Are you still treating AI like a glorified spell-checker?
If so, you're missing out on at least 90% of its power and that power can save your team hours, boost productivity, and give your business a serious edge.
In this solo masterclass, host Isar Meitis walks through the real-world, tactical ways you should be using AI, especially ChatGPT, Claude, and Gemini to automate complex workflows, enhance proposal writing, improve learning, and cut down time spent on repetitive tasks.
From uploading documents the right way to building custom GPTs that can practically write your RFPs for you, this episode is packed with actionable techniques that most professionals have no idea exist — and it's all tailored for business use.
In this session, you’ll discover:
- The 4-part prompt framework that stops hallucinations and delivers reliable citations
- How to build and share a prompt library for your team’s efficiency
- The secrets of NotebookLM, Study Mode, and Custom GPTs for business use
- Real examples of interactive AI tools that go beyond text — from simulations to training games
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.
GMT20250922-133742_Record (00:00):
Hello and welcome to the Leveraging AI
(00:01):
Podcast, a podcast that sharespractical ethical ways to
leverage AI to improveefficiency, grow your business
in advance your career.
This Isar Mateis, your host, andI'm fighting some really bad
allergies, so I really apologizein advance for my stuffy nose,
but I promise you this is goingto be worth it.
This is going to be anincredibly tactically valuable.
Episode for probably each andevery one of you.
(00:24):
And while it is based on aworkshop that I recently did at
my son's schools to all theteachers in like a teacher
enrichment day, it will behighly applicable for everything
you do at work because we'regoing to touch on more advanced
functionality of the AI modelsthat you all use every single
day, whether it's ChatGPT, orGemini or Claude.
(00:44):
One of the interesting thingswe've learned last week is from
the release of the research doneby OpenAI and by anthropic,
sharing how people actually usetheir models.
And what the model found is thatfirst of all.
Less than half the people useChachi PT for actually work.
A lot of them are using it forpersonal reasons, but the people
(01:04):
who are using it for work, over70% of them use it to help them
write better.
Meaning how to write betteremails, how to edit things that
they wrote in, in differentdocuments and so on.
And that is a very, very smallportion of what AI can actually
do for you.
And so some of the capabilities,more advanced capabilities that
are built and baked.
Into your regular day-to-day AItools are not being used by most
(01:28):
people, and we're going to covera lot of them today.
And yes, we're going to coverthem through the lens of
teaching in a school, but Ipromise you I will give relevant
examples for each and every oneof them, how they could be
applied in business as well.
At the beginning of theworkshop, we worked mostly on
prompting and what are the bestpractices for prompting, which
I'm going to skip because Iassume that most of you in the
(01:50):
audience know how to prompt.
But the next thing we dove into,and I'm gonna be sharing my
screen, but I will explain toyou everything that is on the
screen.
So immediately after prompting,we dove into how to prompt using
reference materials.
When you upload documents toChachi, PT Claude, Gemini, and
so on, there are ways to makethe effort significantly better.
So let's talk about.
(02:10):
What is the difference betweenusing a document and not using a
document in each and every oneof these models, and what kind
of documents you can upload?
Well, first of all, you canupload anything.
You can upload images, PDF, worddocuments, et cetera, and they
could be any kind of documentyou can imagine.
These could be reports thatyou're getting from your team.
These could be reports you'regetting from your clients.
This could be exports.
Statuses from existing systemssuch as your C-R-M-E-R-P, et
(02:32):
cetera.
These could be proposals, thesecould be RFPs, this could be
lesson plans.
This could be any kind ofdocument that you want.
Now when using documents withai, there are two potential
problems.
Problem number one is the AIhallucinating and making stuff
up that's not in the document.
Problem number two is that it isgoing to use information that is
not in the document that itfinds in other sources that is
(02:54):
accurate and correct, but it'snot necessarily the information
from the document.
So when you are using documentsin an AI platform, you want to
add several things to yourprompt.
So I add four components to theprompt and I'm gonna tell you
what they are, and then we'regonna look at an example, and
then I'm gonna tell you how tomake it easier for you every
single time.
So the first part of thisedition is use only information
(03:16):
from the document.
The second part is complimentaryto that, is do not use any
information.
Not from the document.
And I know that sounds a littleredundant, but it reduces the
chances of the AI pullinginformation from a source that
is not the document that youprovided it.
So these are prettystraightforward.
Number three is really critical,and this is the one a lot of
people miss.
(03:36):
Number three saying, if theinformation I'm asking about is
not available in the document,simply say.
Insufficient information orwhatever you wanna put in those
quotation marks, you can say,not available or whatever other
thing.
Why is that important?
Because AI systems are built toplease us, period.
They want to provide you ananswer if you ask the question
(03:56):
and the answer does not exist inthe document, there's a decent
chance it is gonna either makeit up or find it from another
source because it wants toanswer your question.
By the way, we covered a lot ofthat in the recent episode this
past Saturday when it comes tohow can AI labs trade their
models better to reduce that?
But either way, it may provideyou the wrong information
instead of telling you itdoesn't know.
(04:18):
However, if you're telling itwhat to do.
When it does not find theinformation, that is a better
way to please you because it'sactually following your
instructions.
So telling it what to do withinformation is not available is
actually a very good thing forreducing hallucinations.
The fourth.
Component that is also veryimportant is asking for clear
citations.
So what kind of things I ask forin the citations?
(04:39):
I have a standard format.
The format is telling me thename of the document because in
many cases I would upload morethan one the page.
It appears on the name of theheading, and an actual quote
from the document that helpsanswer the question.
And if there's more than one.
Give me several differentcitations with several different
quotes.
Why do I do that?
I do that for several differentreasons.
Reason number one, it forces theAI to ground itself in the
(05:02):
actual document it needs to finda quote, reason number two,
which is as important.
It is going to save me timeverifying the information.
So while yes, you told it to useonly the document and you told
it not to go anywhere else, andyou ask it for citations and you
ask it to tell you theinformation is not there, it may
still make stuff up, which meansyou still need to verify the
information that it gave you.
(05:23):
But if you know the name of thedocument, the page, the
paragraph, and the exact quote,the time it takes you to verify
the information is.
Five times less than if you hadto figure this out on your own.
So let's look at an actualexample.
In this particular example, whatI've done is I've taken a
government RFP for constructionand renovation.
(05:44):
Why that?
Because government RFPs aremultiple documents and they have
more.
Multiple pages and it's not easyto follow and there's lots of
ins and outs and it's hard tofind information in it.
Those of you who've done anygovernment work knows exactly
what I'm talking about.
Those who haven't, you'll justhave to trust me.
And then the prompt that I useis you're an expert government
proposal writer who specializesin writing winning proposals for
government construction RFPs.
(06:05):
You will be provided with adocument as an attachment, and
several questions that youshould answer.
A few important rules for thisprocess.
Number one, you must useinformation only in capitalized
letters.
From the document I attached.
Number two, do not in capitalletters, use any other sources
for any steps of this task.
Number three, if you cannot findinformation that is directly
(06:27):
related to the questions thatyou are asked in the original
document, simply respond with.
Insufficient information inquotation marks.
Number four.
For every question you're askedabout this document, you must
include in every segment of youranswer a citation in the
following format, document name,page number, heading exact
quote, and then ask that toreview the document.
(06:48):
And then I started askingquestions.
What you see in the results isreally remarkable.
So a great example for this is Iasked it what is the performance
period once the job is awarded?
And the answer I got is perfectbecause it says the performance
period is 365 calendar days fromthe date of notice to proceed in
parentheses, NTP.
Is issued by the contractingofficer.
And then I get an actual quotethat says, period of
(07:11):
performance, 365 calendar daysfrom Notice to Proceed.
And then it tells me the name ofthe document, which is
elicitation number and the, inthe PDF page three continuation
sheet.
And then it says the actualquote.
This is perfect.
And then there's another quotefrom a different section in the
document that is saying the samething.
Now, in addition to the fact Iget something that is very, very
(07:31):
specific, I can go and actuallyverify that.
I can go to page three of thedocument and go to the section
that he talks about and thenfind the exact quote.
And it's gonna take me aboutfive seconds.
It can take me five secondsbecause all I need to do is copy
and paste the quote, paste itinto the search inside the pa,
inside that.
Paste it into the search box inthe PDF document, and I can find
(07:51):
that exact quote.
And if I don't, I know that theAI made it up.
Now, by the way, this worksextremely well with GPT five.
It works very poorly in Gemini.
Why?
I don't know, it still makesstuff up, or Gemini, it still
gives me quotes that don'texist.
But this is a situation rightnow, but this works very, very
well with.
ChatGPT, especially in GPT five.
(08:12):
So if you're using any kind ofdocuments, these tips should be
very helpful to you.
Then you may think, this is alot of work.
If I have to do this everysingle time, I have to write all
these things.
Just add, so much text.
Well first of all, it's gonnasave you a lot of time.
But the trick is, is obviouslyto use a prompt library.
So what is a prompt library?
A prompt library is acentralized location that you
and other people that work withyou.
Whether you team yourdepartment, your entire company
(08:34):
can save prompts, or in thisparticular case, segments of
prompts that work well anddeliver results consistently.
