00:00:00
There are lots of perfect prompt
formulas out there, and I'm sure
00:00:02
you must have heard of them before.
00:00:04
But the more I prompt these advanced AI
models and learn AI prompting, the more
00:00:08
I realize those perfect prompt formulas
might actually be holding you back.
00:00:13
And the recent videos from Anthropic’s
team and my own experience just
00:00:16
confirmed my understanding.
00:00:18
I've even talked with my friend who is
an experienced machine learning engineer
00:00:22
who trains AI models, and his insight
also validates what I've been seeing.
00:00:26
So in this video, I'll share the so
called truth about AI prompting that you
00:00:30
need to stop believing and the better
way to actually approach AI prompting
00:00:35
and to craft more effective prompts.
00:00:38
The first one, assigning AI a role
will always generate better results.
00:00:42
Everyone will tell you that you
must assign a role in the prompt.
00:00:45
Act as an expert consultant, act as
an experienced marketing professional.
00:00:50
I'm not saying that assigning a role is
not useful, but role prompting might not
00:00:55
be always be as effective as you think.
00:00:57
And from my experience, especially when
prompting using those more advanced
00:01:01
models like GPT 4 model 3.5 Sonnet,
the difference is not quite noticeable.
00:01:06
A study about the effectiveness
of role prompting also shows
00:01:09
that assigning roles will not
consistently improve the response.
00:01:13
Indeed, using a 2-shot chain of
thought prompting is even better.
00:01:17
And more interesting, in experiments
to ask AI to act like an idiot
00:01:21
versus a genius, the idiot
prompt actually outperformed.
00:01:25
Another research also reveals that
the role prompting performance are
00:01:28
often unpredictable and use them
when there is a clear and logical
00:01:32
alignment between the task and the role
00:01:34
that is to say, you might not
need a role in every prompt.
00:01:37
Generally for now, I see assigning roles
works better for tasks that require more
00:01:42
creative thinking, or tasks that require
high accuracy like legal document writing.
00:01:48
So don't use roles as a blind
shortcut to try and make AI sound
00:01:52
smarter or more authoritative.
00:01:54
Use roles when they generally reflect the
context or provide a meaningful framework.
00:01:59
And if you want to accelerate your
AI prompting learning curve, I
00:02:02
recommend this valuable resource
from HubSpot, a library of 1,000+ AI
00:02:07
prompts for marketing and productivity.
00:02:09
I put it in the description
for you to download for free.
00:02:11
This library covers most of the common
marketing and business scenarios from
00:02:15
marketing strategies, brand strategies,
to SEO, pay search, and even productivity.
00:02:20
These are not perfect prompt formulas
to follow blindly, but give you
00:02:23
real world examples that you can
learn from and customize, so you
00:02:27
don't have to start from scratch.
00:02:28
And this is why I like it.
00:02:30
These prompts serve practical starting
points to help you develop your own
00:02:33
natural approach to working with AI.
00:02:35
For example, the brand analysis
section, AI really helps at analytical
00:02:39
tasks like analyzing your brand
positioning compared to competitors,
00:02:43
identify gaps in your brand messaging.
00:02:45
You can download this in the
description below for free.
00:02:48
And thank you HubSpot for
sponsoring this video.
00:02:50
The next one.
00:02:51
There is a perfect prompt formula
you should always follow, and
00:02:54
this is absolutely misleading.
00:02:56
While I agree that structuring your prompt
can help because that makes your request
00:03:00
easier for AI to follow your thought
process, they shouldn't be the end point.
00:03:04
We need to understand that why we have
those frameworks at the first place.
00:03:08
It's because we as human want
AI to mimic the way we think.
00:03:11
For example, the RTF framework,
Role Task Format, RISEN framework,
00:03:16
few-shot prompting, chain-of-thought
prompting, whatever they are, these are
00:03:19
All these are trying to mimic
how a real human approach a
00:03:22
problem in a real situation.
00:03:24
So if you only follow these perfect
formulas or technical frameworks or
00:03:28
even just end there, you're limiting the
creative side of these smart AI models.
