with the sameexec instructions on both
(14:28):
platforms.
Now in addition, one more thingthat you can do in custom GPT is
that you cannot do.
In gems is add code to them toconnect them to more advanced
capabilities and connect them toAPI of third party tools,
connect them to MCP servers, etcetera.
So if we look at this particularGPT is connected down here to
the API of data for SEO.
(14:50):
Which is a company that providesdata for SEO, like the name
suggests, but that means that Ican ask questions about keywords
and their availability andcompetitors and their density
and so on across any domain thatI want, within this custom GPT,
and get information aboutmyself, my competitors, and so
on, and Build my future blogpost based on actual real SEO
(15:10):
data right here within chat GPT,without having to go to other
tools and bring information andso on.
This is a very powerfulcapability.
This is called Actions, and it'savailable on the bottom of
custom gpt.
Uh, if you, if I'll open this,you'll see this is just code
that connects it to their MCPserver.
I didn't write the code.
That's a code that they provideto anybody.
So you can go and grab the codeand attach it in here, and you
(15:32):
can then have this kind ofcapability.
The ability to out code into agem just doesn't exist.
So that's another benefit ofcustom GPEs versus Gemini Gems.
The next topic that we're goingto talk about is dashboards.
So I actually think that thebest tool to create dashboards
right now is Claude.
Every time I need to createdashboards, I actually go to
Claude, and Claude does anamazing job in creating visually
(15:53):
pleasing and also very effectivedashboards.
However, I wanted to test Geminion that topic, but before I even
wanted to test it, somethingvery interesting happened.
I was actually working on afinancial analysis, and so the
financial analysis, just as anexample, took, information of a
publicly traded company.
In this particular case, Ibelieve it was Salesforce, and
then it grabbed the informationfrom the internet and it gave me
(16:14):
their information, and I askedit to create a report about it.
So this was a great example onhow you can take raw data off
the internet and create graphslike you can see here, like
fiscal results per quarter, orthen compare that year over year
and just get information just byasking simple questions or even
creating complete reports likeyou can see here inside of
Gemini.
So it created this detailedreport on the financial success
(16:36):
of Salesforce.
In the past X number of quartersbased on the prompting that I
provided in the beginning.
I could have done this obviouslyon any company in the world.
So a very solid solution fromGemini in this particular case,
however, it then suggested onits own to create an interactive
dashboard, and all I have to dois say, yes, I'm interested in
that.
And it created this amazingdashboard from the data that it
has collected based on myprompt.
(16:57):
So you can see it saysSalesforce financial dashboard.
It has key matrix, revenueanalysis, profitability and
efficiency.
Each and every one of them is alink that if I click, it will
scroll down to that particularsection.
Then there's the key findings inbig, bold statuses, and if I
click on each and every one ofthem, it opens another segment
with some more details aboutthis particular topic.
So you can see if I clickdifferent things, it goes to
that.
And then it has revenue andseasonal patterns.
(17:20):
And then it has year over yearcomparison and it created.
All of this on its own.
Based on the conversation that Ihad with it, I did not define
any of the components in thisdashboard.
It just asked me if I would liketo create one.
And when I said yes, it createdthis, which is really, really
cool.
So from that perspective, hugewin.
On the Gemini side, both takingthe initiative and suggesting
it, but also without me definingwhat exactly I want to see in
(17:42):
the dashboard.
It was able to create thisamazing dashboard, but then I
did something else and I wentthe next step.
I said, I have these amazingdashboards that I've created in
Claude.
Can I just replicate them inChatGPT and in Gemini and see
which one does a better job?
So I started with ChatGPT.
And all I did is I provided itthe same exact information that
I provided to Claude, and Iprovided it a screenshot of the
(18:05):
dashboard, so it got the inputfile that I'm going to upload,
basically the same exact filethat I'm using on the cloud
side, and you've got ascreenshot.
Of the entire page.
So a full page screenshot, notjust the top of it, of the
dashboard and all the componentsthat are in it.
And then I ask it and I said,you're an expert in dashboard
application creation.
I'm going to provide you thefollowing, basically what I
explained to you.
And I ask it to create thedashboard.
(18:26):
And it did, and it never works.
So when I click preview.
The first step is correct.
It actually asked me to uploadthe file and start the analysis,
but every time I do this,there's some kind of a bug and
the actual dashboard doesn't rundespite the fact it's the same
exact file.
I, I gave it as the sample datain order to build the initial
dashboard.
So that was my experience withChatGPT.
I tried it several times and Igot the same issue every single
(18:48):
time.
Gemini, on the other hand, gotthe same exact instructions.
I literally copied and pastedthe instruction.
I also copied and pasted, uh,the same screenshot from the
Claude Dashboard, and instead ofcourse, I will create the
dashboard and it went ahead andcreated an amazing dashboard.
It's working, it's fullyfunctional.
I can upload the file, so let'splay the game.
I will click here and I will goand find the file.
(19:08):
Here we go.
Click open.
Click start analysis and boom, Ihave a working dashboard and you
can see that it has, uh, themonthly sales trend and it has
year over year growth and it hasthe top 20 clients and their
trend in the past month comparedto the trailing 12 months.
And it has the distributionbetween the top leading clients
and there's the filters on top.
And I didn't have to explain anyof this.
I literally just showed it ascreenshot of the other
(19:29):
dashboard.
So if I click now, uh, channelone.
You can see that it changes thenumbers.
I can click and say, oh, I justwanna look at Q1, uh, for that.
So it will change that as well.
And you can see that it changesonly to Q1 across the different
years, uh, and so on and soforth.
So all of this is working.
I can go to clients and I canselect the different clients
across different regions andeverything.
We'll move and updateaccordingly across all the
(19:50):
different aspects of thedashboard, really fantastic and
without almost any effort.
So on this particular example,definitely a huge win for
Gemini, for creating the rightcode, for understanding the
need, and for creating a veryuseful and functioning
dashboard.
With very few steps.
Uh, the only thing I had toactually correct was the graph
on the bottom that had all thecompanies, uh, if there are no
(20:10):
filters are applied.
So if I do this, it had like agazillion companies, which made
no sense.
It was too crowded and all I hadto do is ask it to only include
the top 20 companies, uh, in thegraph and everything else just
worked out of the box withalmost no instructions.
So big win.
To Gemini on this particularcreation of the dashboard
compared to Chad G.
Pt that despite all my attempts,was never able to make this work
(20:31):
And now our next topic is deepresearch.
Both tools have deep researchcreating capabilities.
Both tools are the two leadingtools from a deep research
perspective right now.
From my perspective and myexperience, and I use both of
them all the time, I also trythe other tools.
Very regularly.
So about about once every otherweek, I will try one of the
other tools or some of the othertools as well just to see how
(20:51):
they're doing.
So whether that's, perplexity orDeep seek or Claude and I will
try their deep se, deep researchas well.
Still, ChatGPT and Gemini have avery big lead when it comes to
deep research, but let's comparethe two.
And so in this particularexample, what I was looking for
is I was looking for AI softwaresolutions that can help
architects, interior designers,smart home integrators and
builders, and other people whowork with floor plans and stuff
(21:13):
like that.
Because I have multiple clientsin that universe and I wanted to
see what it will find.
So I wrote along in detailed.
Prompt.
I copied the same exact promptto both tools and I let it run.
So you can see the prompt isabout a page and a half long
explaining exactly what I need.
The way it works with Gemini, ittells you what it's going to do,
and then you can say that youapprove the plan or not.
If you don't, it will ask you toclarify what else do you want or
(21:33):
what do you want different?
It's a little different on theChatGPT universe.
It always asks you questions so.
After you write the promptagain, it's the same exact
prompt I copied and pasted it.
it said, to get started, couldyou please clarify the
following?
Are you focused only on AIpowered software, or should I
include any advanced software,even if not AI based that
supports automation, rules,logic, blah, blah, blah.
So it asked me all thesequestions that I gotta answer in
(21:55):
order to clarify exactly whatI'm looking for.
Then both started the research.
Uh, usually Gemini takes a lotlonger when he does the process,
but both tools will work on itfor a while.
And then this is the result Igot from ChatGPT.
So you can see it starts with atable that shows the different
tools and what functionalitythey have.
But what you'll see is thatalready here there's issues
(22:16):
where it's not showing, uh, someof the table.
And if you scroll down, thenyou'll see many of these aspects
instead of being a part of thetable as they should, are just
written in text with thevertical line in between them.
So it didn't really create itcorrectly.
So I have some of thefunctionality here compared, but
then all the other ones are inthis weird format that I cannot
(22:36):
do anything with.
So I messed that up.