You need somebody who's gonna.
Manage that library and it canlive wherever is a shared place.
This could be on Slack, thiscould be on Notion, this could
be on teams.
This could be on a shareddocument on Google Drive.
It doesn't matter.
(08:54):
The way I use it is I use a toolcalled Magical Chrome Extension.
I'm not.
Affiliated with them.
But I really like the tool andI've been using it way before AI
has existed.
All it is is a text expenderthat lives within Chrome.
What do I mean by that?
What I mean by that is you cansee the little M button, those
of you who are watching thescreen, there's a little MI code
on the side of my Chrome browserand when I, and I don't even
(09:18):
have to click on it or doanything.
So in this particular case, whenI showed you how I add these
four rules, I have shortcuts foreach and every one of these.
Prompts.
So if I go to the prompt lineand all I have to type is AI doc
in one word, which is myshortcut that I invented for it
to bring the next little prompt.
So when I type AI doc into theprompt, this entire prompt shows
(09:38):
up everything that I read beforebetween number one and number
four with asking it to do allthese different things.
So every time.
I upload a document to chat GPTor to any of the other tools.
I type AI doc and this additionto the prompt shows up.
How does this work?
Again, if I click and expandthis, you can see that there's
multiple templates and each andevery one of them has a long
segment of a prompt, like youcan see here.
(10:00):
When I'm sharing it, you can seethe whole four steps of the
prompt and it has a shortcut.
Every time I tapped I, everytime I type the shortcut
anywhere.
In Chrome, which means any toolthat I use, the whole prompt
will show up.
So the way I use this, I usethis across everything that I
do.
I'm sharing this in differentfolders with different teams in
the company, and everybody hasaccess to prompts or segments of
(10:21):
prompt little Legos, buildingblocks that we can use for
different things that we need todo.
So if you don't have a promptlibrary for yourself or for your
team right now, this is a greatplace to start while adding
capabilities that everybody canuse and benefit and save a lot
of time while getting consistentresults.
But there are other ways to getconsistent results from ChatGPT
or these other tools, two waysin ChatGPT to do the is using
(10:43):
projects or using custom gpt.
So what are projects?
Projects are these librariesthat exist right under where gpt
are, right above where yourconversations are on the left
menu bar.
Now, in addition to being afolder where you can just put
different chats, there areactually many universes of
context.
Meaning you can create a projectfor a project, you can create a
(11:03):
project for a client, you cancreate a project for a supplier,
you can create a project for aproduct, you can create a
project for a sprint, whateverit is that is a new topic that
you need reference informationabout, and you need that
reference information again andagain and again.
So in this particular case, I'mshowing you how I'm using it for
my clients.
You can see this is a sampleclient project, but I have one
for each and every one of myclients.
And you can see here that itsays.
(11:26):
That I added multiple documents.
I added a document about theculture of the company.
Some of it I've written, some ofit comes from their website.
You can see the course proposalI sent them, the consulting
agreement I have with them adocument that summarizes the
people from the company.
So I have an AI that does aresearch on each and every one
of them from LinkedIn and othersources.
It creates a document about thepeople I work with that's in
there as well.
And then a workshop proposal Igive them, and then the About us
(11:48):
section from A PDF from theirwebsite.
So now every conversation I havein here, instead of just from
the regular new chat, has all ofthat information in there.
I can also add instructions, andso the instructions will give it
even more guidance and morecontext about this particular
project or folder.
So every time I'm creating a newconversation, just a regular
(12:10):
conversation, it has.
All of that as a context, andI'm not starting from scratch,
meaning I don't have to add allof that information manually for
the AI to know about theproject, the client, the the
product, whatever it is thatyou're working on.
So this is one way to save youprompting while still getting
better results.
The other way is obviously tocreate a custom GPT.
What is a custom GPT.
A custom GPT in addition tobeing reference information or
(12:33):
context information for the ai.
It also provides a set ofinstructions, that set of
instructions it can be anythingyou want.
So you can see for me here, I'vegot multiple of these and I got
probably got 25 more.
So if you go to GPT up here, youcan go to see all the gpt that
exist in the GPT store and thereare millions of them that other
people or companies havecreated.
(12:55):
But you can also create your owngpt.
If you go to my gpt, you'll seethat I have.
I don't know, 20, 30 of themthat you can create on your own.
So how do you create A GPT?
You just click all the createGPT button.
You could do it from here or youcould do it from the previous
steps.
So if you're just on the GPTpage, there's a create button
here, and all it has is you needto give it a name, you need to
give it a description.
These don't really matter otherthan you knowing what they do.
(13:15):
And then you write instructions.
These instructions will tell itexactly what to do.
So this could be writing aproposal for you.
I don't write my own proposals.
I write my proposals.
What does it have as far as theinstructions?
It has a very detailed processand as far as knowledge, you can
see where it says knowledge.
Here, you can attach files.
It has two files.
It has a very comprehensivetemplate that I created
(13:37):
specifically for that, that hasall the different aspects that I
provide as services and goods tomy clients.
And it has a PDF that I send tocompanies who ask me about my
services, who have moredescription or longer
descriptions about each andevery one of the things that I
provide.
And the AI instructions tell itto use both these things, but
relate to.
What happened in a recording ofa conversation?
(13:59):
So the input to this GPT isactually a transcript of a call
or calls with my client andemails that I shared with them
that are related to the servicesthat they want me to do for
them, and then just writes theproposal.
The benefit of that are immense.
First of all.
I'm now writing a proposal inabout 15 minutes instead of
about two and a half hours thatit took me before.
These are still 10 to 15 pageslong proposals.
(14:21):
Second, it is amazing atcapturing the actual words and
the way the client describestheir problems, and it makes
that positive proposal.
Now, in theory, everybody shouldknow how to do that.
It definitely does it betterthan I can do it myself.
And the last thing is that itknows how to combine all of this
together in a fluid seamlessway.
That just looks good.
(14:41):
The formatting is great and Ineed to invest very little time.
I still read each and every oneof them the proposal is created
in Canvas.
Those of you who don't knowCanvas, it is a mode in Chachi
PT that opens the actualdocument on the right, and then
this document is editable.
So you can highlight sectionsand ask the AI to fix it for
you, or you can fix it yourself.
So very, very quickly, I gofrom.
(15:02):
Conversations with clients to awritten proposal that then are
just copy and paste into mytemplate on Google Docs and I
can send them the proposal.
So this is just one example, butyou can use it in multiple other
ways.
Going back to why I'm sharingthis with you, we talked about
three ways to shorten the timeto getting great results from
ai, and it all has to do withgiving it the right instructions
(15:22):
in the right context.
So one is having a promptlibrary where you have either
entire prompts or segments ofprompts ready to go.
The second is using projectswhere you just give it context,
and context is king when itcomes to any AI tools.
And number three is creatingcustom GPTs where the
instructions are already builtin, and then you're just gonna
give it the standard input thatit's expecting and it's gonna
(15:43):
give you the right output.
by the way, if you're not aChatGPT user, the same exact
concept exists in Claude and inGemini.
In Gemini, they're called gems.
In Claude, they're calledprojects just to confuse it from
the projects in ChatGPT.
But projects in Claude are morelike custom GPTs rather than
projects because you can givethem detailed instructions for
them to follow every singletime.
(16:03):
And so.
You can do that on the otherplatforms as well.
If you want to learn more on howto use custom gpt, we've done an
entire episode about it.
It was episode 1 75.
It's called Stop Wasting Time,automate Repetitive Task with
Custom gpt.
So you can go back to episode 175 and listen to the entire
episode.
But now let's continue.
So there are two other greattools that you can use in order
(16:25):
to get information ready orlearn new information that you
want to learn, whether it's foryour personal good or for
business needs.
One of them is Notebook LM fromGoogle, and the other is study
mode from all the differenttools.
So let's start with NotebookLM.
What is NotebookLm?
So NotebookLM is a tool thatallows you to drop in multiple
sources that you can see here onthe left.
By the way, you can get to it bygoing to notebook lm.google.com,
(16:49):
and then you can start creatingnew notebooks.
If I go back to my main page,you can see that I have dozens
of these notebooks alreadycreated.
I create.
Probably on average, at leastone a day, sometimes more.
Every time I have access to newinformation, whether it's news
about ai, research papers thatcome out or anything else that I
wanna know about, I just createa new folder.
But in this particular case,I've done it about the school
(17:10):
that I was working because Iwanted to show them information
about the school.
So I dropped several differentweb pages and two PDF files from
the county about the educationin our area, and it created
immediately.
A quick summary of all of it,but in addition, on the right
side, you have these pre-cannedthings that you can create.
You can create an audiooverview, which basically
becomes a podcast that explainsto you in five to 50 minutes
(17:33):
everything that is in theinformation that is there, and
it's not reading theinformation.
It's nearly a podcast.
It's a conversation between twopeople talking about the subject
that you have given it.
The second one is videooverview, which is an upgrade of
that, which is really, reallycool.
So.
It is the same concept.
So think about it as a lecture,but with slides.
So it creates its own slides andthe slides actually come from
the content that you'veprovided.
