In this video, I share the flowchart I use to decide when to use AI and when not to, so you can save time without losing your ability to think critically and make good judgments.
=== About Dr Justin Sung === Dr. Justin Sung is a world-renowned expert in self-regulated learning, a certified teacher, a research author, and a former medical doctor. He has guest lectured on learning skills at Monash University for Master’s and PhD students in Education and Medicine. Over the past decade, he has empowered tens of thousands of learners worldwide to dramatically improve their academic performance, learning efficiency, and motivation. #education#kidslearning
AI Generated·May contain errors
In this video, I'll teach you how to use AI
without letting it think for you.
AI, and especially LLMs, obviously save you time.
But by now, there's enough research saying
that using it either too much or in the wrong way
will make you dumber.
So how do you use AI to save time
and meet whatever demands workplaces on you
while maintaining or even enhancing your ability
to make good judgments, think critically,
and ultimately be more valuable in the workplace?
As a learning and cognitive performance coach,
Answering this question has been a major focus
for myself and for the clients that
I coach for the last two years.
And after working with hundreds of managers, software developers,
business owners, doctors, accountants,
teachers, professionals of every industry,
I've made this framework that I use for deciding
when to use AI and when not to.
This is so important that I even flew my own team
into Melbourne to host a in-person training workshop,
making sure that my own team is AI-enhanced
rather than AI-dependent.
So this is my AI, or more specifically,
LLM usage flowchart.
And so before you use ChatGPT or Claude,
run through this flowchart first.
And the first place we start is asking ourselves the question,
do I have the knowledge or skills to complete this task
without using an LLM
and everything else on this flowchart starts from this one question.
Hey, it's Justin from the future. I was watching this video
and I thought that having some more
examples would help you guys understand it.
So I'm just going to walk through some examples and
interject as we go. So what does it actually mean to have the knowledge
or skills to complete a
task? So a task could be anything.
For example, one of the tasks that I have to do everyday
part of my job is to create product strategy.
So we're launching new products, we're managing
We're managing products, we're maintaining products.
What is the strategy that we're going to do?
What's the business plan around it?
So do I have the knowledge and skills
to create that strategy,
to create these plans that are well-considered?
So this is an example of like a high-level task.
There could be a mid-level task,
for example, managing my team.
Do I have the knowledge and skills to manage my team
to create some kind of solution design
for a particular feature?
A low-level task might be setting up something
on a web server, writing a specific email.
These are much more just
Getting down and doing the job,
it's really executing things, it's creating some kind
of really tangible output.
So probably most of the tasks that you
do on a daily basis are going to be things that
are within your wheelhouse of skills.
You already have the knowledge and skills to do it,
you've probably been doing it for
years, you have experience.
And then there's a smaller subset of tasks
that are going to be new, that are going to
be novel, new technologies you're interfacing with,
new projects and challenges that you've
never tackled before.
So these would be things where you may
not have the knowledge and skills to do it.
Let's carry on with the flow chart.
If the answer is no, I don't currently have the knowledge
or the skills to complete this task without using an LLM,
the next question we ask ourselves is,
will I in the future need this knowledge
to make decisions on any work to come?
So I may not have the knowledge right now,
will I in the future need that knowledge?
Are people going to be asking me questions about this work
or my decision or my strategy
in the weeks and months to come?
If I get ChatGPT to generate a plan for me today,
two weeks from now, am I going to have to change that plan?
Now, let's say that the answer to this is still
no. Then the final question we ask ourselves,
are the consequences of this work that I'm doing
or the decision that I'm
making substantial if I were to get this wrong?
This kind of situation I would
call don't know doesn't matter.
You don't know how to do it and it doesn't matter
what you do anyway.
And in this situation it's probably fine to use
an LLM or an AI to just get that task done.
And this type of situation is actually frankly pretty
common especially if you're in a small team where there isn't
a dedicated specialist to handle
every little thing. You might get some random tasks thrown
at you that you have no real
knowledge or experience around.
It's a pretty one-off event so you just ask Claude or ChatGPT
how to do it, you follow the instructions, you do it.
This is actually one of only two outcomes
Where it's probably fine to just use AI and LLMs to do it
and all the other permutations that we're
going to cover in this flow chart,
it's probably best to leave AI or LLM to a later step or just
not use it at all. And one of the reasons,
actually, I think one of the most important
reasons not to use an LLM to complete a task
or to make a decision has nothing to do with
saving time or efficiency or how effective the AI is.
And actually, it's to do with trust.
And we can see this just by changing
the answer for this final question.
If the consequences of this are in fact substantial,
then this situation falls within
the don't know but it matters scenario.