But overall, then it startedwith generative.
Design tools for architects, uh,and gave me several different
options.
Then it gave me, uh, AI toolsfor interior designer and space
planning, and then it gave me afew options here, uh, and so on
and so forth.
So overall, a detailed, helpfulreport that is aligned with what
I requested or kind of alignedwith what I requested, and I
(22:58):
explained why.
What I mean by kind of aligned,if you go back to the prompt,
what I ask it to start with, Iask it to start with doing a
research.
On what kind of positions, whatkind of roles exist in these
companies, and what are the usecases of each and every one of
those people?
And only then go and start theresearch on the actual tools to
be able to check if they can dothese functions that these
(23:19):
different roles have to do.
So, if we go to the Geminiversion of the deep research,
you will see that it followedthe instructions.
Perfectly said part one, theprofessional landscape of floor
plan utilization, and then ithas objectives, and then it has
chapter one, core design andplanning professions.
And then it has architecturaldesigners and drafters, and it
has what their needs are andwhat they need to do and so on
(23:39):
and so forth.
And it went to interiordesigner, floor plan related
tasks, space planning.
So it literally followed.
In detail exactly what Irequested.
So it started with a much betterunderstanding of what these
people need before it startedthe actual research on the
people.
Now maybe ChatGPT did the samething, but it not expose that in
its report.
So it's very hard for me toknow.
So you can see all of this, likeI'm still scrolling and
(24:00):
scrolling and scrolling.
For those of you who're notwatching this.
All of this, like pages andpages are just analysis of what
are the roles and what keyfunctions they need that these
software will do for them.
And then comes part to of the AIsoftware market, floor plan,
automation landscape.
And then it has the objectivesand then it has all the
different tools, very similar toway cha PT did it, divided to
the different segments and whatare the benefits and the pros
(24:22):
and cons of each and every oneof the software and the pricing
and so on.
Very detailed report.
It also created a table in theend, like I requested, which G
PT put in the beginning.
But in this particular case, thetable is full and complete.
So all of it is in the table.
And it included all thedifferent companies and tools
that it found necessary.
The other very cool thing thatexists if you use Gemini, is
that almost every output that ithas, you have a one click export
(24:42):
to the relevant Google platform.
In this particular case, it'sGoogle Sheets, so if I click on
export to Google Sheets, it willbuild a Google Sheet file for me
with this table in it, and Idon't have to copy paste,
realign and so on.
It's the same thing with openingreports as documents and so on.
So it's a very helpful, usefultool, uh, for Gemini.
So in general, and again, I usedeep research several times a
(25:03):
week, every single week formultiple topics.
Gemini is number one right nowwhen it comes to doing deep
research, and ChatGPT is good.
But it's definitely in a secondplace, and this was just one
example.
It's a very clear example of whyGemini is better overall.
By the way, Gemini also foundand researched more tools than
ChatGPT.
ChatGPT researched 18 differenttools that it reviewed, and
(25:23):
Gemini did 23 different toolsthat it reviewed.
Part of it might be because itdid the first part of the
research as I requested.
The next topic that is a lot offun is creating images.
So if you remember X number ofmonths ago, I think it was May
Cha, GPT came out with itsamazing image generation
capabilities.
The world was going crazy withcreating Ghibli images of
themselves, their family, theircoworkers, and so on.
(25:44):
And then two weeks ago, Geminicame out with Nano Banana, which
is also known as Gemini 2.5flash image, which creates
amazing images and knows how tokeep consistency and so on.
And I wanted to try anexperiment that I did just after
Chet PT came out and tried toreplicate it with Gemini.
And so the experiment was totake a relatively low resolution
image of a product of theinternet and create an ad with
(26:06):
it.
In this particular case, it is aNeutrogena sunscreen 70 SPF.
And again, those of you'rewatching this, this is an image
of the internet.
It has a white background and arelatively low quality image of
the sunscreen in its tubewithout any kind of background,
and what I did is I first askedit, please provide a line by
line text of what is written onthis product.
(26:26):
This helps you afterwards getconsistent results when it's
recreating it.
So it gave me the exact thingthat was on it, and then I said,
now I'd like you to create acloseup professional product,
photography of a female handholding the product.
The background is the ocean andthe beach, but it is blurry as
the focus is on the hand.
And the product.
Nikon Z seven 50 millimeter lensaperture 2.8.
And then the product should say,and then I pasted what he told
(26:48):
me that's written on the productand it's created an amazing
image.
It looks exactly like theprompt.
Suggest professional photographyoff the beach with the right
lighting, with a female handholding the product.
It looks perfect.
But if we zoom in.
If we zoom in on the text, yousee the text says, dermatologist
(27:09):
recommended brand.
It says Beach defense, water andsun, sunscreen, lotion, broad
spectrum, all the stuff that itsays.
But here on the bottom, whereit's where the original one
said, water resistance, this onesays, which is not exactly the
same, and instead of 80 minutesit says zero eight minutes.
So it missed one small aspect.
(27:29):
Again, if you don't zoom intothe fine print, you will not
know the difference.
But then I went on and I askedit to change the SPF 70 to 50
just for fun.
And it did that, and it actuallycaptured both places that it's
written.
Then I went ahead and asked itto change, the SPF from 70 to
50, just to see if you will doit.
So you can see in the originalimage it says broad spectrum SPF
70, and then it says 70 in a bigfont beneath that.
(27:52):
And then in the new image itdoes says 50, but where it says
broad spectrum, it doesn't havea number anymore.
Just as broad spectrum, SPF.
And then I ask it, please redothe image and change the girl's
nail polish to the USA flagpattern.
And it did amazing.
It actually did that and it keptthe 50 SPF, but now you can see
that the broad spectrum says SPFthree, and you can see that the
(28:13):
water resistant is now allgibberish of what it said
before.
It doesn't even say waterresistant.
It says something else and itsays 50 minutes.
So things are starting to driftaway from the original image,
but the quality is still verygood.
And if you don't dive into thevery close details, it looks.
Perfect, and it looksprofessional.
Then I went further and said,extend the image on the bottom
to make it a nine by 16 imageand write in a fun summary font.
(28:36):
4th of July sale, Neutrogena,bogo, and it did exactly what I
asked.
It extended the picture to aportrait format, and then it
added in two different fonts,the sale.
Then the bottom font wasn'tclear enough, so I asked it to
change that to the same font,and I got to something that I
can definitely use as aprofessional ad.
The only problem, as we've seenbefore is if we dive in, you
will see that now even the stuffthat was written on the bottom
(28:58):
is a little bit messed up and itinvented new words, that are
supposed to be written there andit invented the amount of time
that it protects you from that.
So again, from as an ad, itstill looks great.
From the very fine details, it'sstill not perfect.
I try to mimic the same exactprocess with Gemini 2.5 flash
image, also known as nanobanana.
I first of all started the sameexact way I started in ChatGPT.
(29:20):
I pasted the same exact imageand I asked it to create the
line by line text.
And then I used the same exactfont that says, I would like you
to create the closeupprofessional product photography
of a female hand holding theproduct.
And they gave me something thatlooks like a female hand holding
the product, but it wassomething completely different
than what I had before.
It still looks like a tube ofsunscreen lotion, but it doesn't
look anything like the originalone.
(29:41):
And then it caught my attentionthat I forgot to turn on the
image generation tool, whichmeans it's still using, in this
particular case, it's oldImagine four engine instead of
the image generation capability,which is hidden down in the
tools.
So in the prompt menu where youwrite your prompt, there are
tools and if you click on that,there's an option to create
images and there's a littlebanana.
(30:01):
Icon next to it, and that's whatyou wanna select.
So then I started from thebeginning with the correct tool,
and you can see here that itsays image and it has the banana
next to it.
And I pasted the same stuff.
And this time it created aperfect image.
It actually has all the textcorrect.
Including the water resistant inthe 80 minutes.
The quality of the image from apixelation perspective is
slightly lesser than what it wason the ChatGPT side.
(30:25):
But, uh, both of these can thenuse different kind of upscales
to solve the problem.
So I don't see that as a bigissue.
Uh, but overall a great image.
Definitely as good as the oneChachi created as the first
step.
It even did something very cooland it added little water
droplets on the actual tube tomake it look as if somebody's at
the beach actually holding it,which I found really cool.
It makes it look a lot morerealistic and there's shading on
(30:47):
it.
It's just perfect from an imageperspective.
Then I went the same exact stepthat I did with ChatGPT.
I said, I would like you torecreate the image with SPF of
50 instead of 70.
And in this particular case, itactually changed the 50 SPF part
on the fine print where it saysbroad spectrum SPF.