(17:54):
So if we look at this particularcase about how does the school
report system work, I will sharewith s so you can hear what it's
actually saying.
And if I click play, so everyyear Florida drops this big
report card for all of itspublic schools and each one gets
a simple letter, grade A, B, C,D, or F.
But what is really behind thatsingle letter today we're gonna
(18:15):
dig into then divide by thetotal possible points, simple as
that slides are changing ourstudents taking on more advanced
classes.
And finally, you've got thegraduation rate, which is
exactly what it sounds,explaining what's happening.
A whopping s those sections.
So this is a five minute and 48seconds long video that is more
like a PowerPoint presentation,but it's narrated by a
professional that knowseverything about the content
(18:37):
that I gave it.
It's five minutes and 48 secondsinstead of reading three entire
websites and two very long PDFfiles in order to learn.
And information that I asked itabout.
So that is very helpful.
Other options that exist, thereis a mind map.
Uh, if you wanna build like astructure of something based on
the information, you can use itfor proposals as well.
So a mind map can show you thestructure of the proposal and
(18:58):
even components in it and so on.
Uh, you can create multipletypes of reports, so if you
click on reports, you'll seethat I have different kinds.
Uh, create your own briefingdoc, study guide, blog post, and
there's many other suggestionsthat it's going to suggest
specifically for the contentthat I uploaded.
You can create flashcards tolearn, so I will show you that
as well.
So you can see I createdflashcards for the school, and
(19:18):
this particular case, it'sgeneral questions about the
school.
In what year was Lake YolaCharter School failed it?
And then if you, uh, click onit, it will flip it and you can
see so flashcards that can helpyou learn any content.
And it created 50 questions forme to learn from these
flashcards.
Again, this could be for anybusiness related things.
This could be for onboarding ofnew employees.
This could be for internaltraining.
This could be for any, whateverneed.
(19:40):
Uh, you have either or that, oryour personal needs.
And the third thing that Icreated is a study guide.
So the study guide has a quiz.
With questions and then it hasanswers, and then it also has
essay format questions that youcan use in order to learn for a
specific topic or in order tohelp other people or your kids
learn for a specific topic.
All of that is free to use fromnotebook lamb.
(20:03):
The other tool that is extremelyhelpful that most people don't
know that exists because it'svery well hidden, that can help
you or other people learn, isactually built into each and
every one of the chat platforms.
So in Chachi pt, there's alittle plus button on the left
side of the prompt line.
If you click on that, it showsyou different options.
It says add photos of files, addfrom Google Drive, deep
research, create.
(20:24):
But on the bottom, there's more.
If you hover above more.
One of the options is study andlearn.
The same exact concept exists inGemini, and the same exact
concept exists in Claude.
If you click on that, it saysthat it's in study mode, and
then instead of giving youanswers, it's gonna help you
learn how to solve the problem.
And I'll give you an example.
In this particular case, Iuploaded a sheet of fifth grade
(20:49):
math worksheet with severaldifferent math problems, and I
told it that I only need tosolve the left column of the
math problem and ask it for helpinside of study mode.
It says, okay, great.
Let's tackle just the leftcolumn together.
Here are the four problems wesee and it gave me the problems,
but then instead of giving youthe answers right now, let's
work step by step.
(21:09):
That way you'll actually get toknow how to do these big
multiplications, step one, warmup an estimation.
So it tells me how to doestimation.
So instead of 1235 times 1432,it says, try to do 1200 times
1400, which is like doing 12times 14, which is actually
relatively easy to do and thenadd to zeros.
But then it tells me how to dothe actual math problem.
(21:31):
So I said, okay, step one of1235 times.
Uh.
1432 just multiply the firstdigits, which is two times five.
So it's asking me, what is that?
So I told him two times.
Five is 10 said, oh, exactly.
Good start since two times fiveis 10, we write down the zero
and carry over the one.
Now let's keep going.
And now it says two times three.
(21:52):
Don't forget to add and carrythe one after you multiply.
Can you try that one?
So I asked why two times threesaid, ah, good question.
Let's slow down.
And make it crystal clear, wemultiply 1235 at 1432.
We, when we multiply the two.
The one digits of the 1332, wemultiplied by every digit of the
(22:13):
1235.
So first we do two times five.
Now we do two times three, whichis the 10th digit of.
So it's a great explanation andit's gonna keep going and help
me do this, ask you to actuallyshow me a graphical
representation of the process soit does, so I can see exactly
where to write the differentdigits.
All of that instead of giving methe answer.
So this is study mode thatexists in all these tools.
So if you wanna learn somethinginstead of just getting the
(22:33):
answer, or if you wanna helpyour kids or people in your
department learn somethingwithout just getting them the
answer, that's a great way to dothat.
But then we get to the reallycool stuff.
So beyond all these things,they're just text and that are
very helpful and powerful.
You can do significantly moresophisticated things.
So I created three differentmedia applications that run
within the large language modelsthat can help.
(22:56):
In this particular case, kidslearn stuff, but you can do
exactly the same to teach peoplehow to do different things in
your organization.
So the first one I said, learnmath while playing soccer.
This is something I actuallydeveloped together with my
10-year-old son because when thesummer started, and he really
loves doing math, he asked me togive him math problems and.
Don't know how to give him mathproblems that will actually help
him.
So I said, what if we create agame that will give you math
(23:18):
problems?
Say, oh yeah, I love game.
Let's create a game.
I said, okay, what game wouldyou like to create?
And at that time, he wasactually playing a really low
quality from a graphicsperspective.
A penalty kick shootout game ona website that he likes to play.
And he said, what if we createsomething like this, but with
math problems?
Like, okay, that's a great idea.
And so we created this game.
So how does the game look likefor those of you who don't have
(23:39):
it, I'll actually gonna show youhow it looked like when it runs.
So when it runs, the first thingthat opens is a set of
instructions.
And the instructions say, whatis the game objectives?
Two players compete in a fiveround penalty kick shootout
using math scales, kicker inblue and goalie in orange, and
then how to play each round,both players answer a math
question, kickers answer first.
(24:01):
Then the goalie.
The penalty kick is resolvedbased on both answers after five
rounds.
The play with most goals orsaves wins.
So how does it work if both hitit right?
It's a 50 50 chance, it's arandom outcome.
If the kicker is right and thegoal is wrong, it's a goal.
If the kicker is wrong and thegoal is right, it's a save.
If both are wrong, it's a 50 50chance again.
And then it tells you what themath topics are and then what
(24:23):
the, how the to play the game.
And then you just click go.
And then what you see on thedisplay is that we have a
goalie.
And we have a kicker and we havea goal and we have fans and it's
all relatively low resolution.
But that's how he wanted it.
He wanted it to be a low qualityfrom a graphics perspective
because it's similar to the gamehe was playing.
And then there's a question onthe bottom.
What is 50% of what?
(24:43):
96.
So this particular case, theanswer is 98.
If I kick it says, oh yeah,that's correct.
Now it says, okay, what is 50%of 180 7?
That's the question that the,uh, that the goalie got.
And let's say that goalie getsit wrong.
So I would say it's 55 save, andit says wrong.
The answer is that he gives youthe right answer and the kicker
(25:03):
kicks and the ball goes into thegoal.
So now on the right side, I cansee the history round one
kicker.
What is the answer?
He got the right question, theright answer.
Goalie got it wrong.
And then the goal was scored.
And on top you can see thatthere's one goal round two of
five, and the goalie has zerosaves.
The game goes on and on.
After five rounds, it goes tothe next level and there's gonna
be tougher questions.
And the same on and on for levelthree and so on.
(25:25):
Why is this so powerful?
It's so powerful because it'ssomething I developed together
with my kid.
He was really excited creatingthe game because when we started
the game, it did not look likethis.
The first round of the gamewhere we just started working on
it, looked like this.
So I will show you.
We created it in Claude, and ifI'm gonna scroll all the way to
the top.
(25:45):
And show you the first round ofit.
It was very poor quality.
It did not have the history onthe side, but the concept was
the same.
The kicker was just a square.
The goalie was just a square.
But the concept was alreadythere.
And from that moment went on, wejust kept iterating and giving
it more and more instructionswith ideas.
So you can see how the display,those of you who are just
listening, I promise you, thedisplay just get better and
better.
It gets a little bit of a 3Dvibe.
(26:06):
And the people start lookinglike mini figures instead of
just a square, and then it addsthe actual kicking area.
And slowly we added more andmore stuff.
In the end, we added the fans inthe background, and made the
questions more complicated andmake it not repeat any questions
and added the history on theside.
So he was very excited aboutthis because he was able to
build his own game, and becauseof that, the whole family played
(26:27):
this game for about three days,trying to beat each other in
math questions, but even just inthe process of creating the
game, because every route we hadto solve the problems to see
what happens, then probablydoing, developing the game, he
probably solved.
I don't know, 50 to 80 mathquestions, which is 50 to 80
math questions more than hewould've solved otherwise in the
first week of his summer break.
And so again, you can dosomething very similar for your
(26:50):
employees and you can just findsomething that will excite them
as far as the process or createany other kind of application
that can help your employeesachieve and do different things.
And you can do it in any ofthese platforms.
The same exact thing can bedone.
