ChatGPT is saving instructional
designers a ton of time, and that's why in this video I'm going to show
you what it is and how to use it so that you can make a much bigger
impact as an instructional designer. So if you're new to ChatGPT, then
you are in for a treat because you are about to see the magic
that this tool is capable of. I have a lot of examples lined up for
you where we can see how we're going to use this for storyboarding, idea
generation and a lot of other stuff. And I'm even going to get into
some advanced use cases at the end for those of you who have
been using ChatGPT for a while. So no matter where you're at with
this tool, there's going to be some good stuff for you here, and I would
say, If you haven't used ChatGPT at all, then go to chat.openai.com. I'll share that link in the
description and create an account and maybe even play around with ChatGPT
while you're watching this video. This is a very good tool to get
on board with, and like I said, there is massive time-saving
potential with many common ID tasks. So let's just dive right into it. So what is ChatGPT? One way to think about it is as a
very advanced form of auto complete. So like when you're texting on
your phone or using some of these new Google apps, like Google Docs
or Google Slides, you'll notice it starts completing the words for you. ChatGPT is kind of like that, but
times a million because it's trained on a vast, vast, vast amount of data,
kind of like the entire internet worth of data, and it is really smart at
understanding the context of what you're asking for and what you want from it. But why it's like auto complete is
it basically guesses the probability of the next word it should generate,
and it can generate tons and tons of words that all make sense
contextually, like in that fashion. So it's kind of like auto
complete, but really, really smart. And again, trained on a ton of data. So it is a smart cookie to say the least. So let's just look at it
in action and let's look. This is ChatGPT here. So again, create an account if
you haven't already, you need to get hands on with this. But let's see what ChatGPT has
to say ChatGPTabout ChatGPT. So we can say, 'Hey, ChatGPTexplain
ChatGPT to me like I'm five years old', and then here we go. Sure thing, kiddo. So this is all again, artificially
intelligently generated. T this is AI if you're brand
new to this, but I realize this description was kind of too long. I'm not going to read
this description to you. It wasn't what I wanted. So it's kind of like a chat conversation. So I say, 'try again, but explain
it to a professional instructional designer and use an analogy that
they would be familiar with. Your language should still be concise and
conversational, and your total response should be no longer than one paragraph.' So you can see how I can give
it some more details to get closer to what I'm looking for. So it's like a nice little
virtual assistant in that way. And then ChatGPT says, 'absolutely'. You can think of this as an
interactive digital textbook. Much like how I use multimedia. And again, you can pause and read
this if you want, but this was still too instructional designy for me. I'm like, this isn't exactly
what we're going for. So I said, 'try again, but explain
it to an average adult instead of a professional instructional designer.'
And this is what we're going for. You can think of ChatGPT as a
digital -- pen pal - can't read today - that's incredibly well read. It's like having a conversation
with someone who has read a vast library of information, books,
articles, websites, you name it. You can ask it questions, have it explain
concepts to you, or even write stories. So, there you go. You can read the rest of this
if you'd like, but that is a good way to think about it. This really smart friend or assistant,
more like an assistant for professional purposes, who has tons of information
at its disposal, and it, again, is really, really smart and near instant. Again, I've generated this stuff
for you so you don't have to watch me typing stuff out for this
demo, but play around with this. This stuff gets generated within seconds. It's not like you need to wait around
for a response, so very impressive. All right, so how do we use ChatGPT? So you've got a little demo of some of
the techniques I've been using there, but I'm just going to share some here. And then when we look at more examples,
you'll see all of this in practice. So we want to tell
ChatGPT what its role is. So you are a professional instructional
designer who's going to help me write a storyboard, for example. You want to make it really clear what
you're asking ChatGPT for in that moment. You can also tell it to ask you
questions if it needs more information. And I'm going to just go back into
the examples because I think that's where the real good stuff is. So, if you are brand new to ChatGPT,
notice on the left here, this is just one of these conversations. You can have many of these different
conversations open at once, and the different conversations
will hold their own context. So obviously as I try again, It
knows what all these previous messages said so that it knows
what I'm actually asking it to do. If I were to, say this same message
in a different thread, I'm going to get something very different. So it is contextual in that way. All of the previous messages help
inform what the next response will be. So it can keep up with the