It missed that and it kept thatas 70.
So I would say that in thisparticular case, it is equal to
what chat GPT did because ChaGPT didn't write the SPF at all
(31:09):
on the five print, and itactually did, Gemini did better
by preserving the font of the 50over 70 when ChatGPT changed it
a little bit.
Then I followed the sameprocess.
Please redo the image and changethe girl's name polish to the US
flag.
And he did that and it kepteverything else intact.
And in this particular case, itis also maintaining everything
that was written on thepackaging.
(31:29):
So you can see that thedermatologist recommended, uh,
brand is still okay, and you seethat the fine print here still
says.
Water resistant 80 minutes, sobetter than ChatGPT did on
keeping the actual product, butthen is where it starts failing
and I asked it to extend thebottom of the image to make it a
nine by 16 image and write thefont on the bottom and it just
doesn't change the aspect ratio.
(31:49):
That's something I found withnano banana that it just doesn't
do.
Does not agree to change theaspect ratio of images.
So it created the font.
It was very hard to read becauseit created the colors of the
flag within inside the text,which is actually very cool, but
it makes it hard to actuallyread, ask it to remove it.
And in the second attempt itactually did.
So the final output is verysimilar to what we had with GPT,
(32:10):
but with several different, butwith several major differences.
One is that it did not extendthe bottom of the image, so the
text actually hides a little bitof the actual product, which is
not what I wanted, and it is notwhat I had with ChatGPT two is
in the very final image.
You can see that there's a lotmore pixelation around the text
on the actual tube and it's notas clear, and it's starting to
(32:32):
have these weird artifacts so itdrifts further and further away
from the original image as youkeep redoing the product, that's
something you just should beaware of.
So what is the verdict here?
I would say this is a tiebecause ChatGPT did a better job
in the aspect ratio and placingthe font in the correct place,
and Gemini did a better job inmaintaining the exact text and
(32:52):
the accuracy on the product, atleast in the first few steps.
But then I said, okay, thisdeviation from the.
Original text in Gemini onlyhappened after multiple steps of
making different corrections inthe image.
What if I try to one shot it?
What if I allow it to create theimage with a more detailed
prompt, with just one step.
So in this particular case, Ibasically took all the prompts
(33:13):
and combined them together.
I gave it the same originalimage and I said, make it nine
by 16.
The background needs to be theocean blurry, blah, blah, blah.
Changed the SPF from 50 to 70.
The girl's name Polish needs tobe the pattern of the US flag,
and then exactly what it needsto say, and it actually created
it with one go other than theaspect ratio.
So I got.
Something very similar to thefinal ad with keeping the
(33:34):
accuracy of the text on theproduct.
So this is great because I don'thave to then fix it in Photoshop
and so on, but it did not extendthe aspect ratio, so I was able
to find a solution to part ofthe problem, but not for the
entire problem.
The expect ratio is justsomething that Gemini will not
do at this point.
And then the final thing that Itested was how good are these
(33:55):
tools in actually reading myinformation across multiple of
my tools that it is connectedto?
So Gemini is a Google product.
It has access to GoogleCalendar, Google Drive, and
Gmail.
And ChatGPT, if you connect itthrough its connectors, also has
access to your calendar and yourGmail account.
So I have lots and lots ofbusiness travel coming in.
The next two months, I'm doingmultiple in-person workshops to
(34:18):
companies.
I'm speaking at severaldifferent conventions and
events, and so I'm going to betraveling a lot.
So what I did is I started inChatGPT and I said I would like
to check my calendar and myemails and all the proposals
that I've sent in the past sixmonths, and look for any
engagements between now and theend of November that requires me
to travel.
I would like you to create atable with the following
(34:38):
columns, starting date, endingdate, destination activity, and
details, and then I explain whatdo I want in the details.
And the way you do this in chat,GPT is when you click on the
plus button, in the promptsection of chat GPT, there is a
button that says Use connectors.
And if you click on that.
There's a dropdown menu whereyou can connect different
connectors like Canva andDropbox and so on, and there's a
(34:59):
lot of other connectors that youcan go and choose from.
But what I connected is Gmailand Google Calendar, and I asked
you to find, uh, information andthen it only found two pieces of
information for my travel.
It only found two travel thingsthat I need to do in the next
two months, which I wish thatwas the case.
It's more like, seven or eight,but it did a very good job on
those too.
So it gave me the dates, uh, thestarting date and the ending
(35:21):
date, and the location and theactivity, and a very well
detailed plan on exactly what amI doing and what, who am I
meeting and where I'm supposedto go.
And if there was hotelinformation or stuff like that,
all it is is in there, includingthe booking reference of the,
uh, hotel that I'm staying inand so on.
So.
Not great, but not horribleeither.
It found a few of the events,uh, and gave me a detailed
(35:42):
information about each of theevents when I did the same thing
for Gemini.
Which in theory should do abetter job because it has native
access to all my stuff.
It found three different travelout of like the seven or eight
that it was supposed to find,and it provided a similar level
of information.
As far as the details, itdoesn't look.
As nice and clean as the viewinside of Chat pt.
(36:04):
But it definitely has the samequality and the same level of
details as far as all the thingsthat I need and the information
in it.
And again, it has the button toexport to Google Sheets with one
click, which is helpful.
So from my perspective on thisparticular aspect, they both.
Failed because they both did notprovide even half of the travel
that he was supposed to find.
(36:24):
And the information isdefinitely there across the
different tools.
To be fair, Gemini has access tomore information, but it because
it can also see my Google Driveand it can find all the
proposals that I've writtenthere as well, not just as
attachments in Gmail, but theinformation should be available
on both platforms.
So both tools had the option tobe successful in this and they
were not.
(36:44):
So what is the bottom line ofall of this?
We tried multiple different usecases.
Some are more basic, some aremore advanced capabilities, and
we're trying to compare the twodifferent platforms and in some
aspects also clawed as well.
And the reality is that each andevery one of them has pros and
cons.
Some things Gemini did better,like the dashboards and coding,
(37:05):
uh, tools have actually workedbetter on Gemini than they
worked on chat GPT.
Also, as far as findinginformation, it found a little
more information in my GoogleDrive and so on.
It definitely did a much betterjob on deep research.
ChatGPT, on the other hand, dida better job with custom GPT
compared to gems, actually amuch better job across all the
different aspects from creatingthem in a regular conversation
(37:27):
through running more complexones and getting consistent
results through the ability touse code.
So pros and cons on each side,there is no one clear.
Winner because each and everyone of them has different
capabilities that it's actuallygood at.
And the same is actually true ifyou expand that to Claude and
Grok, which I also use regularlyfor different aspects.
So what does that mean to youand your business?
(37:48):
Well, if you have to pick onetool, I would say pick the tool
that is best connected to youruniverse.
So if you are a Google user,over time, you will gain more
and more benefits by usingGemini.
And yes, I know cha.
GPT has connectors and Claudehas MCP and connectors and so
on.
But we have to assume thatGoogle will make better,
seamless integrations of theiruniverse into their world.
(38:10):
And in general, Gemini is a verypowerful and capable platform.
The only thing, it wasdefinitely not as good at.
Is gems.
And so that's the only thingthat you're giving up on to an
extent.
You can still create gems, theycan still run automations.
I built a lot, uh, for myselfand I definitely build a few
for, different clients, but it,if I had a choice, I will build
those in custom GPT because itjust runs better and provides,
(38:33):
uh, more capabilities.
If you're not in the Googleuniverse, you can choose
whatever tool you want.
Then ChatGPT is definitely anoption.
However, if you are willing toprovide different people access
to different tools, depending onthe use cases that they have,
and you heard me talk about thisin this podcast a thousand
times.
It's all about the use case.
Right.
You don't start by selecting atool.
(38:53):
You start by finding the usecases in which AI can provide
you significant value, and youcan define value however you
want.
It's all okay.
This could be time, this couldbe money, this could be
happiness of your employees.
This could be anything youdefine as valuable to you and
your business, but.
Find something that providesvalue to your business that AI
can help you solve.
And then find the best tool thatdoes that in the best way.
(39:15):
And this might be doing sub deepresearch like I did for the
architects, uh, and interiordesigners, right?
You may run that and you mayfind custom specific tools for
specific tasks.
Or you can just test yourself,the main platforms like Cha pti
and Gemini and Rock and Claudeand so on, and some of the
Chinese models and open sourcemodels and see which one does
that particular thing betterthan the other tools.
(39:35):
And then in two to three months,you can test it again if new
models come out and you wannamake sure you're not staying
behind.