Not just on Claude, but alsoChatGPT and Gemini.
So let's move on and look at thenext example.
The next example is me wantingto show that you could go beyond
(27:13):
just math.
So in this particular case itwas talking to Greek
philosophers.
So you can talk to Socrates,Play-Doh and Aristotle.
How did I do that?
So in this particular case,before playing the game, I wanna
show you how the game wascreated.
So the game was created becauseI did not know exactly where to
go.
So I actually consulted with theAI again in this particular
case, Claude, afterwards, I'llshow you another example, but I
(27:36):
started with what is covered inthe curriculum of social studies
for sixth to eighth grade.
And then he told me what iscovered in each and every one of
them.
And then I asked in the sixthgrade curriculum.
Includes any Greek scholars andthey said, yeah, we talk about
philosophers and scientists andso on.
And I said, okay, please createa detailed curriculum of sixth
grade through eighth grade.
Uh, need to know about Socrates,Plato, and Aristotle.
(27:57):
And I gave it to it as a deepresearch project.
So it's kind of the same in allof them.
There's a function that's calleddeep research, and when you
select that, instead of justgiving you an answer, it will go
and research.
Multiple websites.
In this particular case, you cansee 291 sources that it actually
read.
In order to provide me thatinformation, it took seven
minutes and 17 seconds, so a lotfaster than it will take me to
(28:20):
go through 291 sources, and itgave me a detailed review.
Of everything that I need toknow as far as what is taught in
six to eight grades.
And you can see that's a very,very, very long document.
It's probably 20 pages long.
And based on that, I ask it torecommend interactive games that
can develop and run inside ofClaude that would allow six to
eight graders to interact withthose Greek philosophers.
(28:40):
And then it gave me severaldifferent options.
I picked one option.
And I ask it to use voiceinstead of just text.
And we ended up with a game thatif I scroll all the way, well
actually, I don't need to scrollbecause I have the actual thing
here.
So if we look at the way thegame looks, uh, it asks me for
my name, I'll give it my name,ISAR, and then we refer to me
and it says, voice on.
I will keep the voice on andyou'll be able to hear what's
happening now.
(29:01):
Welcome to ancient Athens Za.
Choose a philosopher to beginyour quest for wisdom.
So I have Socrates, Plato, andAristotle.
Let's go with Plato.
Welcome future guardian ofwisdom.
Let us explore what lies beyondthe shadows of ignorance.
Are you ready to see the lightof truth?
Is what we see in virtualreality real?
How do we know what's trulyreal?
So now I can type my answer, Iguess, if we feel.
(29:27):
And see it in all senses.
It should be real.
Right?
Question mark.
Now Plato is thinking.
It's gonna give me his nextanswer.
Ah, my young friend.
But what if I made you feel asthough you were falling in a
dream so vivid that all yoursenses believed it completely?
Would that dream fall be realsimply because you felt it?
(29:49):
You see, our senses can be likeprisoners chained in a cave,
mistaking shadows on the wallfor reality itself when the true
forms of things exist in a realmbeyond what we merely touch and
see.
Now what you see here is areally cool combination of the
data from the deep research thatknows about Plato's Cave and the
shadows on the wall, which isstuff that he talks about a lot
combined with relating it to myanswer.
(30:11):
And the reason I developed thisparticular game in Claude is
because out of these threemodels, Claude, Chachi, and
Gemini.
The only one that can actuallyuse the Claudeed thinking brain,
meaning the actual largelanguage models, as it's running
code is Claudeed.
So really what's happening herebehind the scenes is when you
saw Plato thinking, quoteunquote, it's actually Claudeed
from its code in the artifactthat it created on its own.
(30:35):
Is calling the Claude API behindthe scenes to provide these
answers, which enable this verylively conversation.
Again, you can use this to trainyour people or allow them to get
any information about anythingyou want, whether it's how to
apply for a PTO as far as an HRquestion, or how to set up a
specific server in whatever ITenvironment or whatever the case
(30:57):
may be.
You can give it informationfrom.
Data that you give it or throughthe research process and then
allow it to think through it,review the information, and
provide you answers in a gamevoice conversation or any other
kind of solution that you wantin a few minutes of development
and having something that youcan share with your entire
employees.
I find this extremely valuableand extremely helpful, and
(31:19):
again.
99% of people don't know thatthat's possible.
So now you're one of a, so nowyou're a part of a really small,
unique group that actually knowsthat this is possible and can
take a benefit of it.
The third elastic example thatI've used is to show how you can
use a sophisticated graphicdisplay to help people
understand more advancedconcepts.
Now, some of you may know thisalready, but I was a F 16 pilot
(31:41):
back in the day many years ago,so I'm very well connected to
flight.
I was a flight instructor aswell.
I also have a private, licenseand so teaching people how
flight works is something I'mvery passionate about because
nobody thinks about it.
But when you board a 7 47, A 747, a takeoff is weighing more
than 400.
Metric tons.
(32:02):
That's over 800,000 pounds.
And we never ask ourself, howthe hell does this thing get
airborne?
That doesn't make any sense.
Well.
Simple physics make it possible.
But that simple physics actuallybreaks into four different
forces.
You have lift, weight, thrustdrag, which actually impact how
much lift an airplane cangenerate.
(32:22):
So what I did here is I createda mini simulators that allows
people to explore theseconcepts.
So it starts with the first pagewith just different ideas.
What are the four forces?
What is an angle of attack?
Uh, what is air speed altitudeand how these things balance.
But then the cool thing isthere's an experiment, but there
is an experimentation section,and in that section on the
(32:43):
right, you see an actual diagramof the wing and you can see the
air flowing much faster on topof the wing than it is flowing
on the bottom of the wing, whichis the whole trick because high
speed generates low pressure,which then basically pulls the
airplane up.
But the other thing on the leftside, there are sliders.
If I change the sliders as anexample of angle of attack, you
can see the wing rotating andyou can see two things that are
happening.
(33:03):
It's gonna tell us that from alevel flight we're now climbing
and it says current situationlift is greater.
Then wait, so the plane ascendsagainst gravity.
Cool.
Right.
The other thing that you can seethat it starts showing.
These turbulence behind thewing.
So if I keep increasing theangle of attack, eventually you
sit.
Now it's going to descend.
Why?
Because the plane is notgenerating enough lift to
(33:24):
overcome the weight, and I keepon going.
It says too much turbulence format a high angle of attack.
Lift decreases below weight anda stall can occur.
This is really cool.
Now the other thing, if I takethis down back to a normal angle
of attack and I increase thealtitude.
Then you can see another thingthat's happening.
The lift is going to decreasebecause now the air is thinner.
(33:44):
It's gonna tell me that currentsituation, the plane is not
generating enough lift toovercome the weight.
Higher altitude means thinnerair, reducing air density, and
therefore reducing lift.
So I can play around with thesliders and see how that impacts
the actual lift on the wing.
That's very cool.
But again, you can take this toanything in business that you
can explain in a graphical wayand create it in a visual way
(34:04):
that will help people interactwith it and better understand
how it works.
This is not rocket scienceeither.
Going back, this was created inChatGPT and not in Claude.
Just to show you that ChatGPTcan do the same thing.
By the way, Gemini can do thesame thing as well, and if I
scroll back all the way to thetop, you can see what I asked it
for.
I said, I am a science teacherin the eighth grade, and we are
learning about the physics ofelevation in airplanes and so on
(34:25):
and so forth.
I asked it for ideas on how wecould do that.
It asked me several differentquestions because I asked it to
ask me question.
I said, before you startprogramming this, uh, please let
me know if you have anyquestions.
So it asked me clarificationquestions about the level of
physics I want to go into,whether, what kind of aircraft I
want, what kind of units I wantto use, and so on and so forth.
And then we together, afteranswering the question, got to
(34:45):
the conclusion what we wanna do.
Version one was really ugly andnot very helpful, and I iterated
back and forth until we got tothe version you just saw.
So what does that mean to eachand every one of you in your
business?
It means that all these tools,chat, bt, Gemini, clo, et
cetera, can do a lot more thanmost people know that they can
do.
All you have to do is know howto use the tools.
So, several different things wecovered today.
(35:06):
One is using a prompt library tosave your better prompts.
Two is learning how to usereference material in a way that
will dramatically increase thechances of getting the correct
answer, and that will help youverify the information much
faster.
We looked at Notebook LM as away to summarize lots of
information and get a lot ofdifferent outputs out of it.
We looked at study mode insideof the models that allows you to
(35:27):
learn stuff instead of just getan answer, and we looked at
creating interactive.
Dashboards, learningenvironments, call them,
whatever you wanna call them,that you can use across multiple
aspects of your business andpersonal life in order to help
people around you, whether it'sstudents, your kids, yourself,
your employees learn thingsbetter and faster.
I really hope you found thishighly impactful.
(35:48):
I really hope that you're gonnatest and play with this
preferably today, so you don'tforget, uh, any of the things
that we talked about and sharethis with other people.
This is, again, a highly.
Tactical episode, these are theones that I love the most, and
you can share it with otherpeople that can learn how to do
these kind of things and developit for things they are
interested in creating orteaching again, either in
business or in school or intheir personal life.