conversation in that way. So I have all these different ones open. I opened them in tabs up here. I'm going to go to this conversation
where I wanted ChatGPT to help me generate a storyboard. So notice here I give it this role: 'You
are a corporate instructional designer who needs to create a course on instructional
design best practices,' right? We tell it its role and we tell
it what our goal is; creating a course on ID best practices. 'Generate the first page of a storyboard
for this course in a table format'. So notice I'm being precise there, not
just generate a storyboard, generate the first page of a storyboard. So we want to be nice and
specific, in a table format. 'The columns should
include...', so on and so forth. 'The storyboard will be given to a
developer who will develop the tool to develop the course,' it should
say, 'in Articulate Storyline 360. So programming notes
should work in that tool'. So notice it starts trying to develop
this storyboard, but then, in this cell here, it starts listing these bullet
points and it breaks the table format. Not sure why I did that. We obviously don't want it to do that. So then I send a message, 'try
again, but do not use bullet points because they broke the table format'. So you can talk to it
like a person in that way. It will understand what
you're asking it for. And then, how nice. ChatGPT apologized, and then now we
get this in an actual table format. So let's look at how this is. So we have the title of the slide. We have some content on the screen. It's welcoming us to the course. And then visual notes, logo of the
company, animated arrow pointing to the next button, and then some programming
notes about colors, font, error animation. This isn't perfect. This definitely is not perfect. I would not suggest using a
simple prompt like this to generate an entire storyboard. It's probably not going to go over well,
but this is just at a base version. Not terrible, right? Maybe you can get some ideas from this. Let's look at another slide. 'What is instructional design? Visual notes: image of a brain with
gears working inside, animate the gears turning.' Kind of silly, kind of
rudimentary, but still interesting to see. For slide four on the importance of
instructional design, it suggests using icons 'to represent efficiency,
effectiveness, and appeal'. So I just want to show you another
example of how you can keep asking follow-up questions to get more specific. You can say, or I said for slide
four, 'what should the icons look like?' That might not be clear to me. What is an icon representing efficiency? So then it tells you, 'the efficiency
one should be a clock or stopwatch with motion lines indicating fast speed. For effectiveness, it should be a
bullseye or target.' So again, this is just one example, but you can see how
this kind of thing can start being useful in an instructional design context. So I'm going to hop back over here. You can also tell it which tone to use. I think you saw me do
that in the first example. 'Use a conversational, concise
tone.' That's generally what I'm telling ChatGPT to use. And you can give it parameters. So again, you saw that in
the first example, right? Like, 'do this, the total response
should be no longer the one paragraph'. So parameters like that really help
because you don't want to get stuff that's super long, especially if you're using
this to save time, and especially when it comes to summaries and things like that. You also can give it context. So again, giving it context around
the goal, around the audience that you're serving, and so on and so forth. The more information that you include with
the prompt, generally speaking, the better ChatGPT will be able to give you the
kind of response that you're looking for. If you just give it a one sentence
really basic prompt, it's going to be a pretty generic output that you get. And this is one of my
favorite things to do. Actually, no, we're going to get
there soon, but ask it to try again if it gets something wrong
or if you need more information. And you've already seen me do
that a lot in these examples. If it doesn't, if you type in a
prompt and it gives you something and you're like, 'oh, this isn't great. Like, this tool is lame. I'm just going to go do something else'. You can keep the conversation
going, tell it what you need. You're basically training ChatGPT
in a sense, in each one of these threads that you create. You're training it to respond
to what you're looking for each new time you start a new chat. So don't be afraid to do that. So I want to show you how we can use
these ChatGPT best practices to take this storyboard example forward and get closer
and closer to what we're looking for. And again, I think these practices
alone will help you go much further in your ChatGPT journey. So here is the next prompt I started with. So I started a new chat. And I said, 'you are a corporate
instructional designer who needs to create a course on ID best practices. Your audience includes new
IDs at a large tech company. The course should take
15 minutes to complete. The output should be in table format'. So you can see here,
I'm telling it its role. I'm telling it what our goal
is to create this course. I'm telling it who the audience is and
where they work, and I'm giving some other parameters and context: 'the course
should take 15 minutes to complete. The output should be in a table. The columns should include...'