And as long as you understandthe concepts and you understand
the risks, and you understandwhat is working and not is not
working and what you need totest for in these tools, then
switching the tools should bevery easy.
You don't have to be married toany of these tools for a very
(39:55):
long time.
You can switch if you understandthe concepts and you've built
the right processes around it.
That's it for today.
I hope you found this, valuable.
This is just a currentcomparison between ChatGPT and
Gemini, which are the twoleading AI platforms right now
in means of usage, and you nowkind of know.
What each and every one of themdoes a little better.
Your use cases might bedifferent, so you have to test
(40:16):
it on your own, but this givesyou an idea, maybe on how you
can test it.
So I hope you found thisvaluable.
If you have, please give us athumbs up in a five star rating
on your favorite podcastingplatform, whether it's Apple
Podcast or Spotify, which iswhere most of you listen to the
show, share it with otherpeople.
This is how you can help otherpeople learn how to use AI
properly, is by sharing thispodcast.
(40:36):
Literally open your phone rightnow unless you're driving.
Click on the share button andtype the names of a few people
that you know, either onWhatsApp or Instagram or
LinkedIn or text messages,however you wanna share it with
people or email and just shareit with a few people that can
benefit from this podcast aswell.
I will be really grateful if youdo that, and until next time,
have an awesome rest of yourweek.
Hello and welcome to the Leveraging AI
Podcast, the podcast that sharespractical and ethical ways to
leverage AI to improveefficiency, grow your business,
and advance your career.
This is Isar Metis, your host,and we've got a very interesting
episode for you today.
The reason it's interesting isit is based on my own personal
experience that is tied to aworkshop I have delivered two
weeks ago.
(00:20):
So two weeks ago, I was invitedto provide a several hour
workshop at a very large reallyknown tech company in San
Francisco.
The company that invited me usesGoogle as their office
infrastructure enhanced therequest was to focus most of the
workshop on Gemini as well asother Google AI tools, and
that's what I did.
(00:40):
But to do that, I had to take alot of the workshops that I've
been teaching to many othercompanies and customize it and
test it on the Geminienvironment, which gives me a
very unique insight to how it iscompared to other tools,
especially ChatGPT.
Now I've learned a lot in theprocess and I thought it's gonna
be highly valuable to you too,to learn from my experience from
(01:02):
what I've learned from trying toconvert use cases from Claude
and ChatGPT, mostly ChatGPT intothe Gemini universe.
And so this episode is not goingto be structured in a very
specific flow because these areunique specific use cases.
We're just gonna go use case byuse case and compare what the
results were and what were thepros and cons of each and every
one of these platforms.
There is a winner, which we'regonna get to in the end, but
(01:25):
let's get started with thecomparison.
(02:10):
The first thing that I want totalk about is a concept that is
critical for the success ofanything you do with ai, and
that is the concept of context.
These tools lack context, right?
They don't know you, they don'tknow your company, they don't
know your industry, they don'tknow your unique expertise, and
because of that, a lot of theanswers you're going to get are
generic.
So the more the tools.
(02:31):
Know about you, the moreinformation you provide them in
order to learn about youruniverse, the more specific, the
more accurate, the more relevantanswers that you're going to
get.
That is true for all thesetools, which means that if you
want to get accurate answers,you need to write very long
prompts every single time,unless you can teach the model
stuff about you that it willremember for the long run.
And that is a very bigdifference right now between
(02:53):
ChatGPT and.
All the other models, but inthis particular case, Gemini as
well.
So let's look at what I'mreferring to.
So what I did here is Ibasically asked ChatGPT, what do
you know about me?
A very simple andstraightforward question.
You can do that yourself to testit yourself.
You can try it with differentmodels.
But let's see what ChatGPT knowsabout me.
And this is just a questionthat's, there's nothing else in
(03:13):
this prompt.
And it says, here's a structuredoverview of what I know about
you based on our pastconversations and context.
You've shared background, formerF 16 pilot in the Israeli Air
Force.
Later on, instructor in the AirForce Academy transitioned to
business and technology with 15plus years as a CEO experience
in startups.
People post and largecorporations last minute travel,
hotel beds group,, moved to theUS about 20 years ago,
(03:36):
professional journey built andscaled multiple companies,
including taking one public andselling another.
Founded Multiplai in 2023,focused on AI education and
consulting.
Provides, and so on and soforth.
Then it has, content education,which are the things that I'm
focusing on right now.
It talks about this podcast, theLeveraging AI podcast, about our
newsletter, about the hangoutswe do every single Friday, and
about my collaboration with theCrummer School of Business,
(03:59):
which is the business school Idid my executive MBA at, and now
I've been collaborating withthem for the past year and a
half in delivering AI educationto their audience.
And then our other initiativethat I'm involved in, market and
audience, where it is, who amI'm catering to, personal notes
about myself, about my family,what I'm passionate about, what
my family does, and so on.
And some key takeaways.
All of that, it knows about me,which means when I'm asking it
(04:22):
anything, it uses that referenceinformation in order to provide
more relevant answers to me.
I've done the same exact thingwith Gemini and asked Gemini
what it knows about me.
So same question.
What do you know about me?
And the answer is, based on ourcurrent interaction, here's what
I know, just as important, whatI do not know.
What I know, our currentconversation, and then it says
what we talked about in thatparticular conversation, time
(04:43):
and location.
Based on the informationprovided to my system.
I know it is Thursday, August7th at 10:49 AM technical
information, very little.
And what it doesn't know, itdoesn't know my name, it doesn't
know my age.
It doesn't know where I live.
It doesn't know my emailaddress.
It doesn't know my browsinghistory.
It's a lot of other stuff itdoesn't know.
So basically it doesn't knowanything about me, which is good
(05:03):
from maybe a securityperspective.
It's bad from a contextperspective to.
Any conversations I'm going tohave.
So I actually take the time fora very long time now, investing
in teaching Chachi piti aboutme, about my business, about my
tone, about my customers, aboutthe proposals that I write,
about the services that Ideliver.
It knows a lot more than itactually shared in that one
summary, and because of that canprovide me much better
(05:26):
information.
The other thing that I use forcontext in chat g PT that does
not exist in the Gemini universeis I use projects.
Projects are on the left side,right under,, the custom GPT
segment just above your.
Chats.
And in there you can createthese folders.
And in these folders you willhave different conversations,
but each conversation can haveits own unique context universe.
(05:48):
So you can see in thisparticular case, this is a
sample client project.
It's not a real client, but Ihave one for each and every one
of my clients.
In there you will see that Ihave different kinds of
information.
I have proposals, for thatparticular client that I've sent
this client I have.
Different files that I'vecreated.
So the culture of the companybased on my experience and based
on their website, what proposalsI've sent them.
(06:08):
So you can see what that is.
The consulting agreement I havewith them, the people in the
company that I engage with andtheir profiles from LinkedIn,
the workshop proposal that Isent them, the about US PDF from
their website.
So more context about thatparticular client.
So then when I haveconversations in this folder, it
has a lot more context aboutthis particular which again is
not something you can do inGemini.
So that's on context, which isdefinitely a huge win on the
(06:33):
chat GPT side.
So let's move on to the next usecase.
The next use case is usingcustom GPTs versus using gems.
So the concept is very similar.
I'll explain in one sentencewhat it is.
So custom gpt and gems allow youto write instructions that you
wanna repeat time and timeagain.
So every time you have arepetitive thing you want to do
with ai, whether it's writing aproposal, analyzing a specific
(06:55):
document, comparing two Excelfiles, uh, creating content, et
cetera, et cetera, every one ofthose can become either a custom
GPT on the chat, GPT Universe orGEMS on the Gemini universe.
I recorded an entire episodeabout it.
And if you want to dive into thetopic of custom gpt, you can go
back to episode 1 75.
It is called Stop Wasting Time,automate repetitive Tasks with
(07:17):
Custom gpt so you can go backand listen more about that
topic.
But there is a similar conceptthat is more or less the same in
the Gemini universe called gems,where you can provide
information and so on.
So the way I create and the wayI teach how to create both
custom GPTs and gems is not bywriting the instructions on your
own, but rather by having aconversation with cha GPT or in
(07:37):
Gemini and then asking ChatGPTat the end of the conversation.
Once you explain to iteverything that you want, once
you ask it questions, once youwork inside the conversation.
On the actual output and youachieve the correct output, then
I would go and say, I would likeyou to now turn everything that
we've done, including the finalsuccessful outcome.
So ignore the stuff that didn'twork into highly detailed
instructions for a custom GPT ora gem, because it creates better
(08:00):
instructions than I can,especially after we've done the
entire exercise and we'regetting to the outcome that we
want.