(36:10):
If you found this episodehelpful, please subscribe to the
podcast.
We really appreciate all thepeople who are subscribed.
There's tens of thousands ofyou, but every single person.
But if you are not one of them,then please join the people who
are already subscribe, share itwith other people who can
benefit from it.
And if you really in a goodmood, then give us a five star
rating in a short description ofwhy you like this podcast on
(36:32):
Apple Podcast and or Spotify.
And until next time, have anamazing rest of your week.
Hello and welcome to the Leveraging AI
(00:01):
Podcast, a podcast that sharespractical ethical ways to
leverage AI to improveefficiency, grow your business
in advance your career.
This Isar Mateis, your host, andI'm fighting some really bad
allergies, so I really apologizein advance for my stuffy nose,
but I promise you this is goingto be worth it.
This is going to be anincredibly tactically valuable.
Episode for probably each andevery one of you.
(00:24):
And while it is based on aworkshop that I recently did at
my son's schools to all theteachers in like a teacher
enrichment day, it will behighly applicable for everything
you do at work because we'regoing to touch on more advanced
functionality of the AI modelsthat you all use every single
day, whether it's ChatGPT, orGemini or Claude.
(00:44):
One of the interesting thingswe've learned last week is from
the release of the research doneby OpenAI and by anthropic,
sharing how people actually usetheir models.
And what the model found is thatfirst of all.
Less than half the people useChachi PT for actually work.
A lot of them are using it forpersonal reasons, but the people
(01:04):
who are using it for work, over70% of them use it to help them
write better.
Meaning how to write betteremails, how to edit things that
they wrote in, in differentdocuments and so on.
And that is a very, very smallportion of what AI can actually
do for you.
And so some of the capabilities,more advanced capabilities that
are built and baked.
Into your regular day-to-day AItools are not being used by most
(01:28):
people, and we're going to covera lot of them today.
And yes, we're going to coverthem through the lens of
teaching in a school, but Ipromise you I will give relevant
examples for each and every oneof them, how they could be
applied in business as well.
At the beginning of theworkshop, we worked mostly on
prompting and what are the bestpractices for prompting, which
I'm going to skip because Iassume that most of you in the
(01:50):
audience know how to prompt.
But the next thing we dove into,and I'm gonna be sharing my
screen, but I will explain toyou everything that is on the
screen.
So immediately after prompting,we dove into how to prompt using
reference materials.
When you upload documents toChachi, PT Claude, Gemini, and
so on, there are ways to makethe effort significantly better.
So let's talk about.
(02:10):
What is the difference betweenusing a document and not using a
document in each and every oneof these models, and what kind
of documents you can upload?
Well, first of all, you canupload anything.
You can upload images, PDF, worddocuments, et cetera, and they
could be any kind of documentyou can imagine.
These could be reports thatyou're getting from your team.
These could be reports you'regetting from your clients.
This could be exports.
Statuses from existing systemssuch as your C-R-M-E-R-P, et
(02:32):
cetera.
These could be proposals, thesecould be RFPs, this could be
lesson plans.
This could be any kind ofdocument that you want.
Now when using documents withai, there are two potential
problems.
Problem number one is the AIhallucinating and making stuff
up that's not in the document.
Problem number two is that it isgoing to use information that is
not in the document that itfinds in other sources that is
(02:54):
accurate and correct, but it'snot necessarily the information
from the document.
So when you are using documentsin an AI platform, you want to
add several things to yourprompt.
So I add four components to theprompt and I'm gonna tell you
what they are, and then we'regonna look at an example, and
then I'm gonna tell you how tomake it easier for you every
single time.
So the first part of thisedition is use only information
(03:16):
from the document.
The second part is complimentaryto that, is do not use any
information.
Not from the document.
And I know that sounds a littleredundant, but it reduces the
chances of the AI pullinginformation from a source that
is not the document that youprovided it.
So these are prettystraightforward.
Number three is really critical,and this is the one a lot of
people miss.
(03:36):
Number three saying, if theinformation I'm asking about is
not available in the document,simply say.
Insufficient information orwhatever you wanna put in those
quotation marks, you can say,not available or whatever other
thing.
Why is that important?
Because AI systems are built toplease us, period.
They want to provide you ananswer if you ask the question
(03:56):
and the answer does not exist inthe document, there's a decent
chance it is gonna either makeit up or find it from another
source because it wants toanswer your question.
By the way, we covered a lot ofthat in the recent episode this
past Saturday when it comes tohow can AI labs trade their
models better to reduce that?
But either way, it may provideyou the wrong information
instead of telling you itdoesn't know.
(04:18):
However, if you're telling itwhat to do.
When it does not find theinformation, that is a better
way to please you because it'sactually following your
instructions.
So telling it what to do withinformation is not available is
actually a very good thing forreducing hallucinations.
The fourth.
Component that is also veryimportant is asking for clear
citations.
So what kind of things I ask forin the citations?
(04:39):
I have a standard format.
The format is telling me thename of the document because in
many cases I would upload morethan one the page.
It appears on the name of theheading, and an actual quote
from the document that helpsanswer the question.
And if there's more than one.
Give me several differentcitations with several different
quotes.
Why do I do that?
I do that for several differentreasons.
Reason number one, it forces theAI to ground itself in the
(05:02):
actual document it needs to finda quote, reason number two,
which is as important.
It is going to save me timeverifying the information.
So while yes, you told it to useonly the document and you told
it not to go anywhere else, andyou ask it for citations and you
ask it to tell you theinformation is not there, it may
still make stuff up, which meansyou still need to verify the
information that it gave you.
(05:23):
But if you know the name of thedocument, the page, the
paragraph, and the exact quote,the time it takes you to verify
the information is.
Five times less than if you hadto figure this out on your own.
So let's look at an actualexample.
In this particular example, whatI've done is I've taken a
government RFP for constructionand renovation.
(05:44):
Why that?
Because government RFPs aremultiple documents and they have
more.
Multiple pages and it's not easyto follow and there's lots of
ins and outs and it's hard tofind information in it.
Those of you who've done anygovernment work knows exactly
what I'm talking about.
Those who haven't, you'll justhave to trust me.
And then the prompt that I useis you're an expert government
proposal writer who specializesin writing winning proposals for
government construction RFPs.
(06:05):
You will be provided with adocument as an attachment, and
several questions that youshould answer.
A few important rules for thisprocess.
Number one, you must useinformation only in capitalized
letters.
From the document I attached.
Number two, do not in capitalletters, use any other sources
for any steps of this task.
Number three, if you cannot findinformation that is directly
(06:27):
related to the questions thatyou are asked in the original
document, simply respond with.
Insufficient information inquotation marks.
Number four.
For every question you're askedabout this document, you must
include in every segment of youranswer a citation in the
following format, document name,page number, heading exact
quote, and then ask that toreview the document.
(06:48):
And then I started askingquestions.
What you see in the results isreally remarkable.
So a great example for this is Iasked it what is the performance
period once the job is awarded?
And the answer I got is perfectbecause it says the performance
period is 365 calendar days fromthe date of notice to proceed in
parentheses, NTP.
Is issued by the contractingofficer.
And then I get an actual quotethat says, period of
(07:11):
performance, 365 calendar daysfrom Notice to Proceed.
And then it tells me the name ofthe document, which is
elicitation number and the, inthe PDF page three continuation
sheet.
And then it says the actualquote.
This is perfect.
And then there's another quotefrom a different section in the
document that is saying the samething.
Now, in addition to the fact Iget something that is very, very
(07:31):
specific, I can go and actuallyverify that.
I can go to page three of thedocument and go to the section
that he talks about and thenfind the exact quote.
And it's gonna take me aboutfive seconds.
It can take me five secondsbecause all I need to do is copy
and paste the quote, paste itinto the search inside the pa,
inside that.
Paste it into the search box inthe PDF document, and I can find
(07:51):
that exact quote.
And if I don't, I know that theAI made it up.
Now, by the way, this worksextremely well with GPT five.
It works very poorly in Gemini.
Why?
I don't know, it still makesstuff up, or Gemini, it still
gives me quotes that don'texist.
But this is a situation rightnow, but this works very, very
well with.
ChatGPT, especially in GPT five.
(08:12):
So if you're using any kind ofdocuments, these tips should be
very helpful to you.
Then you may think, this is alot of work.
If I have to do this everysingle time, I have to write all
these things.
Just add, so much text.
Well first of all, it's gonnasave you a lot of time.
But the trick is, is obviouslyto use a prompt library.
So what is a prompt library?
A prompt library is acentralized location that you
and other people that work withyou.
Whether you team yourdepartment, your entire company
(08:34):
can save prompts, or in thisparticular case, segments of
prompts that work well anddeliver results consistently.
You need somebody who's gonna.
Manage that library and it canlive wherever is a shared place.
This could be on Slack, thiscould be on Notion, this could
be on teams.
This could be on a shareddocument on Google Drive.
It doesn't matter.
(08:54):
The way I use it is I use a toolcalled Magical Chrome Extension.
I'm not.
Affiliated with them.
But I really like the tool andI've been using it way before AI
has existed.
All it is is a text expenderthat lives within Chrome.