So you notice we're getting very, very specific here. The storyboard will be
given to a developer, right? You're already familiar with this, but
I also give some other details: 'ensure that any interactivity is intentional'. In other words, it helps the learner
practice and assess the knowledge and skills that they're learning. Also, 'ensure that none of the
animations or visuals are distracting'. So I'm trying to teach it, 'Hey, here's
how to be a good instructional designer'. 'Follow Mayer's multimedia principles
when generating the storyboard. And then finally, take your time,
work on it step by step, and ask me any questions in case you need more
information.' I've seen this shared as a best practice, telling it to take its
time and work through this piecemeal. You can do some additional
research on that if you like, like some prompting best practices. Prompt engineering is the term that
everyone's using for this type of thing. But I just sent a lot at it
to see what it came up with. And I would say it's a little
better than the first one, right? We have nice content on
the screen listed here. We have this narration and voiceover
and some visual notes: 'gentle fade in of the logo and text. Bullet points appear one by
one as they're narrated'. Look at this: 'use Mayer's
signaling principle. Highlight each phase as it is
discussed.' So, it can reference a subject matter expertise, like it
can apply some of these principles. And this honestly isn't bad, again,
feel free to pause and read this or just try a prompt like that for
yourself and see what it comes up with. But you're probably not just
going to bring this into the Storyline and develop it. Another example: 'applying ADDIE
scenario text and options for response'. That's not really giving us a lot. What is the text? What are the options for response? And as I've shown you, you can of course
ask questions about this to get those questions and options in response,
and we'll look at that in my favorite approach for working on stuff like this. This third example, check this prompt out. 'You are a corporate instructional
design consultant who will help me create a storyboard for new instructional
designers who need to learn how to apply ID best practices to create
effective learning experiences.' So pretty specific with the role I'm giving
it and the context I'm giving of how it will help me and what our goal is. 'I need this course to be creative
and engaging, and I'll develop it using Articulate Storyline 360. Can you please ask me questions to
ensure that you have the necessary information to help me write a development
ready storyboard?' So this is one of my favorite things to do to make sure
that I'm giving ChatGPT what it needs so that it can give me what I need. So this right here, 'can you please
ask me questions to ensure you have the necessary info?' That right there
is going to really change the quality of what it is that you're getting out. So here we go. 'Absolutely. I'm happy to help you. Here are the questions I need
to ask.' So look at all this. 'What are the main learning
outcomes you'd like? Can you tell me about the
background of the learners? What topics will be covered? How long is the course?' So
this list goes on and on and on. A little weird thing as it goes,
like '7, 8, 9, 0, 1, 2, 3'. That's okay. I'm obviously not going to read you
all of this, but then I take the time to answer each one of these questions. I gave it some very basic
learning objectives. I gave it some basic background on
the audience, and I went through and I answered every single one of these
questions that it asked me just by following the format it gave me, even
though again, it's '9 0 1 2 3', but it understands based on the pattern. And then it says, 'thanks. I'll help you grab the
storyboard,' and here we go. So now it's much more custom to our
specific needs and goals with this course. So you can see here. Scene by scene. I'm not going to read it, but it is
more personalized to our audience. You can check it out if you want, but
it's still at a kind of high level. You can notice, 'slide 3.1,
overview of the ADDIE model. Onscreen text and diagram'. It doesn't tell you what that actually is. This is more of an outline,
a higher level kind of thing. Look at this scene for
Gagne's nine events. 'SIide 4.1, overview, 4.2 to 4.1. Detailed explanation of each event,
scenario based learning interactive scenarios where they can apply each
event in a real world context.' And it's saying that because that's
one of the strategies I said it wanted to use, if you go ahead and
pause and read this if you want to. But, so then what I did next, just to
show you, 'can you generate a detailed storyboard in table format four slides,
4.2 to 4.5, and these are like the Gagne's nine events slides I just mentioned. Columns should include slide number
on screen, elements, narration, visual notes, and programming notes'. And then it says, 'sure, here you go'. So you can see that when
we - well, let's see. So onscreen elements, we have the
text description of event one. We have this narration
about Gagne's, first event. And then, 'illustrate with
attention grabbing iconography. Button to be clearly highlighted'. And programming notes: 'button triggers
a layer or popup where a scenario related to gain attention is presented'. And you can see how you can keep asking
questions and getting more and more in depth with what you're going for. And I did that. I said, 'try again, but
make it more specific. Include the exact text that should be
included on the screen and explain what each icon or visual should look like'. So here we go. Now we have the exact text that
we could essentially be copying and pasting onto the screen. And then for visual notes instead
of, 'oh, use a relevant icon,' it says, 'use an icon of a light
bulb or an exclamation mark to signify the gaining of attention'. So again, this is fairly rudimentary
and I'd probably take this further and say, 'hey, we don't want all of
this text on the screen and narration. We want to rely more
on visuals than text'. But like I said, you're essentially
training these different ChatGPT chats based on your needs and your expertise. So it's not replacing you in that way. You do need to be a competent
instructional designer because you are babysitting this tool essentially,