I've done this in chat GPTmultiple times, and then I've
tried it in Gemini and itdoesn't work as good.
So let's look at an example fromeach and every one of those.
So in this particular case, thisis from a presentation I've done
for the Israeli parliament andI've started the prompt with you
(08:21):
are an expert data researcher inthe Israeli Parliament tset.
We are planning to use a customGPT to help us in the research
and summarization of data.
We would like to build a team ofseparate personalities that will
participate in this process, etcetera, et cetera, et cetera.
I'm writing in this prompt, andthen we're going back and forth
and it gives me suggestions ofwhat these people can be and
what they will do and what aregoing to be the responsibilities
and the roles of each and everyone of them, and what's gonna be
(08:42):
the research process and what'sgonna be the focus on different
things.
And then I ask it for a list ofresponsibilities for each and
every one of these people.
So it went ahead and did that.
So now I have a long list ofresponsibilities.
And then in the end, after goingthrough the entire process, I
said, please create a detailedinstructions for our custom GPT
that will create all thesepersonas.
The GPT should receive an inputthat would be a topic for the
Nset team would like toresearch.
(09:04):
The team should first discussthe topic debate.
And agree on the way forward, etcetera, et cetera, et cetera.
And he wrote these amazinginstructions.
So you can see there's systemdescription.
This is the first thing that itcreated.
Then it created a team overview.
It gave names to each and everyone of the people.
What's their role, what's theirstyle?
So they can have differentpersonalities that will support
the team.
Then a step-by-step workflow.
(09:24):
Step one, topic intake and groupdiscussion.
What are the actions that theyshould follow?
Step two, data collection.
Step three.
Domain analysis, what eachpeople, which, and it even has a
breakdown of who's leading whataspect of the process based on
the personas and the roles andthe actions that they need to
take and the responses that theyneed to provide.
Step four, political and publiclens.
Step five, quantitativeanalysis.
(09:46):
Step six, synthesis anddrafting.
Step seven, risk and ethic.
Review, and then who is the leadon that and the actions for all
of that Step eight final review.
And then there's a communicationformat on how they should
communicate with one another.
And then what is the outputstructure of the report that
they should generate andadditional capabilities.
So, this is an incredible set ofinstructions that, again, there
is no way I could have writtenthis as accurate and as
(10:07):
detailed.
And then I did a similar processin Gemini asking it to create
instructions for a specificbusiness process, not exactly
the same.
And when I did that, it actuallygave me.
Instructions that were okay.
They weren't horrible, but theydefinitely weren't great.
And it was weird because it waskind of like relating it in
third person, and I'll explainwhat I mean.
So my prompt was, I would likeyou to create instructions for a
gem that I can run inside ofGoogle Slides that will ask the
(10:30):
user this set of questions andtopic, blah, blah, blah, blah,
blah.
And then it created a set ofinstructions that should be able
to work, but it invented thingsthat just don't exist.
So as an example, it said thisGem will open in the sidebar in
Google Slides.
It will guide the user through aseries of five steps, asking
questions and so on and soforth.
But the reality is it cannotopen on the sidebar.
So I wrote back, I said, I don'tbelieve the sidebar can have a
(10:53):
next button.
I believe it is a simple chatinterface.
Am I wrong?
but then it said you are rightto question the specifics as the
interface isn't free foreigncanvas.
However, it's more powerful thana simple interface, and you can
absolutely have a next button,which by the way is not true.
At least I wasn't able to see ittrue, but it gave me
instructions that just don'twork.
(11:13):
So, instead of just giving meinstructions for a gem, it's
referring to what needs to bethe instructions in the gem,
which is not horrible.
And in many cases it actuallyworked okay, but it's definitely
not as good as the solution thatI got from chat GPT.
Then I tested something else.
I tested a very complexmulti-step custom GPT that I've
created a while back for aclient, and I was trying to
(11:35):
recreate it with the same exactinstructions on Gemini.
So let's look at that.
So before we dive into this,let's talk about what this
multi-step process does, andit's actually gonna blow your
mind just by itself.
Because what it does is it takesa inventory file that is
completely fake.
I made it up.
All the numbers are made up, butit's an inventory file from
multiple sources showinginventory of different SKUs in
(11:57):
different locations for thecompany.
And it is using it to create asales brief for the salespeople
to explain to them whatinventory we have, what
inventory was were out of, inwhich regions of the world and
so on.
So this is something that manycompanies struggle with, right?
So the ability to takeinformation and compare the
inventory from yesterday toinventory from today, the
(12:19):
inventory from last week to theinventory from this week, and
based on that update salesinformation that they can be
sent to salespeople.
This is a process that manycompanies do completely manually
today, and I've developed acustom GPT that does the entire
process, but it's a five stepprocess, so all I have to do
right now is upload the oldinventory file, the new
inventory file, and the oldsales file to the custom GPT,
(12:41):
and then it follows all thedifferent steps.
It does step one, step two, stepthree, step four compares
different things, uh, where itis, what are the sizes, what are
the SKUs, it's the same packagetype, uh, and so on and so forth
to verify all that information.
And then when it's done withthat step, it goes to step
three, and then it removes allthe roles that.
Do not have the relevant rows inthe other file anymore.
So basically we have it in thesales files, but they don't
(13:02):
exist in the inventory fileanymore.
So it compares three differentfiles back and forth.
It updates all the quantity andthe numbers and the SKUs and
everything to provide me a newupdated file, and then it
provides it to me.
As a CSV file that I candownload to either upload to
whatever CRM system that I'mrunning or to email to my team.
(13:23):
Again, the CSV file can then beconverted to anything that I
want.
So this is an extremely powerfulmulti-step process that I have
tested consistently multipletimes and provides.
Accurate and amazing resultsevery time that I run it.
You can imagine similarprocesses across multiple
aspects of businesses where youhave multiple data sources that
you need to align, compare, andprovide some kind of an output
(13:44):
based on all of them, and itworks flawlessly inside of
ChatGPT.
However, when I try to run it onGemini with the same exact
instructions it literally getsstuck.
It doesn't do anything.
It says, of course, I willprocess the files, uh, according
to the steps provided.
Please provide a moment while Iperform the data comparison,
filtering, merging, andupdating.
I'll provide you with a finalupdated file for download
(14:06):
shortly.
That sounds very promising.
Only nothing happens, and so.
I tried several different ways.
I changed the instructions alittle bit.
Again, the instructions werecopied and pasted, uh, and yet
it does not work.
The inputs are the same exactfiles that I gave ChatGPT.
So in this particular case,again, a huge win for ChatGPT
compared to Gemini when it comesto actually running a multi-step
complex process, with the sameexec instructions on both
(14:28):
platforms.
Now in addition, one more thingthat you can do in custom GPT is
that you cannot do.
In gems is add code to them toconnect them to more advanced
capabilities and connect them toAPI of third party tools,
connect them to MCP servers, etcetera.
So if we look at this particularGPT is connected down here to
the API of data for SEO.
(14:50):
Which is a company that providesdata for SEO, like the name
suggests, but that means that Ican ask questions about keywords
and their availability andcompetitors and their density
and so on across any domain thatI want, within this custom GPT,
and get information aboutmyself, my competitors, and so
on, and Build my future blogpost based on actual real SEO
(15:10):
data right here within chat GPT,without having to go to other
tools and bring information andso on.
This is a very powerfulcapability.
This is called Actions, and it'savailable on the bottom of
custom gpt.
Uh, if you, if I'll open this,you'll see this is just code
that connects it to their MCPserver.
I didn't write the code.
That's a code that they provideto anybody.
So you can go and grab the codeand attach it in here, and you
(15:32):
can then have this kind ofcapability.
The ability to out code into agem just doesn't exist.
So that's another benefit ofcustom GPEs versus Gemini Gems.
The next topic that we're goingto talk about is dashboards.
So I actually think that thebest tool to create dashboards
right now is Claude.
Every time I need to createdashboards, I actually go to
Claude, and Claude does anamazing job in creating visually
(15:53):
pleasing and also very effectivedashboards.
However, I wanted to test Geminion that topic, but before I even
wanted to test it, somethingvery interesting happened.
I was actually working on afinancial analysis, and so the
financial analysis, just as anexample, took, information of a
publicly traded company.
In this particular case, Ibelieve it was Salesforce, and
then it grabbed the informationfrom the internet and it gave me
(16:14):
their information, and I askedit to create a report about it.
So this was a great example onhow you can take raw data off
the internet and create graphslike you can see here, like
fiscal results per quarter, orthen compare that year over year
and just get information just byasking simple questions or even
creating complete reports likeyou can see here inside of
Gemini.