What do I mean by that?
What I mean by that is you cansee the little M button, those
of you who are watching thescreen, there's a little MI code
on the side of my Chrome browserand when I, and I don't even
(09:18):
have to click on it or doanything.
So in this particular case, whenI showed you how I add these
four rules, I have shortcuts foreach and every one of these.
Prompts.
So if I go to the prompt lineand all I have to type is AI doc
in one word, which is myshortcut that I invented for it
to bring the next little prompt.
So when I type AI doc into theprompt, this entire prompt shows
(09:38):
up everything that I read beforebetween number one and number
four with asking it to do allthese different things.
So every time.
I upload a document to chat GPTor to any of the other tools.
I type AI doc and this additionto the prompt shows up.
How does this work?
Again, if I click and expandthis, you can see that there's
multiple templates and each andevery one of them has a long
segment of a prompt, like youcan see here.
(10:00):
When I'm sharing it, you can seethe whole four steps of the
prompt and it has a shortcut.
Every time I tapped I, everytime I type the shortcut
anywhere.
In Chrome, which means any toolthat I use, the whole prompt
will show up.
So the way I use this, I usethis across everything that I
do.
I'm sharing this in differentfolders with different teams in
the company, and everybody hasaccess to prompts or segments of
(10:21):
prompt little Legos, buildingblocks that we can use for
different things that we need todo.
So if you don't have a promptlibrary for yourself or for your
team right now, this is a greatplace to start while adding
capabilities that everybody canuse and benefit and save a lot
of time while getting consistentresults.
But there are other ways to getconsistent results from ChatGPT
or these other tools, two waysin ChatGPT to do the is using
(10:43):
projects or using custom gpt.
So what are projects?
Projects are these librariesthat exist right under where gpt
are, right above where yourconversations are on the left
menu bar.
Now, in addition to being afolder where you can just put
different chats, there areactually many universes of
context.
Meaning you can create a projectfor a project, you can create a
(11:03):
project for a client, you cancreate a project for a supplier,
you can create a project for aproduct, you can create a
project for a sprint, whateverit is that is a new topic that
you need reference informationabout, and you need that
reference information again andagain and again.
So in this particular case, I'mshowing you how I'm using it for
my clients.
You can see this is a sampleclient project, but I have one
for each and every one of myclients.
And you can see here that itsays.
(11:26):
That I added multiple documents.
I added a document about theculture of the company.
Some of it I've written, some ofit comes from their website.
You can see the course proposalI sent them, the consulting
agreement I have with them adocument that summarizes the
people from the company.
So I have an AI that does aresearch on each and every one
of them from LinkedIn and othersources.
It creates a document about thepeople I work with that's in
there as well.
And then a workshop proposal Igive them, and then the About us
(11:48):
section from A PDF from theirwebsite.
So now every conversation I havein here, instead of just from
the regular new chat, has all ofthat information in there.
I can also add instructions, andso the instructions will give it
even more guidance and morecontext about this particular
project or folder.
So every time I'm creating a newconversation, just a regular
(12:10):
conversation, it has.
All of that as a context, andI'm not starting from scratch,
meaning I don't have to add allof that information manually for
the AI to know about theproject, the client, the the
product, whatever it is thatyou're working on.
So this is one way to save youprompting while still getting
better results.
The other way is obviously tocreate a custom GPT.
What is a custom GPT.
A custom GPT in addition tobeing reference information or
(12:33):
context information for the ai.
It also provides a set ofinstructions, that set of
instructions it can be anythingyou want.
So you can see for me here, I'vegot multiple of these and I got
probably got 25 more.
So if you go to GPT up here, youcan go to see all the gpt that
exist in the GPT store and thereare millions of them that other
people or companies havecreated.
(12:55):
But you can also create your owngpt.
If you go to my gpt, you'll seethat I have.
I don't know, 20, 30 of themthat you can create on your own.
So how do you create A GPT?
You just click all the createGPT button.
You could do it from here or youcould do it from the previous
steps.
So if you're just on the GPTpage, there's a create button
here, and all it has is you needto give it a name, you need to
give it a description.
These don't really matter otherthan you knowing what they do.
(13:15):
And then you write instructions.
These instructions will tell itexactly what to do.
So this could be writing aproposal for you.
I don't write my own proposals.
I write my proposals.
What does it have as far as theinstructions?
It has a very detailed processand as far as knowledge, you can
see where it says knowledge.
Here, you can attach files.
It has two files.
It has a very comprehensivetemplate that I created
(13:37):
specifically for that, that hasall the different aspects that I
provide as services and goods tomy clients.
And it has a PDF that I send tocompanies who ask me about my
services, who have moredescription or longer
descriptions about each andevery one of the things that I
provide.
And the AI instructions tell itto use both these things, but
relate to.
What happened in a recording ofa conversation?
(13:59):
So the input to this GPT isactually a transcript of a call
or calls with my client andemails that I shared with them
that are related to the servicesthat they want me to do for
them, and then just writes theproposal.
The benefit of that are immense.
First of all.
I'm now writing a proposal inabout 15 minutes instead of
about two and a half hours thatit took me before.
These are still 10 to 15 pageslong proposals.
(14:21):
Second, it is amazing atcapturing the actual words and
the way the client describestheir problems, and it makes
that positive proposal.
Now, in theory, everybody shouldknow how to do that.
It definitely does it betterthan I can do it myself.
And the last thing is that itknows how to combine all of this
together in a fluid seamlessway.
That just looks good.
(14:41):
The formatting is great and Ineed to invest very little time.
I still read each and every oneof them the proposal is created
in Canvas.
Those of you who don't knowCanvas, it is a mode in Chachi
PT that opens the actualdocument on the right, and then
this document is editable.
So you can highlight sectionsand ask the AI to fix it for
you, or you can fix it yourself.
So very, very quickly, I gofrom.
(15:02):
Conversations with clients to awritten proposal that then are
just copy and paste into mytemplate on Google Docs and I
can send them the proposal.
So this is just one example, butyou can use it in multiple other
ways.
Going back to why I'm sharingthis with you, we talked about
three ways to shorten the timeto getting great results from
ai, and it all has to do withgiving it the right instructions
(15:22):
in the right context.
So one is having a promptlibrary where you have either
entire prompts or segments ofprompts ready to go.
The second is using projectswhere you just give it context,
and context is king when itcomes to any AI tools.
And number three is creatingcustom GPTs where the
instructions are already builtin, and then you're just gonna
give it the standard input thatit's expecting and it's gonna
(15:43):
give you the right output.
by the way, if you're not aChatGPT user, the same exact
concept exists in Claude and inGemini.
In Gemini, they're called gems.
In Claude, they're calledprojects just to confuse it from
the projects in ChatGPT.
But projects in Claude are morelike custom GPTs rather than
projects because you can givethem detailed instructions for
them to follow every singletime.
(16:03):
And so.
You can do that on the otherplatforms as well.
If you want to learn more on howto use custom gpt, we've done an
entire episode about it.
It was episode 1 75.
It's called Stop Wasting Time,automate Repetitive Task with
Custom gpt.
So you can go back to episode 175 and listen to the entire
episode.
But now let's continue.
So there are two other greattools that you can use in order
(16:25):
to get information ready orlearn new information that you
want to learn, whether it's foryour personal good or for
business needs.
One of them is Notebook LM fromGoogle, and the other is study
mode from all the differenttools.
So let's start with NotebookLM.
What is NotebookLm?
So NotebookLM is a tool thatallows you to drop in multiple
sources that you can see here onthe left.
By the way, you can get to it bygoing to notebook lm.google.com,
(16:49):
and then you can start creatingnew notebooks.
If I go back to my main page,you can see that I have dozens
of these notebooks alreadycreated.
I create.
Probably on average, at leastone a day, sometimes more.
Every time I have access to newinformation, whether it's news
about ai, research papers thatcome out or anything else that I
wanna know about, I just createa new folder.
But in this particular case,I've done it about the school
(17:10):
that I was working because Iwanted to show them information
about the school.
So I dropped several differentweb pages and two PDF files from
the county about the educationin our area, and it created
immediately.
A quick summary of all of it,but in addition, on the right
side, you have these pre-cannedthings that you can create.
You can create an audiooverview, which basically
becomes a podcast that explainsto you in five to 50 minutes
(17:33):
everything that is in theinformation that is there, and
it's not reading theinformation.
It's nearly a podcast.
It's a conversation between twopeople talking about the subject
that you have given it.
The second one is videooverview, which is an upgrade of
that, which is really, reallycool.
So.
It is the same concept.
So think about it as a lecture,but with slides.
So it creates its own slides andthe slides actually come from
the content that you'veprovided.
(17:54):
So if we look at this particularcase about how does the school
report system work, I will sharewith s so you can hear what it's
actually saying.
And if I click play, so everyyear Florida drops this big
report card for all of itspublic schools and each one gets
a simple letter, grade A, B, C,D, or F.
But what is really behind thatsingle letter today we're gonna
(18:15):
dig into then divide by thetotal possible points, simple as
that slides are changing ourstudents taking on more advanced
classes.
And finally, you've got thegraduation rate, which is
exactly what it sounds,explaining what's happening.