and just helping outsource some of the busy work or idea generation. That might typically take you a long
time, but you still need your expertise. Yeah, that's still important, I would say. So hopefully those examples
are pretty helpful. We're goning to move on to some
other use cases here, but again, it's not perfect, but hopefully
this helps you see how you can use a tool like this for common ID tasks. All right, so here are the list of
use cases as I've seen them, and if you have other ways you've been using
it, please share them with us in the comments because we're all looking
for ways to use this to help us save time and be more effective as IDs. So, research and ideation. I think that is a great way to use
ChatGPT, especially because of some of the limitations we'll get into a bit later. Course design, of course, right? Writing, learning objectives,
storyboards, all of that good stuff. Job searching and portfolio
work and communication. And we're going to take a deeper
look into each one of these four primary categories of use cases. All right. So research and ideation. You can essentially ask ChatGPT
about topics you'd like to learn more about, similar to
how you'd be Googling something. But ChatGPT is nice because you
can really have it dumb things down for you if you want it to. And you already saw that example with
the ChatGPT explanation, how we went from explaining it like I'm five to
explaining to me like an average adult. Here's another example with Kubernetes,
and I can't even explain this to you. It's just a complex topic I've
come across in one of the courses I was developing one time. So I said, 'can you explain it to me like
I'm in eighth grade by using an analogy? Please respond in one paragraph or less'. And it gives an analogy about
controlling a bunch of toy cars. Kubernetes is a smart
traffic cop for these cars. And I'm like, okay, that's nice,
but I feel like that's too simple. I still don't fully know what this
is in a useful way for designing a course with that analogy. I said, 'can you write a poem about
Kubernetes to help me understand?'. And this is really just a demo
that you can do this kind of stuff. You can say, 'write me a song by Eminem
about Kubernetes', and then we'll do that and it will sound like an Eminem song. So try, you can even try asking
it to write songs or do things in styles of artists that you like. And it's pretty amazing
what will come out. So, I'm not going to read this poem
to you, obviously, but it rhymes. It's well done. It can be quite creative
on those kind of tasks. I said, 'now explain it to me like
I'm an average adult', because that still wasn't really helping me. And this is too technical, right? 'Open source container orchestration
platform.' I don't know what that is. I don't have that background. So I said, 'try again, but explain
it to me like I know nothing about software development'. And even if you do, you can use
prompts like this to get it to simplify things and try different ways and,
this explanation was a lot better. 'You have a bunch of different tasks you
need to complete, but they're complex. Instead of doing each task manually,
you have a smart assistant called Kubernetes to help you out. It's a personal organizer, project manager
for your tasks.' And I'm not going to read this to you, but I just want to show you
what a workflow like this might look like where you are trying to make sense of a
topic, especially if you're designing a course on some kind of technical software
or engineering topic and you're new to it. Of course, lean on your
subject matter experts. Definitely do not rely on what
you're getting from ChatGPT alone because it can hallucinate or give
you some inaccurate information. We're going to talk more about that
when we get to limitations, but it can help a lot, make sense of stuff
even -- we'll look at some more, but sorry, I'm obviously pretty excited
about the ChatGPT and AI stuff in general, so I'm getting ahead of myself. Another way to use this for research
and ideation is to ask for ideas about projects you're working on
or problems you're trying to solve. I love using ChatGPT for this. It's honestly probably
one of the best use cases. And look at how awesome this is. So, I told ChatGPT, 'you are an
instructional design consultant who specializes in designing engaging,
award-winning e-learning experiences. I'm working on a course about
cybersecurity and I need your help making it really wow my audience. Can you list five suggestions
about how I can make the e-learning experience more unique and engaging?'
It gives me some suggestions. I'm not goning to go through them
all, but basically, gamification, interactive scenarios, rich
multimedia, so on and so forth. So I was interested in
the gamification piece. And it mentioned some examples,
quizzes, challenges, leaderboards, rewards ,batches, not very
helpful in and of itself. But then I say, 'can you list five
relevant ways that I could gamify the learning experience by giving specific
examples that relate to the subject matter of my course?' So again, I hope you see
I'm using some of those strategies, right? I want it to be more specific. I want more context. I'm saying 'list five ways' instead
of just 'tell me some ways' because I want to give it some of those
parameters, and then here we go. 'Certainly. Here are five ways'. And these ideas are honestly
pretty awesome if you ask me. Like, 'cybersecurity challenge levels. Divide your course into different
levels or missions, each representing a specific cybersecurity concept or skill. As they go through, they unlock new
content and face more complex challenges'. 'Virtual hacker hunt. Create a virtual scavenger hunt
where learners have to find hidden vulnerabilities or security flaws
in a simulated environment.' I mean, that sounds fun. 'Security quests and badges. Write a series of quests or missions
related to different cybersecurity topics', and you get a virtual
badge for each achievement. Honestly, not a bad idea. 'Cybersecurity escape room' and
'cybersecurity trivia showdown'. So feel free to pause this if you want,
or again, try prompts like this for yourself, but, pretty impressive, right? Like the idea generation stuff is
pretty sweet, and if you're not going to use it for anything else, this is at
least a good way to use it if you want to up your game as an instructional