So it created this detailedreport on the financial success
(16:36):
of Salesforce.
In the past X number of quartersbased on the prompting that I
provided in the beginning.
I could have done this obviouslyon any company in the world.
So a very solid solution fromGemini in this particular case,
however, it then suggested onits own to create an interactive
dashboard, and all I have to dois say, yes, I'm interested in
that.
And it created this amazingdashboard from the data that it
has collected based on myprompt.
(16:57):
So you can see it saysSalesforce financial dashboard.
It has key matrix, revenueanalysis, profitability and
efficiency.
Each and every one of them is alink that if I click, it will
scroll down to that particularsection.
Then there's the key findings inbig, bold statuses, and if I
click on each and every one ofthem, it opens another segment
with some more details aboutthis particular topic.
So you can see if I clickdifferent things, it goes to
that.
And then it has revenue andseasonal patterns.
(17:20):
And then it has year over yearcomparison and it created.
All of this on its own.
Based on the conversation that Ihad with it, I did not define
any of the components in thisdashboard.
It just asked me if I would liketo create one.
And when I said yes, it createdthis, which is really, really
cool.
So from that perspective, hugewin.
On the Gemini side, both takingthe initiative and suggesting
it, but also without me definingwhat exactly I want to see in
(17:42):
the dashboard.
It was able to create thisamazing dashboard, but then I
did something else and I wentthe next step.
I said, I have these amazingdashboards that I've created in
Claude.
Can I just replicate them inChatGPT and in Gemini and see
which one does a better job?
So I started with ChatGPT.
And all I did is I provided itthe same exact information that
I provided to Claude, and Iprovided it a screenshot of the
(18:05):
dashboard, so it got the inputfile that I'm going to upload,
basically the same exact filethat I'm using on the cloud
side, and you've got ascreenshot.
Of the entire page.
So a full page screenshot, notjust the top of it, of the
dashboard and all the componentsthat are in it.
And then I ask it and I said,you're an expert in dashboard
application creation.
I'm going to provide you thefollowing, basically what I
explained to you.
And I ask it to create thedashboard.
(18:26):
And it did, and it never works.
So when I click preview.
The first step is correct.
It actually asked me to uploadthe file and start the analysis,
but every time I do this,there's some kind of a bug and
the actual dashboard doesn't rundespite the fact it's the same
exact file.
I, I gave it as the sample datain order to build the initial
dashboard.
So that was my experience withChatGPT.
I tried it several times and Igot the same issue every single
(18:48):
time.
Gemini, on the other hand, gotthe same exact instructions.
I literally copied and pastedthe instruction.
I also copied and pasted, uh,the same screenshot from the
Claude Dashboard, and instead ofcourse, I will create the
dashboard and it went ahead andcreated an amazing dashboard.
It's working, it's fullyfunctional.
I can upload the file, so let'splay the game.
I will click here and I will goand find the file.
(19:08):
Here we go.
Click open.
Click start analysis and boom, Ihave a working dashboard and you
can see that it has, uh, themonthly sales trend and it has
year over year growth and it hasthe top 20 clients and their
trend in the past month comparedto the trailing 12 months.
And it has the distributionbetween the top leading clients
and there's the filters on top.
And I didn't have to explain anyof this.
I literally just showed it ascreenshot of the other
(19:29):
dashboard.
So if I click now, uh, channelone.
You can see that it changes thenumbers.
I can click and say, oh, I justwanna look at Q1, uh, for that.
So it will change that as well.
And you can see that it changesonly to Q1 across the different
years, uh, and so on and soforth.
So all of this is working.
I can go to clients and I canselect the different clients
across different regions andeverything.
We'll move and updateaccordingly across all the
(19:50):
different aspects of thedashboard, really fantastic and
without almost any effort.
So on this particular example,definitely a huge win for
Gemini, for creating the rightcode, for understanding the
need, and for creating a veryuseful and functioning
dashboard.
With very few steps.
Uh, the only thing I had toactually correct was the graph
on the bottom that had all thecompanies, uh, if there are no
(20:10):
filters are applied.
So if I do this, it had like agazillion companies, which made
no sense.
It was too crowded and all I hadto do is ask it to only include
the top 20 companies, uh, in thegraph and everything else just
worked out of the box withalmost no instructions.
So big win.
To Gemini on this particularcreation of the dashboard
compared to Chad G.
Pt that despite all my attempts,was never able to make this work
(20:31):
And now our next topic is deepresearch.
Both tools have deep researchcreating capabilities.
Both tools are the two leadingtools from a deep research
perspective right now.
From my perspective and myexperience, and I use both of
them all the time, I also trythe other tools.
Very regularly.
So about about once every otherweek, I will try one of the
other tools or some of the othertools as well just to see how
(20:51):
they're doing.
So whether that's, perplexity orDeep seek or Claude and I will
try their deep se, deep researchas well.
Still, ChatGPT and Gemini have avery big lead when it comes to
deep research, but let's comparethe two.
And so in this particularexample, what I was looking for
is I was looking for AI softwaresolutions that can help
architects, interior designers,smart home integrators and
builders, and other people whowork with floor plans and stuff
(21:13):
like that.
Because I have multiple clientsin that universe and I wanted to
see what it will find.
So I wrote along in detailed.
Prompt.
I copied the same exact promptto both tools and I let it run.
So you can see the prompt isabout a page and a half long
explaining exactly what I need.
The way it works with Gemini, ittells you what it's going to do,
and then you can say that youapprove the plan or not.
If you don't, it will ask you toclarify what else do you want or
(21:33):
what do you want different?
It's a little different on theChatGPT universe.
It always asks you questions so.
After you write the promptagain, it's the same exact
prompt I copied and pasted it.
it said, to get started, couldyou please clarify the
following?
Are you focused only on AIpowered software, or should I
include any advanced software,even if not AI based that
supports automation, rules,logic, blah, blah, blah.
So it asked me all thesequestions that I gotta answer in
(21:55):
order to clarify exactly whatI'm looking for.
Then both started the research.
Uh, usually Gemini takes a lotlonger when he does the process,
but both tools will work on itfor a while.
And then this is the result Igot from ChatGPT.
So you can see it starts with atable that shows the different
tools and what functionalitythey have.
But what you'll see is thatalready here there's issues
(22:16):
where it's not showing, uh, someof the table.
And if you scroll down, thenyou'll see many of these aspects
instead of being a part of thetable as they should, are just
written in text with thevertical line in between them.
So it didn't really create itcorrectly.
So I have some of thefunctionality here compared, but
then all the other ones are inthis weird format that I cannot
(22:36):
do anything with.
So I messed that up.
But overall, then it startedwith generative.
Design tools for architects, uh,and gave me several different
options.
Then it gave me, uh, AI toolsfor interior designer and space
planning, and then it gave me afew options here, uh, and so on
and so forth.
So overall, a detailed, helpfulreport that is aligned with what
I requested or kind of alignedwith what I requested, and I
(22:58):
explained why.
What I mean by kind of aligned,if you go back to the prompt,
what I ask it to start with, Iask it to start with doing a
research.
On what kind of positions, whatkind of roles exist in these
companies, and what are the usecases of each and every one of
those people?
And only then go and start theresearch on the actual tools to
be able to check if they can dothese functions that these
(23:19):
different roles have to do.
So, if we go to the Geminiversion of the deep research,
you will see that it followedthe instructions.
Perfectly said part one, theprofessional landscape of floor
plan utilization, and then ithas objectives, and then it has
chapter one, core design andplanning professions.
And then it has architecturaldesigners and drafters, and it
has what their needs are andwhat they need to do and so on
(23:39):
and so forth.
And it went to interiordesigner, floor plan related
tasks, space planning.
So it literally followed.
In detail exactly what Irequested.
So it started with a much betterunderstanding of what these
people need before it startedthe actual research on the
people.
Now maybe ChatGPT did the samething, but it not expose that in
its report.
So it's very hard for me toknow.
So you can see all of this, likeI'm still scrolling and
(24:00):
scrolling and scrolling.
For those of you who're notwatching this.
All of this, like pages andpages are just analysis of what
are the roles and what keyfunctions they need that these
software will do for them.
And then comes part to of the AIsoftware market, floor plan,
automation landscape.
And then it has the objectivesand then it has all the
different tools, very similar toway cha PT did it, divided to
the different segments and whatare the benefits and the pros
(24:22):
and cons of each and every oneof the software and the pricing
and so on.
Very detailed report.
It also created a table in theend, like I requested, which G
PT put in the beginning.
But in this particular case, thetable is full and complete.
So all of it is in the table.
And it included all thedifferent companies and tools
that it found necessary.