Okay, so I'll give you another example and
I'll give you examples of all
the other permutations as we go. But for now,
what's an example of something where you
don't know how to do it, you're not going
to need to know how to do it in the future,
but instead of it being inconsequential like
For example, when someone asked me to set something up
on a new web server, like I've never done that before.
I just looked it up and I just did it.
I don't need that knowledge, using an LLM was fine.
So in this situation where it's don't know but it matters,
an example of that is setting up our new privacy policies.
So that's something that I don't have a legal background.
I don't know a lot about like GDPR
and privacy compliance related stuff
outside of just like whatever our lawyers give us.
But that stuff really matters,
especially these days with AI
You know people are very wary about data privacy,
so getting that stuff right is really important.
But once it's set up,
I'm not really going to need to do anything with that in the future.
And so if I were to just get Claude or ChatGPT to just
generate it for me, and then I just chuck
it up, I am not really confident that that's going to be good enough.
The risk exposure to that is quite large.
If it were wrong, which I, because I don't have
the expertise, I have no way of checking
If it were wrong, the consequences would
be significant, and I'm not willing to take
that risk exposure.
And so, therefore, that would fit
into this scenario where I don't know, but it matters.
And what you need to do in this scenario is either learn more
so that you gain the knowledge
to do this, hire someone that can do this, or replan.
Change your plan so that you do not have
this consequence or risk exposure.
And if for whatever reason, this is not possible.
At the very least, if you are going to use an AI
or an LLM to do this anyway, then do a pre-mortem.
This is a type of analysis where you assume
that your decision was incorrect, that your plan failed.
You got the bad outcome and you ask yourself, why did that happen?
But you do that in advance.
With a post-mortem, you do that after
you've already gotten the bad outcome.
You failed, why did I fail?
With a pre-mortem, you say, if it were to fail, why might that be?
And the important thing is that you do this even
when the suggestion seems to make sense.
ChatGPT might give you this great plan or recommendation,
whatever it is, and when
you read it, you're like, yeah, this makes a lot of sense to me.
You need to do the pre-mortem anyway,
because if you are in this situation where you don't
know, it also means you don't have the insight
to know whether what the AI generated for
you is right or wrong.
I consistently am reminded of this because
when I use an LLM on things that I have deep
expertise and knowledge about,
typically things around learning science, cognitive science,
I disagree with 30 to 40% of the things that it tells me.
When I use it for things that I'm not very knowledgeable about,
I only disagree with like 2% or 5% of the things.
Probably what's happening is that
because I know more about this field,
I'm able to spot the issues in it.
But it's a chicken and egg situation
because when I use an LLM on things
that I don't know much about,
not only am I not able to detect
what is right and what is wrong,
I also don't know enough to give it
the right prompts in the first place.
So when I'm using an LLM
on something where I don't know about it, I'm now operating a
situation where it is more likely to give me bad advice
and I am less likely to see
the issues.
And if you use AI in this way,
where you don't know how to do it, but the consequences actually
matter, then the true cost is more than just making the wrong decision.
is that it actually erodes trust
that other people have in you.
And I think this is the big thing
that a lot of people look over.
They don't appreciate how significant this is.
When I give someone a task,
as a team leader, as a manager,
really what I'm saying is,
here's this thing that I'm worried about.
I am trusting you to deliver on this outcome for me.
And at the end of the day,
What I am trusting is actually your judgment.
I am not going to give that responsibility to someone
who doesn't have the knowledge, the skill, and the experience.
Why?
Because I cannot trust that someone without that knowledge,
skill, and experience
is going to make the right judgment.
That they're going to pick up on the errors
or the nuances of the context.
But this is the thing that I have heard.
Countless times, especially
when I talk to senior leaders and managers of teams,
is that as people use tools like ChatGPT and Claude more aggressively,
but more liberally,
they also start getting lazier with their judgment.
And so what I hear from managers
and employers and leaders
and all the industries that I've worked with is that
they can no longer trust
when they get a member of their team to do something.
The issue is not that someone used an LLM to do the task.
The issue is that there is a lack
of transparency around how much higher order human expertise
went into the judgment and critiquing
and refinement when they were using the LLM.
And I personally feel this viscerally as well.
There are members in my own team
who I know very rarely use an LLM for a lot of their tasks.
I know that they've applied that same standard
that I know they're capable of to their work.
Whereas there are some people who I think,
is this you speaking, or is this the LLM speaking?
Because I don't think you, who has the standard
that I know you're capable of, would have said this.
And once you lose that trust, you lose everything.
So that's if the consequence matters versus doesn't matter.
Let's work our way back from the top.
What if you don't have the knowledge,
but the answer to this question,
will I in the future need this knowledge to make decisions?
Is now, yes, in the future I will.
This is a situation where you don't know right now,
but you're going to need to know.