designer and design more engaging stuff. Pretty good for that. I was excited when I
saw it, as you can tell. And then another way to use it
for research and ideation is to ask ChatGPT to summarize content. So this is where I was getting
ahead of myself, right? ChatGPT can hallucinate information
sometimes, but one way to help that is to feed it the source
information you want it to work with. So, for example, here I just copied
some content from this Wikipedia article about instructional design. And at first I just copied and pasted
the entire article and I was like, 'I want you to summarize this'. But then it comes back and
says, 'hey, this is too long. We can't work with that'. And we'll get into some of the limitations
around length a bit later in the video, but I essentially wound up having
to just copy this models section. So I said to ChatGPT, 'can you
please summarize the following article in two paragraphs or less? Your tone should be conversational
and your word choice should be suitable for an eighth grade student. Here is the article', and then I
just pasted it in directly from Wikipedia, and then here we go. We get a summary, right? The ADDIE model, here's what it stands
for, here's what we do in each phase. As you can tell, it's not two paragraphs. So sometimes ChatGPT just doesn't
want to listen to what you're asking for it, and you have to
keep working with it in that way. So I said, 'please try again, but
restrict your summary to one paragraph.' So, it discusses instructional
design models, including ADDIE, which it gets into what ADDIE does. Another one is rapid prototyping,
and it gives a brief explanation. It essentially did
summarize this pretty well. I have a nice sentence about each one
of these models, and I can do that instead of reading this entire thing. So that is just a quick example. Obviously you're going to be summarizing
stuff that's a lot longer than this piece that I copied and pasted, but you
can imagine how, as an instructional designer, if you're given a PDF or you're
given some PowerPoints that don't make a lot of sense to you, you can put some
of that content into ChatGPT and get summaries or get it rewritten in ways
that are easier for you to process or understand, and maybe even ways that
you can then present it to your learner. Of course, there are some potential
privacy and security risks with this, which we will get to in limitations. Maybe I should have started with
limitations since it's coming up so much. We're moving along. So you can also use
ChatGPT for course design. And this is some of the biggest
time-saving potential I would say. So you can use it to generate
learning objectives and we of course have to take a look at that. So here's the prompt. 'You are an instructional designer
with 20 years of experience and a PhD in instructional systems and learning
technologies.' This is important, and I'm going to give some context here. ChatGPT by default is not very good
at generating learning objectives. I've tried it on my own and they want
to say things like, 'oh yeah, understand this, or learn this or know this'. Luke Hobson did some content about this
too, and was dealing with similar issues. So, ChatGPT doesn't really write learning
objectives like an instructional designer should be writing learning objectives,
because you learn in instructional design 101 to avoid those vague, fluffy verbs
when it comes to learning objectives. So I've seen people mentioning different
strategies for how they go about it. I'll link to some other content and
resources from other creators who have been working on this ChatGPT in ID stuff
in the description, so you can check it out, but I figured I would take a shot
at using an approach like this just with being very well-defined about the role,
because an instructional designer with that experience and a PhD is not going
to make those very amateur mistakes when it comes to learning objectives. So that was my hypothesis. And then I said, 'I need your help writing
learning objectives for a course that I'm developing on ID best practices. Please list three to five learning
objectives that would be suitable'. So it said, 'certainly'. Sure, yeah. Here we go. It was too much, right? I don't want to read all of this. I want this to save me time. I'm not trying to make sense
of whatever format this used. So here's another potentially
more advanced technique, I guess you can call it. I said 'try again, but
use the following format. Each learning objective should
replace an X in the following format, and each learning objective should
be no longer than one sentence'. So I'm essentially telling you,
I want you to use this format. I'm sure there are more concise ways
to put this, but I just want it to be very clear to the AI that I want you
to replace these Xs with one sentence along learning objectives, and they
should be completing this stem. This is much more aligned with
how we would write them as IDs. And then I even gave an example. So this is something you can do
and something that will help a lot in the quality of the output
that you get from ChatGPT. So I said, 'for example, an appropriate
response may look something like this: By the end of the course, learners will be
able to...', and then notice I replaced those X's with simple learning objectives. Straightforward, one
sentence learning objectives. All right, so ChatGPT apologized
and then gave the list of revised learning objectives. 'By the end of this course, learners will
be able to analyze and evaluate different ID models for their suitability in
diverse learning contexts, conduct a needs assessment design, and develop effective
learning materials', and the list goes on. This should be broken down further. Again, this is a very
big learning objective. This could be a learning goal for an
entire learning program, but I would hope, with some of these strategies that we've
been discussing, you can see how we could break each one of these down further and
further and further to get closer and closer to what it is that we're looking
for and get more information from ChatGPT. All right. We can also use it to
create course outlines. Again, that's pretty similar to
what we've already been looking at. We can use it to generate
learning activity ideas. So similar to how we use that idea
generation example, you can do that for learning activities, role-playing