The other very cool thing thatexists if you use Gemini, is
that almost every output that ithas, you have a one click export
(24:42):
to the relevant Google platform.
In this particular case, it'sGoogle Sheets, so if I click on
export to Google Sheets, it willbuild a Google Sheet file for me
with this table in it, and Idon't have to copy paste,
realign and so on.
It's the same thing with openingreports as documents and so on.
So it's a very helpful, usefultool, uh, for Gemini.
So in general, and again, I usedeep research several times a
(25:03):
week, every single week formultiple topics.
Gemini is number one right nowwhen it comes to doing deep
research, and ChatGPT is good.
But it's definitely in a secondplace, and this was just one
example.
It's a very clear example of whyGemini is better overall.
By the way, Gemini also foundand researched more tools than
ChatGPT.
ChatGPT researched 18 differenttools that it reviewed, and
(25:23):
Gemini did 23 different toolsthat it reviewed.
Part of it might be because itdid the first part of the
research as I requested.
The next topic that is a lot offun is creating images.
So if you remember X number ofmonths ago, I think it was May
Cha, GPT came out with itsamazing image generation
capabilities.
The world was going crazy withcreating Ghibli images of
themselves, their family, theircoworkers, and so on.
(25:44):
And then two weeks ago, Geminicame out with Nano Banana, which
is also known as Gemini 2.5flash image, which creates
amazing images and knows how tokeep consistency and so on.
And I wanted to try anexperiment that I did just after
Chet PT came out and tried toreplicate it with Gemini.
And so the experiment was totake a relatively low resolution
image of a product of theinternet and create an ad with
(26:06):
it.
In this particular case, it is aNeutrogena sunscreen 70 SPF.
And again, those of you'rewatching this, this is an image
of the internet.
It has a white background and arelatively low quality image of
the sunscreen in its tubewithout any kind of background,
and what I did is I first askedit, please provide a line by
line text of what is written onthis product.
(26:26):
This helps you afterwards getconsistent results when it's
recreating it.
So it gave me the exact thingthat was on it, and then I said,
now I'd like you to create acloseup professional product,
photography of a female handholding the product.
The background is the ocean andthe beach, but it is blurry as
the focus is on the hand.
And the product.
Nikon Z seven 50 millimeter lensaperture 2.8.
And then the product should say,and then I pasted what he told
(26:48):
me that's written on the productand it's created an amazing
image.
It looks exactly like theprompt.
Suggest professional photographyoff the beach with the right
lighting, with a female handholding the product.
It looks perfect.
But if we zoom in.
If we zoom in on the text, yousee the text says, dermatologist
(27:09):
recommended brand.
It says Beach defense, water andsun, sunscreen, lotion, broad
spectrum, all the stuff that itsays.
But here on the bottom, whereit's where the original one
said, water resistance, this onesays, which is not exactly the
same, and instead of 80 minutesit says zero eight minutes.
So it missed one small aspect.
(27:29):
Again, if you don't zoom intothe fine print, you will not
know the difference.
But then I went on and I askedit to change the SPF 70 to 50
just for fun.
And it did that, and it actuallycaptured both places that it's
written.
Then I went ahead and asked itto change, the SPF from 70 to
50, just to see if you will doit.
So you can see in the originalimage it says broad spectrum SPF
70, and then it says 70 in a bigfont beneath that.
(27:52):
And then in the new image itdoes says 50, but where it says
broad spectrum, it doesn't havea number anymore.
Just as broad spectrum, SPF.
And then I ask it, please redothe image and change the girl's
nail polish to the USA flagpattern.
And it did amazing.
It actually did that and it keptthe 50 SPF, but now you can see
that the broad spectrum says SPFthree, and you can see that the
(28:13):
water resistant is now allgibberish of what it said
before.
It doesn't even say waterresistant.
It says something else and itsays 50 minutes.
So things are starting to driftaway from the original image,
but the quality is still verygood.
And if you don't dive into thevery close details, it looks.
Perfect, and it looksprofessional.
Then I went further and said,extend the image on the bottom
to make it a nine by 16 imageand write in a fun summary font.
(28:36):
4th of July sale, Neutrogena,bogo, and it did exactly what I
asked.
It extended the picture to aportrait format, and then it
added in two different fonts,the sale.
Then the bottom font wasn'tclear enough, so I asked it to
change that to the same font,and I got to something that I
can definitely use as aprofessional ad.
The only problem, as we've seenbefore is if we dive in, you
will see that now even the stuffthat was written on the bottom
(28:58):
is a little bit messed up and itinvented new words, that are
supposed to be written there andit invented the amount of time
that it protects you from that.
So again, from as an ad, itstill looks great.
From the very fine details, it'sstill not perfect.
I try to mimic the same exactprocess with Gemini 2.5 flash
image, also known as nanobanana.
I first of all started the sameexact way I started in ChatGPT.
(29:20):
I pasted the same exact imageand I asked it to create the
line by line text.
And then I used the same exactfont that says, I would like you
to create the closeupprofessional product photography
of a female hand holding theproduct.
And they gave me something thatlooks like a female hand holding
the product, but it wassomething completely different
than what I had before.
It still looks like a tube ofsunscreen lotion, but it doesn't
look anything like the originalone.
(29:41):
And then it caught my attentionthat I forgot to turn on the
image generation tool, whichmeans it's still using, in this
particular case, it's oldImagine four engine instead of
the image generation capability,which is hidden down in the
tools.
So in the prompt menu where youwrite your prompt, there are
tools and if you click on that,there's an option to create
images and there's a littlebanana.
(30:01):
Icon next to it, and that's whatyou wanna select.
So then I started from thebeginning with the correct tool,
and you can see here that itsays image and it has the banana
next to it.
And I pasted the same stuff.
And this time it created aperfect image.
It actually has all the textcorrect.
Including the water resistant inthe 80 minutes.
The quality of the image from apixelation perspective is
slightly lesser than what it wason the ChatGPT side.
(30:25):
But, uh, both of these can thenuse different kind of upscales
to solve the problem.
So I don't see that as a bigissue.
Uh, but overall a great image.
Definitely as good as the oneChachi created as the first
step.
It even did something very cooland it added little water
droplets on the actual tube tomake it look as if somebody's at
the beach actually holding it,which I found really cool.
It makes it look a lot morerealistic and there's shading on
(30:47):
it.
It's just perfect from an imageperspective.
Then I went the same exact stepthat I did with ChatGPT.
I said, I would like you torecreate the image with SPF of
50 instead of 70.
And in this particular case, itactually changed the 50 SPF part
on the fine print where it saysbroad spectrum SPF.
It missed that and it kept thatas 70.
So I would say that in thisparticular case, it is equal to
what chat GPT did because ChaGPT didn't write the SPF at all
(31:09):
on the five print, and itactually did, Gemini did better
by preserving the font of the 50over 70 when ChatGPT changed it
a little bit.
Then I followed the sameprocess.
Please redo the image and changethe girl's name polish to the US
flag.
And he did that and it kepteverything else intact.
And in this particular case, itis also maintaining everything
that was written on thepackaging.
(31:29):
So you can see that thedermatologist recommended, uh,
brand is still okay, and you seethat the fine print here still
says.
Water resistant 80 minutes, sobetter than ChatGPT did on
keeping the actual product, butthen is where it starts failing
and I asked it to extend thebottom of the image to make it a
nine by 16 image and write thefont on the bottom and it just
doesn't change the aspect ratio.
(31:49):
That's something I found withnano banana that it just doesn't
do.
Does not agree to change theaspect ratio of images.
So it created the font.
It was very hard to read becauseit created the colors of the
flag within inside the text,which is actually very cool, but
it makes it hard to actuallyread, ask it to remove it.
And in the second attempt itactually did.
So the final output is verysimilar to what we had with GPT,
(32:10):
but with several different, butwith several major differences.
One is that it did not extendthe bottom of the image, so the
text actually hides a little bitof the actual product, which is
not what I wanted, and it is notwhat I had with ChatGPT two is
in the very final image.
You can see that there's a lotmore pixelation around the text
on the actual tube and it's notas clear, and it's starting to
(32:32):
have these weird artifacts so itdrifts further and further away
from the original image as youkeep redoing the product, that's
something you just should beaware of.
So what is the verdict here?
I would say this is a tiebecause ChatGPT did a better job
in the aspect ratio and placingthe font in the correct place,
and Gemini did a better job inmaintaining the exact text and
(32:52):
the accuracy on the product, atleast in the first few steps.
But then I said, okay, thisdeviation from the.
Original text in Gemini onlyhappened after multiple steps of
making different corrections inthe image.