A whopping s those sections.
So this is a five minute and 48seconds long video that is more
like a PowerPoint presentation,but it's narrated by a
professional that knowseverything about the content
(18:37):
that I gave it.
It's five minutes and 48 secondsinstead of reading three entire
websites and two very long PDFfiles in order to learn.
And information that I asked itabout.
So that is very helpful.
Other options that exist, thereis a mind map.
Uh, if you wanna build like astructure of something based on
the information, you can use itfor proposals as well.
So a mind map can show you thestructure of the proposal and
(18:58):
even components in it and so on.
Uh, you can create multipletypes of reports, so if you
click on reports, you'll seethat I have different kinds.
Uh, create your own briefingdoc, study guide, blog post, and
there's many other suggestionsthat it's going to suggest
specifically for the contentthat I uploaded.
You can create flashcards tolearn, so I will show you that
as well.
So you can see I createdflashcards for the school, and
(19:18):
this particular case, it'sgeneral questions about the
school.
In what year was Lake YolaCharter School failed it?
And then if you, uh, click onit, it will flip it and you can
see so flashcards that can helpyou learn any content.
And it created 50 questions forme to learn from these
flashcards.
Again, this could be for anybusiness related things.
This could be for onboarding ofnew employees.
This could be for internaltraining.
This could be for any, whateverneed.
(19:40):
Uh, you have either or that, oryour personal needs.
And the third thing that Icreated is a study guide.
So the study guide has a quiz.
With questions and then it hasanswers, and then it also has
essay format questions that youcan use in order to learn for a
specific topic or in order tohelp other people or your kids
learn for a specific topic.
All of that is free to use fromnotebook lamb.
(20:03):
The other tool that is extremelyhelpful that most people don't
know that exists because it'svery well hidden, that can help
you or other people learn, isactually built into each and
every one of the chat platforms.
So in Chachi pt, there's alittle plus button on the left
side of the prompt line.
If you click on that, it showsyou different options.
It says add photos of files, addfrom Google Drive, deep
research, create.
(20:24):
But on the bottom, there's more.
If you hover above more.
One of the options is study andlearn.
The same exact concept exists inGemini, and the same exact
concept exists in Claude.
If you click on that, it saysthat it's in study mode, and
then instead of giving youanswers, it's gonna help you
learn how to solve the problem.
And I'll give you an example.
In this particular case, Iuploaded a sheet of fifth grade
(20:49):
math worksheet with severaldifferent math problems, and I
told it that I only need tosolve the left column of the
math problem and ask it for helpinside of study mode.
It says, okay, great.
Let's tackle just the leftcolumn together.
Here are the four problems wesee and it gave me the problems,
but then instead of giving youthe answers right now, let's
work step by step.
(21:09):
That way you'll actually get toknow how to do these big
multiplications, step one, warmup an estimation.
So it tells me how to doestimation.
So instead of 1235 times 1432,it says, try to do 1200 times
1400, which is like doing 12times 14, which is actually
relatively easy to do and thenadd to zeros.
But then it tells me how to dothe actual math problem.
(21:31):
So I said, okay, step one of1235 times.
Uh.
1432 just multiply the firstdigits, which is two times five.
So it's asking me, what is that?
So I told him two times.
Five is 10 said, oh, exactly.
Good start since two times fiveis 10, we write down the zero
and carry over the one.
Now let's keep going.
And now it says two times three.
(21:52):
Don't forget to add and carrythe one after you multiply.
Can you try that one?
So I asked why two times threesaid, ah, good question.
Let's slow down.
And make it crystal clear, wemultiply 1235 at 1432.
We, when we multiply the two.
The one digits of the 1332, wemultiplied by every digit of the
(22:13):
1235.
So first we do two times five.
Now we do two times three, whichis the 10th digit of.
So it's a great explanation andit's gonna keep going and help
me do this, ask you to actuallyshow me a graphical
representation of the process soit does, so I can see exactly
where to write the differentdigits.
All of that instead of giving methe answer.
So this is study mode thatexists in all these tools.
So if you wanna learn somethinginstead of just getting the
(22:33):
answer, or if you wanna helpyour kids or people in your
department learn somethingwithout just getting them the
answer, that's a great way to dothat.
But then we get to the reallycool stuff.
So beyond all these things,they're just text and that are
very helpful and powerful.
You can do significantly moresophisticated things.
So I created three differentmedia applications that run
within the large language modelsthat can help.
(22:56):
In this particular case, kidslearn stuff, but you can do
exactly the same to teach peoplehow to do different things in
your organization.
So the first one I said, learnmath while playing soccer.
This is something I actuallydeveloped together with my
10-year-old son because when thesummer started, and he really
loves doing math, he asked me togive him math problems and.
Don't know how to give him mathproblems that will actually help
him.
So I said, what if we create agame that will give you math
(23:18):
problems?
Say, oh yeah, I love game.
Let's create a game.
I said, okay, what game wouldyou like to create?
And at that time, he wasactually playing a really low
quality from a graphicsperspective.
A penalty kick shootout game ona website that he likes to play.
And he said, what if we createsomething like this, but with
math problems?
Like, okay, that's a great idea.
And so we created this game.
So how does the game look likefor those of you who don't have
(23:39):
it, I'll actually gonna show youhow it looked like when it runs.
So when it runs, the first thingthat opens is a set of
instructions.
And the instructions say, whatis the game objectives?
Two players compete in a fiveround penalty kick shootout
using math scales, kicker inblue and goalie in orange, and
then how to play each round,both players answer a math
question, kickers answer first.
(24:01):
Then the goalie.
The penalty kick is resolvedbased on both answers after five
rounds.
The play with most goals orsaves wins.
So how does it work if both hitit right?
It's a 50 50 chance, it's arandom outcome.
If the kicker is right and thegoal is wrong, it's a goal.
If the kicker is wrong and thegoal is right, it's a save.
If both are wrong, it's a 50 50chance again.
And then it tells you what themath topics are and then what
(24:23):
the, how the to play the game.
And then you just click go.
And then what you see on thedisplay is that we have a
goalie.
And we have a kicker and we havea goal and we have fans and it's
all relatively low resolution.
But that's how he wanted it.
He wanted it to be a low qualityfrom a graphics perspective
because it's similar to the gamehe was playing.
And then there's a question onthe bottom.
What is 50% of what?
(24:43):
96.
So this particular case, theanswer is 98.
If I kick it says, oh yeah,that's correct.
Now it says, okay, what is 50%of 180 7?
That's the question that the,uh, that the goalie got.
And let's say that goalie getsit wrong.
So I would say it's 55 save, andit says wrong.
The answer is that he gives youthe right answer and the kicker
(25:03):
kicks and the ball goes into thegoal.
So now on the right side, I cansee the history round one
kicker.
What is the answer?
He got the right question, theright answer.
Goalie got it wrong.
And then the goal was scored.
And on top you can see thatthere's one goal round two of
five, and the goalie has zerosaves.
The game goes on and on.
After five rounds, it goes tothe next level and there's gonna
be tougher questions.
And the same on and on for levelthree and so on.
(25:25):
Why is this so powerful?
It's so powerful because it'ssomething I developed together
with my kid.
He was really excited creatingthe game because when we started
the game, it did not look likethis.
The first round of the gamewhere we just started working on
it, looked like this.
So I will show you.
We created it in Claude, and ifI'm gonna scroll all the way to
the top.
(25:45):
And show you the first round ofit.
It was very poor quality.
It did not have the history onthe side, but the concept was
the same.
The kicker was just a square.
The goalie was just a square.
But the concept was alreadythere.
And from that moment went on, wejust kept iterating and giving
it more and more instructionswith ideas.
So you can see how the display,those of you who are just
listening, I promise you, thedisplay just get better and
better.
It gets a little bit of a 3Dvibe.
(26:06):
And the people start lookinglike mini figures instead of
just a square, and then it addsthe actual kicking area.
And slowly we added more andmore stuff.
In the end, we added the fans inthe background, and made the
questions more complicated andmake it not repeat any questions
and added the history on theside.
So he was very excited aboutthis because he was able to
build his own game, and becauseof that, the whole family played
(26:27):
this game for about three days,trying to beat each other in
math questions, but even just inthe process of creating the
game, because every route we hadto solve the problems to see
what happens, then probablydoing, developing the game, he
probably solved.
I don't know, 50 to 80 mathquestions, which is 50 to 80
math questions more than hewould've solved otherwise in the
first week of his summer break.
And so again, you can dosomething very similar for your
(26:50):
employees and you can just findsomething that will excite them
as far as the process or createany other kind of application
that can help your employeesachieve and do different things.
And you can do it in any ofthese platforms.
The same exact thing can bedone.
Not just on Claude, but alsoChatGPT and Gemini.
So let's move on and look at thenext example.
The next example is me wantingto show that you could go beyond
(27:13):
just math.
So in this particular case itwas talking to Greek
philosophers.
So you can talk to Socrates,Play-Doh and Aristotle.
How did I do that?
So in this particular case,before playing the game, I wanna
show you how the game wascreated.
So the game was created becauseI did not know exactly where to
go.