exercises, so on and so forth. You can write content for
your storyboards and scripts. Again, huge time saver. If you get a good conversation going
where ChatGPT is giving you the type of stuff that you're looking for,
you can use it to write character dialogue and case study details. I'm not going to go into examples of
all of this stuff because it's going to get quite long and some of this stuff
will be a bit repetitive, but using the strategies that you've already seen me
covering, I hope will give you enough to start experimenting with these
types of things, and you can use it to generate good multiple choice questions. This is a big one. This one is good, especially because
some people have a hard time writing good distractors or just writing good
multiple choice questions in general. So I think I'm going to do an entire
video on how to use ChatGPT for writing good multiple choice questions, so I'm
not going to dive too deep into it here. That one will be its own video,
just like some of the other stuff we're going to look at. But that is a great way to use ChatGPT. All right, moving right along to job
searching so you can use it to revise your resume to brand yourself as an ID and
present your experience in the best light. Joanna's going to do a full video on this
and the channel, but you can imagine, if you've seen any of the resume content on
this channel, we want to use it to help rewrite your resume to make it use more
instructional design language and less education language, for example, if you're
coming from education or just to highlight your achievements or your responsibilities
in a better, more concise light. So there's a lot of different ways
it can help you with your resume. You can also use it to generate
custom cover letter letters for each role you apply to. I will say be careful with this. We're going to get into one of the
limitations a little bit later, but, we don't want people just throwing
away your materials because it's so obviously written by ChatGPT. And if you follow the suggestions I use
here, and if you have the conversation, you get stuff more and more concise. You're not goint to have that problem. Some of the stuff you won't be able to
tell is were in by ChatGPT if you're using your human mind to guide it
and your human critical thinking and attention to detail and all of that stuff. That's still necessary when
you're working with this tool. It's just a tool. If you're just using ChatGPT to say,
'oh, generate me a cover letter', it's not going to be very good. You need to give it a lot more if you
want something more unique and custom and personal out so that it's not, 'oh yeah,
this is obviously generated by a robot'. And to generate and
revise portfolio content. So this can be literally outlining
your entire portfolio homepage. This can be coming up
with your hero title text. 'I design and develop engaging
e-learning', like those types of things. You can use it to develop something
much more engaging or something much more personal to you, I should say,
if you feed it the right prompts. And you can use it for your portfolio
about page, for your process writeups, even for portfolio project ideas. And Nicole is going to do an entire
video on how you can use artificial intelligence, including ChatGPT,
and a host of other tools to create a really good instructional
designer e-learning portfolio. So that's coming next month. Definitely subscribe because again,
every week we're going to be sending more of this AI in ID content your
way, and this is just the beginning, so definitely subscribe if you're interested. And then for communication, I'm sure you
can imagine how this would be helpful too. You can use ChatGPT to write
professional emails to subject matter experts, stakeholders, hiring
managers, clients, colleagues. It can help you set boundaries in
a professional tone in case like you're feeling more emotional
or unsure about something. You can adjust the tone of things either
to make it more formal and professional or make it more casual and conversational
if you want to go more in that direction. It's really good at rewriting stuff. And that can be helpful
in course design too. Of course, if you want to use a more
conversational tone in your courses, you can have it literally rewrite
entire sections of content for you. And it does a really good job at that. This feels relevant, I was talking to one of my friends this
week about ChatGPT and asking if he was using it at all in his industry. And he's like, 'all the time. I use it to communicate with some
of our clients and executives at those companies and stuff. And my supervisors are being CC'd
on those messages and seeing those emails and they're like, wow,
you are so good at wording stuff. You are really on it with. Yeah, your phrasing, can you actually
start drafting some of our emails that we need to send to these other
executives and CEOs and stuff?' So I just found that hilarious. My friend is getting praised
for his communication, but he's literally using ChatGPT to generate
everything that he's doing. He doesn't even identify as a
great writer or anything like that. So pretty cool how people are
using these tools to stand out and do better at their jobs. All right, so you've heard me mentioning
these limitations and risks quite a bit. Now let's get into them. So if you've been looking into
ChatGPT much you might've heard of the context length or the context
window of these different models. And what it means is essentially
how many words the model can handle in any given request. So it's in terms of tokens and a simple
way to think about it is one token is about three quarters of a word. So the context length for these
ChatGPT models is between 3 and 24,000 words, depending on which
version of ChatGPT you're using. And I should probably show you that. So let me actually pull that up here. So when you go to create a new
chat in ChatGPT, as of when I'm recording this video, you can either
use ChatGPT 3.5 Turbo or GPT four. This one costs 10 or 20 bucks
a month or something like that, and it's more limited. You can only use 25 messages every three
hours, but it is much more advanced. GpT 3.5 is less advanced, and by extension
it has a lower context window, so it's more around that 3000 word mark. And then GPT four, I don't know what