Okay, another example.
What's a situation where we don't know,
but we need to know?
We don't know today,
but in the future we will need to for these decisions.
In my everyday job,
because I'm in charge of creating content
and building product,
I need to know how to create product strategy.
I need to know how to create content.
If I don't know how to do something today,
I could just ask an LLM to tell me exactly what to do.
How do I think about this?
What's the best strategy?
How do I get more views, get a better channel performance,
create a better product?
And it will just tell me something.
That decision might be fine.
But in the future, other people
are relying on me to have that knowledge.
They're going to ask me questions about it.
I need to evaluate if the plan
that we're implementing is actually good or not.
I have ownership on that decision
and all the consequences of that decision.
So it's not enough just for me to have an answer today.
And long-term, it's not efficient for me to ask an LLM
every single time I have a question on this.
Again, because if I don't have the knowledge and expertise,
I am not going to know whether what the LLM is giving me
is good or bad in the first place.
So in a team setting, as a rule of thumb,
anytime someone is depending on you
to deliver on an outcome, an ongoing responsibility,
You probably need to genuinely have the knowledge
and skills to be able to do this.
In an individual setting,
these are the things that you want to have
as your strengths, as your advantages.
Let's say you're preparing for a speech.
You're not really sure how to write this speech.
Well, do you think it's important for you
to be a good writer and a good communicator long-term?
If the answer is yes, you should just get good at that,
even if an LLM could create a better speech for you today.
This is the part where I see a lot of danger,
because there are a lot of individuals
who essentially outsource their individual responsibilities,
like they want to be better at communicating
or emotional self-regulation.
And they're basically, instead of developing those skills,
outsourcing it to an LLM.
And they're actually becoming less capable over time.
Now, what a lot of people will do is they will say,
well, it's going to save me time today,
so I will develop the skill later,
but today I'm just going to use an LLM.
And I would advise you not to do that,
For the reason that I'm about to explain.
People will be asking you questions.
People will be questioning your decision.
You're gonna get more data on your plan or your decision.
And when that data comes around,
you're gonna have to change your plan.
In this situation, if you just use AI and LLM as is,
you're going to trap yourself.
You will create a strategy that you don't know much about.
As you gain more data and you learn more about the strategy
and people start asking you questions,
you're gonna realize you don't know
what you're talking about.
And now guess whose responsibility it is to own
the consequences of that decision?
It's yours.
And in those situations, you either look incompetent or irresponsible.
And it would have been better just to spend
the time upfront to learn about this properly
so that you can fully own the decision
or the recommendation that you gave.
And one easy way that you can avoid getting yourself trapped
in this type of situation
is by clarifying things with a few smart questions upfront.
And one of the most important things to clarify at all times is,
What is the level of urgency?
And this changes your approach significantly.
If it is not very urgent, then as I said,
The best approach is to learn more
and actually gain the knowledge or the expertise or hire
someone who does have that expertise.
If the answer is yes, it is urgent, then you may not
have time to learn about it,
but you may have time to hire someone who has that expertise to
come in on this job and help you out.
You may urgently replan to mitigate the risks.
And if you're using an LLM because you need
to push something through urgently and there
isn't much room to learn, hire, or replan,
then you would do that pre-mortem.
And so here, at least the risks
and the uncertainties are explicitly acknowledged and thought through.
And often what I find is that when someone gives you something
to do and you're not exactly
sure how to do it, but you flag up front.
where your uncertainties are, where you think the risks are,
it can actually cause the senior decision makers to say,
actually, now that we have this information,
we're going to rethink the approach.
And guess what that does for you?
People trust you more.
You become more valuable in that team
because you're the person who put your hand up
and provided crucial information that saved the company
$3 million across the next quarter,
pursuing something that was never going to work in the first place.
And so you can see, so far, most situations
At this point, actually lead to you needing
to just learn more to have the knowledge and skills
or hire or to replan. And at this point,
most people in principle agree with what I'm covering
so far, but they don't agree with it in practice
because it is not realistic to be able to learn
more and gain this knowledge with the time restrictions.
You're busy enough already just
Just doing stuff.
You don't have time to learn how to do this stuff as well,
especially when the industry and the workplace is evolving
to expect you to just do things faster because of AI.
But the more astute of you are seeing there's this catch-22
where if I use AI to do this thing,
I kind of screw myself in the future
because it's degrading my critical thinking ability
and making me dumber.
And this is genuinely what is happening.
What's happening?
I'm seeing.
Hordes of experienced professionals
who have a lot of expertise who are now not the professional
that they once were.
They're not the thinker they once were.
They are not the problems over they once were.
What made them.
Special and unique and valuable,
they have eroded away themselves by over-relying on AI.