What if I try to one shot it?
What if I allow it to create theimage with a more detailed
prompt, with just one step.
So in this particular case, Ibasically took all the prompts
(33:13):
and combined them together.
I gave it the same originalimage and I said, make it nine
by 16.
The background needs to be theocean blurry, blah, blah, blah.
Changed the SPF from 50 to 70.
The girl's name Polish needs tobe the pattern of the US flag,
and then exactly what it needsto say, and it actually created
it with one go other than theaspect ratio.
So I got.
Something very similar to thefinal ad with keeping the
(33:34):
accuracy of the text on theproduct.
So this is great because I don'thave to then fix it in Photoshop
and so on, but it did not extendthe aspect ratio, so I was able
to find a solution to part ofthe problem, but not for the
entire problem.
The expect ratio is justsomething that Gemini will not
do at this point.
And then the final thing that Itested was how good are these
(33:55):
tools in actually reading myinformation across multiple of
my tools that it is connectedto?
So Gemini is a Google product.
It has access to GoogleCalendar, Google Drive, and
Gmail.
And ChatGPT, if you connect itthrough its connectors, also has
access to your calendar and yourGmail account.
So I have lots and lots ofbusiness travel coming in.
The next two months, I'm doingmultiple in-person workshops to
(34:18):
companies.
I'm speaking at severaldifferent conventions and
events, and so I'm going to betraveling a lot.
So what I did is I started inChatGPT and I said I would like
to check my calendar and myemails and all the proposals
that I've sent in the past sixmonths, and look for any
engagements between now and theend of November that requires me
to travel.
I would like you to create atable with the following
(34:38):
columns, starting date, endingdate, destination activity, and
details, and then I explain whatdo I want in the details.
And the way you do this in chat,GPT is when you click on the
plus button, in the promptsection of chat GPT, there is a
button that says Use connectors.
And if you click on that.
There's a dropdown menu whereyou can connect different
connectors like Canva andDropbox and so on, and there's a
(34:59):
lot of other connectors that youcan go and choose from.
But what I connected is Gmailand Google Calendar, and I asked
you to find, uh, information andthen it only found two pieces of
information for my travel.
It only found two travel thingsthat I need to do in the next
two months, which I wish thatwas the case.
It's more like, seven or eight,but it did a very good job on
those too.
So it gave me the dates, uh, thestarting date and the ending
(35:21):
date, and the location and theactivity, and a very well
detailed plan on exactly what amI doing and what, who am I
meeting and where I'm supposedto go.
And if there was hotelinformation or stuff like that,
all it is is in there, includingthe booking reference of the,
uh, hotel that I'm staying inand so on.
So.
Not great, but not horribleeither.
It found a few of the events,uh, and gave me a detailed
(35:42):
information about each of theevents when I did the same thing
for Gemini.
Which in theory should do abetter job because it has native
access to all my stuff.
It found three different travelout of like the seven or eight
that it was supposed to find,and it provided a similar level
of information.
As far as the details, itdoesn't look.
As nice and clean as the viewinside of Chat pt.
(36:04):
But it definitely has the samequality and the same level of
details as far as all the thingsthat I need and the information
in it.
And again, it has the button toexport to Google Sheets with one
click, which is helpful.
So from my perspective on thisparticular aspect, they both.
Failed because they both did notprovide even half of the travel
that he was supposed to find.
(36:24):
And the information isdefinitely there across the
different tools.
To be fair, Gemini has access tomore information, but it because
it can also see my Google Driveand it can find all the
proposals that I've writtenthere as well, not just as
attachments in Gmail, but theinformation should be available
on both platforms.
So both tools had the option tobe successful in this and they
were not.
(36:44):
So what is the bottom line ofall of this?
We tried multiple different usecases.
Some are more basic, some aremore advanced capabilities, and
we're trying to compare the twodifferent platforms and in some
aspects also clawed as well.
And the reality is that each andevery one of them has pros and
cons.
Some things Gemini did better,like the dashboards and coding,
(37:05):
uh, tools have actually workedbetter on Gemini than they
worked on chat GPT.
Also, as far as findinginformation, it found a little
more information in my GoogleDrive and so on.
It definitely did a much betterjob on deep research.
ChatGPT, on the other hand, dida better job with custom GPT
compared to gems, actually amuch better job across all the
different aspects from creatingthem in a regular conversation
(37:27):
through running more complexones and getting consistent
results through the ability touse code.
So pros and cons on each side,there is no one clear.
Winner because each and everyone of them has different
capabilities that it's actuallygood at.
And the same is actually true ifyou expand that to Claude and
Grok, which I also use regularlyfor different aspects.
So what does that mean to youand your business?
(37:48):
Well, if you have to pick onetool, I would say pick the tool
that is best connected to youruniverse.
So if you are a Google user,over time, you will gain more
and more benefits by usingGemini.
And yes, I know cha.
GPT has connectors and Claudehas MCP and connectors and so
on.
But we have to assume thatGoogle will make better,
seamless integrations of theiruniverse into their world.
(38:10):
And in general, Gemini is a verypowerful and capable platform.
The only thing, it wasdefinitely not as good at.
Is gems.
And so that's the only thingthat you're giving up on to an
extent.
You can still create gems, theycan still run automations.
I built a lot, uh, for myselfand I definitely build a few
for, different clients, but it,if I had a choice, I will build
those in custom GPT because itjust runs better and provides,
(38:33):
uh, more capabilities.
If you're not in the Googleuniverse, you can choose
whatever tool you want.
Then ChatGPT is definitely anoption.
However, if you are willing toprovide different people access
to different tools, depending onthe use cases that they have,
and you heard me talk about thisin this podcast a thousand
times.
It's all about the use case.
Right.
You don't start by selecting atool.
(38:53):
You start by finding the usecases in which AI can provide
you significant value, and youcan define value however you
want.
It's all okay.
This could be time, this couldbe money, this could be
happiness of your employees.
This could be anything youdefine as valuable to you and
your business, but.
Find something that providesvalue to your business that AI
can help you solve.
And then find the best tool thatdoes that in the best way.
(39:15):
And this might be doing sub deepresearch like I did for the
architects, uh, and interiordesigners, right?
You may run that and you mayfind custom specific tools for
specific tasks.
Or you can just test yourself,the main platforms like Cha pti
and Gemini and Rock and Claudeand so on, and some of the
Chinese models and open sourcemodels and see which one does
that particular thing betterthan the other tools.
(39:35):
And then in two to three months,you can test it again if new
models come out and you wannamake sure you're not staying
behind.
And as long as you understandthe concepts and you understand
the risks, and you understandwhat is working and not is not
working and what you need totest for in these tools, then
switching the tools should bevery easy.
You don't have to be married toany of these tools for a very
(39:55):
long time.
You can switch if you understandthe concepts and you've built
the right processes around it.
That's it for today.
I hope you found this, valuable.
This is just a currentcomparison between ChatGPT and
Gemini, which are the twoleading AI platforms right now
in means of usage, and you nowkind of know.
What each and every one of themdoes a little better.
Your use cases might bedifferent, so you have to test
(40:16):
it on your own, but this givesyou an idea, maybe on how you
can test it.
So I hope you found thisvaluable.
If you have, please give us athumbs up in a five star rating
on your favorite podcastingplatform, whether it's Apple
Podcast or Spotify, which iswhere most of you listen to the
show, share it with otherpeople.
This is how you can help otherpeople learn how to use AI
properly, is by sharing thispodcast.
(40:36):
Literally open your phone rightnow unless you're driving.
Click on the share button andtype the names of a few people
that you know, either onWhatsApp or Instagram or
LinkedIn or text messages,however you wanna share it with
people or email and just shareit with a few people that can
benefit from this podcast aswell.
I will be really grateful if youdo that, and until next time,
have an awesome rest of yourweek.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Cardiac Cowboys
The heart was always off-limits to surgeons. Cutting into it spelled instant death for the patient. That is, until a ragtag group of doctors scattered across the Midwest and Texas decided to throw out the rule book. Working in makeshift laboratories and home garages, using medical devices made from scavenged machine parts and beer tubes, these men and women invented the field of open heart surgery. Odds are, someone you know is alive because of them. So why has history left them behind? Presented by Chris Pine, CARDIAC COWBOYS tells the gripping true story behind the birth of heart surgery, and the young, Greatest Generation doctors who made it happen. For years, they competed and feuded, racing to be the first, the best, and the most prolific. Some appeared on the cover of Time Magazine, operated on kings and advised presidents. Others ended up disgraced, penniless, and convicted of felonies. Together, they ignited a revolution in medicine, and changed the world.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.