it is right now, they're in the process of expanding it, but at most 24,000
words, but I don't think they've unlocked that for public in ChatGPT. The other difference is, whether
or not you're using the API for like your own app, you can get a bigger
context window in your own app than using ChatGPT here in the browser. Now why is that a problem? That's important. When you are talking to ChatGPT,
when you have a message thread like this, your message and ChatGPT's
response all go into that word limit. And then if you send another message,
it's going to include this message. And the previous one. And the previous one. Okay. So as you go down the thread, ChatGPT is
trying to hold all of this information, all of these words in its context window. So it's not just like, oh,
every message you send has to be less than than 3000 words. It's the entire conversation has to
fit in fewer than 3000 to 24,000 words, depending on a variety of those factors. So you can imagine that's
a pretty big limitation. If I want to send it a 30 page pdf
and then start asking it a bunch of questions and asking it to generate
a storyboard based on that pdf, I'm not going to have enough of a context
window to do something like that. I'm going to run out of space. And what that looks like when in a
chat conversation, like in ChatGPT for example, is once it's past that context
window limit, anything before that is just going to be getting cut off. So you might say in the beginning,
'here's what your goal is. Here's what I want you to help me do'. But if you send it too many messages, it's
going to just forget what that goal was and it's going to start just immediately
becoming much less helpful or useful. So that's something to keep in
mind, that it's a limitation. The good news is researchers have
identified how they can create a language model like this with
1 million token context window. So we're talking about going from 4,000
to 32,000 tokens to over 1 million tokens. And that's where it's obviously
going to be much more powerful and just being able to feed it a ton of
information, ask it a ton of questions, and it has that context for every
single question that you're asking it. So again, this is still early days. This stuff is happening fast, but
the research is very promising, and this will be improving in
the months and years to come. But just keep in mind,
that's the limitation now. Inaccurate info, right? You can probably see this, if we
look at the bottom of ChatGPT, there's a warning at the very
bottom here, really small font. 'ChatGPT may produce
inaccurate information about people, places, or facts. 'And it does that sometimes. It's commonly referred to as
ChatGPT just hallucinating things. That can happen. So take what you do get from ChatGPT
with a grain of salt if you're relying on factual information. That's something that they're working
on, but right now you can't just - it's not like Google where a lot of people
look at stuff on Google and it's like, 'oh yeah, this is the truth'. ChatGPT take with even more grain
of salt than what you see on Google if you're relying on information. And then inconsistency. For the ChatGPT models that we have
access to, not only can the responses vary from thread to thread, it might look
a little bit different and that's okay. That's kind of built into the
tool, but they're also working on the model in the backend. So for example, in the recent weeks or
past month or so, people are saying, 'GPT 4 has become so much stupider'. Essentially they're like, 'it used to be
so much better, but whatever they're doing on the backend, whatever updates they're
doing has made it so much less useful than it was when it first came out'. I don't know if this is fully
accurate, but a lot of people say this, and that's just something too. This stuff is still new. It's still, I don't know if it's
technically in beta, but it might operate differently one month
than it does the previous month. And that changes once
you start using the API. But at least for this public ChatGPT
interface that they make really accessible and really easy to use, it can be
inconsistent depending on when you use it. And then again, I mentioned this as
a limitation or risk just because this is really not a good look. You don't want whatever you're
generating with ChatGPT to be obviously generated by ChatGPT. Again, you can resolve that by using a lot
of the strategies I shared in this video. But I recently had a job application
for my company and a big focus of it was using tools like this and 50% or more
of the responses to a certain question about generating something in my voice, it
was all essentially the same, and it was pretty obvious that everyone used ChatGPT
for, it's like a human would not say this. Someone just typed in a prompt to
ChatGPT, and then just copied and pasted it into the job application. I've heard stories too
about other hiring managers. Certain cover letters, if it's
obviously generated by ChatGPT, it's thrown into the trash. So use your critical thinking thinking. Please don't just copy
and paste this stuff. And again, use these techniques
we've shared to make the output more personalized and more in line with stuff
you would actually say and stand behind. And we'll get into that more in
later videos about using ChatGPT to find jobs and stuff like that. And then I'm not going to get too deep
into this because this could of course be an hour long video in and of itself. But this is the wild west in a sense. There are regulatory concerns. There are security concerns. We don't have a lot of case law
to deal with this stuff yet. We're not really sure how AI is
going to be getting regulated. We're not sure what the copyright
and ownership situation will look like in the months and years to come. Right now, for the most part, a lot
of these tools, what you generate from the tool, it's yours to keep. Whether or not that's going to change
in the future, we'll have to see. And then a lot of companies are hesitant
to use tools like this because of course, they don't want to be uploading
their private, sensitive documents to servers that aren't fully vetted
by their companies and their security teams and all of that good stuff. So obviously use this at your own risk. Don't do anything that your company's
going to be coming after you about. But again, if it's pretty safe, if you're