So I actually consulted with theAI again in this particular
case, Claude, afterwards, I'llshow you another example, but I
(27:36):
started with what is covered inthe curriculum of social studies
for sixth to eighth grade.
And then he told me what iscovered in each and every one of
them.
And then I asked in the sixthgrade curriculum.
Includes any Greek scholars andthey said, yeah, we talk about
philosophers and scientists andso on.
And I said, okay, please createa detailed curriculum of sixth
grade through eighth grade.
Uh, need to know about Socrates,Plato, and Aristotle.
(27:57):
And I gave it to it as a deepresearch project.
So it's kind of the same in allof them.
There's a function that's calleddeep research, and when you
select that, instead of justgiving you an answer, it will go
and research.
Multiple websites.
In this particular case, you cansee 291 sources that it actually
read.
In order to provide me thatinformation, it took seven
minutes and 17 seconds, so a lotfaster than it will take me to
(28:20):
go through 291 sources, and itgave me a detailed review.
Of everything that I need toknow as far as what is taught in
six to eight grades.
And you can see that's a very,very, very long document.
It's probably 20 pages long.
And based on that, I ask it torecommend interactive games that
can develop and run inside ofClaude that would allow six to
eight graders to interact withthose Greek philosophers.
(28:40):
And then it gave me severaldifferent options.
I picked one option.
And I ask it to use voiceinstead of just text.
And we ended up with a game thatif I scroll all the way, well
actually, I don't need to scrollbecause I have the actual thing
here.
So if we look at the way thegame looks, uh, it asks me for
my name, I'll give it my name,ISAR, and then we refer to me
and it says, voice on.
I will keep the voice on andyou'll be able to hear what's
happening now.
(29:01):
Welcome to ancient Athens Za.
Choose a philosopher to beginyour quest for wisdom.
So I have Socrates, Plato, andAristotle.
Let's go with Plato.
Welcome future guardian ofwisdom.
Let us explore what lies beyondthe shadows of ignorance.
Are you ready to see the lightof truth?
Is what we see in virtualreality real?
How do we know what's trulyreal?
So now I can type my answer, Iguess, if we feel.
(29:27):
And see it in all senses.
It should be real.
Right?
Question mark.
Now Plato is thinking.
It's gonna give me his nextanswer.
Ah, my young friend.
But what if I made you feel asthough you were falling in a
dream so vivid that all yoursenses believed it completely?
Would that dream fall be realsimply because you felt it?
(29:49):
You see, our senses can be likeprisoners chained in a cave,
mistaking shadows on the wallfor reality itself when the true
forms of things exist in a realmbeyond what we merely touch and
see.
Now what you see here is areally cool combination of the
data from the deep research thatknows about Plato's Cave and the
shadows on the wall, which isstuff that he talks about a lot
combined with relating it to myanswer.
(30:11):
And the reason I developed thisparticular game in Claude is
because out of these threemodels, Claude, Chachi, and
Gemini.
The only one that can actuallyuse the Claudeed thinking brain,
meaning the actual largelanguage models, as it's running
code is Claudeed.
So really what's happening herebehind the scenes is when you
saw Plato thinking, quoteunquote, it's actually Claudeed
from its code in the artifactthat it created on its own.
(30:35):
Is calling the Claude API behindthe scenes to provide these
answers, which enable this verylively conversation.
Again, you can use this to trainyour people or allow them to get
any information about anythingyou want, whether it's how to
apply for a PTO as far as an HRquestion, or how to set up a
specific server in whatever ITenvironment or whatever the case
(30:57):
may be.
You can give it informationfrom.
Data that you give it or throughthe research process and then
allow it to think through it,review the information, and
provide you answers in a gamevoice conversation or any other
kind of solution that you wantin a few minutes of development
and having something that youcan share with your entire
employees.
I find this extremely valuableand extremely helpful, and
(31:19):
again.
99% of people don't know thatthat's possible.
So now you're one of a, so nowyou're a part of a really small,
unique group that actually knowsthat this is possible and can
take a benefit of it.
The third elastic example thatI've used is to show how you can
use a sophisticated graphicdisplay to help people
understand more advancedconcepts.
Now, some of you may know thisalready, but I was a F 16 pilot
(31:41):
back in the day many years ago,so I'm very well connected to
flight.
I was a flight instructor aswell.
I also have a private, licenseand so teaching people how
flight works is something I'mvery passionate about because
nobody thinks about it.
But when you board a 7 47, A 747, a takeoff is weighing more
than 400.
Metric tons.
(32:02):
That's over 800,000 pounds.
And we never ask ourself, howthe hell does this thing get
airborne?
That doesn't make any sense.
Well.
Simple physics make it possible.
But that simple physics actuallybreaks into four different
forces.
You have lift, weight, thrustdrag, which actually impact how
much lift an airplane cangenerate.
(32:22):
So what I did here is I createda mini simulators that allows
people to explore theseconcepts.
So it starts with the first pagewith just different ideas.
What are the four forces?
What is an angle of attack?
Uh, what is air speed altitudeand how these things balance.
But then the cool thing isthere's an experiment, but there
is an experimentation section,and in that section on the
(32:43):
right, you see an actual diagramof the wing and you can see the
air flowing much faster on topof the wing than it is flowing
on the bottom of the wing, whichis the whole trick because high
speed generates low pressure,which then basically pulls the
airplane up.
But the other thing on the leftside, there are sliders.
If I change the sliders as anexample of angle of attack, you
can see the wing rotating andyou can see two things that are
happening.
(33:03):
It's gonna tell us that from alevel flight we're now climbing
and it says current situationlift is greater.
Then wait, so the plane ascendsagainst gravity.
Cool.
Right.
The other thing that you can seethat it starts showing.
These turbulence behind thewing.
So if I keep increasing theangle of attack, eventually you
sit.
Now it's going to descend.
Why?
Because the plane is notgenerating enough lift to
(33:24):
overcome the weight, and I keepon going.
It says too much turbulence format a high angle of attack.
Lift decreases below weight anda stall can occur.
This is really cool.
Now the other thing, if I takethis down back to a normal angle
of attack and I increase thealtitude.
Then you can see another thingthat's happening.
The lift is going to decreasebecause now the air is thinner.
(33:44):
It's gonna tell me that currentsituation, the plane is not
generating enough lift toovercome the weight.
Higher altitude means thinnerair, reducing air density, and
therefore reducing lift.
So I can play around with thesliders and see how that impacts
the actual lift on the wing.
That's very cool.
But again, you can take this toanything in business that you
can explain in a graphical wayand create it in a visual way
(34:04):
that will help people interactwith it and better understand
how it works.
This is not rocket scienceeither.
Going back, this was created inChatGPT and not in Claude.
Just to show you that ChatGPTcan do the same thing.
By the way, Gemini can do thesame thing as well, and if I
scroll back all the way to thetop, you can see what I asked it
for.
I said, I am a science teacherin the eighth grade, and we are
learning about the physics ofelevation in airplanes and so on
(34:25):
and so forth.
I asked it for ideas on how wecould do that.
It asked me several differentquestions because I asked it to
ask me question.
I said, before you startprogramming this, uh, please let
me know if you have anyquestions.
So it asked me clarificationquestions about the level of
physics I want to go into,whether, what kind of aircraft I
want, what kind of units I wantto use, and so on and so forth.
And then we together, afteranswering the question, got to
(34:45):
the conclusion what we wanna do.
Version one was really ugly andnot very helpful, and I iterated
back and forth until we got tothe version you just saw.
So what does that mean to eachand every one of you in your
business?
It means that all these tools,chat, bt, Gemini, clo, et
cetera, can do a lot more thanmost people know that they can
do.
All you have to do is know howto use the tools.
So, several different things wecovered today.
(35:06):
One is using a prompt library tosave your better prompts.
Two is learning how to usereference material in a way that
will dramatically increase thechances of getting the correct
answer, and that will help youverify the information much
faster.
We looked at Notebook LM as away to summarize lots of
information and get a lot ofdifferent outputs out of it.
We looked at study mode insideof the models that allows you to
(35:27):
learn stuff instead of just getan answer, and we looked at
creating interactive.
Dashboards, learningenvironments, call them,
whatever you wanna call them,that you can use across multiple
aspects of your business andpersonal life in order to help
people around you, whether it'sstudents, your kids, yourself,
your employees learn thingsbetter and faster.
I really hope you found thishighly impactful.
(35:48):
I really hope that you're gonnatest and play with this
preferably today, so you don'tforget, uh, any of the things
that we talked about and sharethis with other people.
This is, again, a highly.
Tactical episode, these are theones that I love the most, and
you can share it with otherpeople that can learn how to do
these kind of things and developit for things they are
interested in creating orteaching again, either in
business or in school or intheir personal life.
(36:10):
If you found this episodehelpful, please subscribe to the
podcast.
We really appreciate all thepeople who are subscribed.
There's tens of thousands ofyou, but every single person.
But if you are not one of them,then please join the people who
are already subscribe, share itwith other people who can
benefit from it.
And if you really in a goodmood, then give us a five star
rating in a short description ofwhy you like this podcast on
(36:32):
Apple Podcast and or Spotify.
And until next time, have anamazing rest of your week.
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.