And so the better solution
to avoid that grim trajectory for yourself
is to just get faster at learning new things.
And most people think that that's not possible,
but I know, as a learning and cognitive performance coach,
that it is possible.
In fact, most people can learn something
in about half the time that it normally takes them.
And so where can you learn how to do this?
Number two, I've got
a free weekly newsletter you might be interested in.
It's three to five minutes to read.
I put in insights and perspectives
and strategies on how to learn faster, how to think clearly
through uncertainty,
how to stay in flow for longer and activate deep flow.
These emails are written by myself without using an LLM
and I do that deliberately because
it's important that you guys are able to trust what I put out
and it is a distillation of
of my years of coaching experience.
Helping people to learn faster and think faster.
So if you're interested in joining that,
I'll leave a link for you in the description below.
It's free to join, you can unsubscribe at any time.
But let's move on to the final branch.
So this part is still incomplete.
What if you do have the knowledge
to do something without the LLM?
In this situation, you should ask yourself,
isn't AI equally good at doing that same thing?
Okay, as a quick tip for you guys,
what are the types of things that AI
tends to be equally good at or better.
Usually, these are low-level execution things.
I think about it like this,
would the average person with a bit of training, a bit of practice
and experience be able to do this?
If the answer is yes, probably an LLM can just do it already.
Where I find that it's...
generally not very good is when we're starting
to think about things that are very multifactorial.
There are a lot of different interpretations.
You would have to spend a lot of time gathering
a lot of experience thinking deeply about
this to come to certain conclusions. And if you ask
10 different experts, they might have 10
different opinions. Sometimes I like to
ask myself the question, is there actually a correct answer to this?
If the answer is no, there's not really a correct answer,
then probably AI is not going to be as good as you
who actually has this experience and judgment.
So you can use this as like a general rule
to save yourself from spending four hours
prompt engineering and LLM
to try and get it to do something for you
only to realize that it's just not that good.
If the answer to this is no,
we are in the situation of
I know, but AI doesn't, and naturally,
what you should do here is do it yourself.
My newsletters fall
into this category where I feel like I know how to do this well, AI
doesn't, I'm just going to do it myself.
Yes, it's going to take a little bit more time,
but again, time is not wasted if it's
achieving the outcome that is important to me,
and if AI is able to achieve a bad outcome
Really fast.
That doesn't really help me.
But what if the answer is yes,
AI is actually equally good as me, then here is potentially
my controversial take on this,
is that you should ask yourself a final question of does
it save a significant amount of time?
And if the answer is yes,
using an AI which is equally good as me would save me a significant
amount of time, then it's probably okay.
If the answer is no,
it doesn't save me a significant amount of time, then in this
situation, I actually believe you should still just do it yourself.
And by doing it yourself, it's going to take more effort,
but you at least preserve your
ability to think and execute,
and it preserves how much other people can trust you.
So I kind of realized that this flow chart
can be a little bit confusing.
So if you want a quick reference version of this
so that you're not flicking back to this video all the time,
I will leave a link to an Instagram post
that talks about this topic.
And if you go to the Instagram post and you click the link
and you leave a comment saying, save my brain,
then I will have an automation that DMs you
a downloadable flow chart of this
that you can save to your own device.
And while you're checking out the post, you may as well
Follow me on Instagram as well if you're not already.
And so here are the two situations.
We have one up here and one down here.
It is probably fine just to use an LLM to do the task,
make the decision, make the plan
if you have the knowledge,
AI is just as good as you and it's going to save you a buttload of time
or you don't know how to do it,
you're never going to need to know anything about this
and it doesn't matter if you do it wrong or not.
If you don't know and it doesn't matter.
But in every other situation, when the consequences matter,
when in the future you need to have knowledge about this,
when it doesn't save you much time,
when AI is not very good at it,
just do it yourself, learn to do it yourself,
hire someone who knows how to do it,
or plan so you don't need to know.
And in the worst case, last resort scenario,
where you are using AI regardless,
please, at the very least,
Do your pre-mortem analysis so that the risks
and the uncertainties are discussed upfront.
And so by using an LLM in this way, it allows you to be AI enhanced.
You still think critically,
you get better at the things that you should get better at,
but the areas where using
an AI genuinely helps you save time, you can use that.
And you can use that with confidence,
knowing that it's not degrading your brain's ability to think.
Now, if you want to take the next step,
which is to learn how to learn things faster,
then I would check out this video here.
It's one of the best videos that you can start with,
especially if you've never really thought about
how to learn faster before.
It's way beyond like memory hacks and mnemonics.
It's a type of learning that you need to get good at
if you wanna be a really fast learner
in a professional real world setting.
So if you're interested, check that video out.
Otherwise, thank you so much for watching
and I'll see you in the next one.