using it for lgeneral idea generation and not sharing confidential information, I
would say, at least to my understanding. AlRight. So those are the limitations. If you can think of more or you have
some insight into any of these, feel free to let us know in the comments
because there are indeed limitations. This is still early days. So just to summarize these steps,
break things down into smaller tasks. So it was pretty ambitious saying,
'just generate me a storyboard'. But we could probably see much better
results if we could say, 'hey, help me get clarity on my learning goal now. Help me identify some learning
objectives for this specific course. Now help me generate a course outline and
then now help me generate the storyboard'. So if you go step by step by step like
that, you're going to see much better results than 'generate me a storyboard',
and it gives you a chance to revise things and refine things as you go to
make sure that you are in alignment with the ai before having it try to
spit out the final product, so to speak. Be clear about what you're looking for. Again, there's a lot of different
ways we can do that, but generally speaking, more details are better. I hope the examples made that clear. And don't be afraid to keep
the conversation going. Ask for what you want. Treat it like a conversation, not
just a one-off prompt and response. This is a chat interface. Treat it like you would treat an
assistant, essentially, or an employee or a team member who's there to help
you do whatever ID tasks you want to do. Now some juicy stuff. This is what I've been spending most
of my time researching and getting hands on wit hthe advanced use cases. This is really exciting to me. So you can use the API for
further customization and incorporate it into other apps. So if you're new new to an API, Open
AI basically said, 'hey, all of this technology, we're going to let you
use this technology for your own applications and you can actually
customize it further to exactly what you want, and you're just going to pay
us a fee based on how many times you send requests to use our servers'. And it's pretty affordable, relative
to how impressive this technology is. So a lot of different companies and
apps are all building these AI features and tools based off of ChatGPT. Well, based off of the open AI language
models, to be more precise, I guess. So there are a couple of
cool things you can do. I mean, one of the simpler ones is you
can tell the response how creative to be. So the lower level ,it can be, you'll
spit out the same thing every time based on the input you give it. Or on the other end it can
be really, really creative. So it can come up with less likely
responses, if that makes sense. That's a really simple thing to do,
but you can also use embeddings, something called embeddings, if you're
a developer and you want to look deeper into this to give it long term memory. So for example, if I want to train
ChatGPT on every single piece of content I've produced, all of the content on
my website and every single YouTube video on my channel, I can do that. I can give, I can train,
essentially train ChatGPT on all of that information and data. And then if you ask it a question,
I can have it respond based on citations in my content, and it's
not going to hallucinate them. It's actually citing this information
as a form of long-term memory. And it can even tell you, 'here,
I cited this part of this YouTube video and this part of this article
to come up with this response'. And that's something that I
would like to build if you would all find that interesting. And I can make that available for
free, so instead of having to research on YouTube or on Google, you can just
ask Devlin AI and it will show you exactly where the relevant content
is based on what you're asking. So that's a really cool use case
that I'm looking into building. And then also, you can fine
tune models to teach them how to respond to certain prompts in
line with what you're looking for. So this is where you basically
say, 'hey, we're noticing the base ChatGPT, it's not doing a very
good job with learning objectives. So I'm going to feed it 500 sample
requests for learning objectives and 500 responses for how those
learning objectives should look. And I'm going to further train the
model based on some more specialized instructional design expertise
that me or my team might have'. So that I think is such an interesting
area for instructional designers. Instead of training people, we are helping
train literal artificial intelligence how to do these very specific tasks and
we can help them do it much better than these general large language models. So a ton of opportunity there. That's why I'm so excited in building
this AI tool that I've mentioned, because we can make it much more specialized
and powerful for instructional designers based on the further training and fine
tuning and embeddings that we'll use. So if you're interested in that stuff
and you're a developer, definitely look into embeddings and fine tuning. Really cool possibilities there and again,
that's what I'm really excited about. So, If you're excited about that too, and
you'd like to use a tool like that, one that is more customized and personalized
for instructional designers, you can go ahead and type devlin.ai in your address
bar and sign up for the wait list. I'm goinging to be doing some surveys,
some focus groups, making sure we're actually building something you want
to see that's going to actually save you time and that you're going to use. So yes, join that wait list and in the
weeks to come, we're going to start doing some of those surveys and focus groups. And in case you couldn't tell by
this video, I am quite excited about what's possible there. AUm, and then as I've mentioned too, we
have a lot more videos on how you can use different artificial intelligence
tools and instructional design. And some more in-depth tools on how
you can use ChatGPT for different common instructional design tasks. That's all coming. So if you're enjoying the AI
content, please subscribe. That's the best way to let us know
that you're enjoying this and to show your appreciation, and we
will see you in the next video. I appreciate you making it to
the end, and I hope you're as excited about this stuff as I am. See you next time.