00:03:34
Just imagine how you solve a
problem in reality without AI.
00:03:38
If you always follow the same set of
rules, you are limiting yourself to always
00:03:42
solve the problem from rigid perspectives.
00:03:44
And this is the problem.
00:03:46
Even for those technical frameworks like
chain of thought promptings, they might
00:03:49
not be needed for simple tasks, but
only for more complex reasoning tasks.
00:03:54
Like this example, you can see
following a rigid framework won't
00:03:57
generate better results in the first
place, and the response without using
00:04:00
any framework is actually better.
00:04:02
And I would say this is even more obvious
when it comes to advanced AI models.
00:04:07
So don't force every prompt into a rigid
template with unnecessary sections.
00:04:12
Do focus on clear communications and
include only the important context
00:04:16
that matters for a specific task.
00:04:18
The next one, longer prompts will
always generate better results.
00:04:22
This is also incorrect.
00:04:24
While for longer prompts, it may mean
you can include more context, but
00:04:28
it doesn't guarantee the results.
00:04:29
in response quality will be better.
00:04:31
In fact, study has shown that there
is a notable decline in LLMs reasoning
00:04:35
performance as the prompt length increase.
00:04:38
This degradation is consistent
across all tested models.
00:04:42
That means when LLMs are loaded with too
much information, they may struggle to
00:04:47
identify and prioritize the most relevant
details, consume more tokens, leading to
00:04:52
inconsistent even low quality in response.
00:04:55
In fact, you should aim to prompt AI in a
way that you can achieve the best possible
00:04:59
results with the least tokens, because the
more tokens spent, it means higher costs.
00:05:05
Even though adding more examples can
usually create better performance,
00:05:08
but it doesn't mean you need
to write super long prompts.
00:05:11
Also, there are still lots of different
factors impacting the output and not
00:05:15
just the number of examples alone.
00:05:17
So don't add excessive context thinking
more contexts must lead to better outputs.
00:05:22
But to select a few high quality,
relevant contexts or examples
00:05:26
that illustrate your task needs.
00:05:28
The next being super polite to AI
will always lead to better results.
00:05:33
Some people say that you need
to be very polite to AI in
00:05:36
order to get a better response.
00:05:38
The reality is it depends.
00:05:40
Although some research shows that
politeness can affect LLM performance,
00:05:44
it emphasized being overly polite
does not guarantee better results.
00:05:48
And from my own experience,
just being polite won't
00:05:51
significantly improve results.
00:05:52
Indeed, they can sometimes
lead to confusions to AI to
00:05:56
understand the crux of the task.
00:05:58
These LLMs are trained at using
the RLHF method, reinforcement
00:06:02
learning with human feedback,
00:06:04
which means involving real humans in
rating the response before they roll out.
00:06:08
And that's why LLM's response is always
being fine tuned to say what should be
00:06:13
acceptable, including politeness and
how it should respond to rude language.
00:06:18
And that's why indeed I find it's
less about being polite or not, it's
00:06:21
about the emotions of the prompts.
00:06:23
LLMs are being LLMs are being trained
to understand human emotion language.
00:06:27
So incorporate emotional context
implies urgency and importance
00:06:31
like using all capital letters.
00:06:33
Study also shows that incorporating
negative stimuli also have the same
00:06:37
impact on LLM performance, like
explicitly express disappointment.
00:06:42
So I'm not telling you to be rude
to AI, just treating AI with respect
00:06:47
is generally a good practice.
00:06:48
So don't add unnecessary polite or
formal language, only add emotional
00:06:53
language when it's necessary
and matters to the task itself.
00:06:57
So how to better approach AI promptings
and craft more effective prompts?
00:07:01
First, we must understand
how AI actually works.
00:07:04
AI doesn't truly understand
context like human do.
00:07:07
It just makes its best prediction
by calculating probabilities based
00:07:12
on the input, it generates response
based on the pattern matching.
00:07:16
So in general, the more specific and
clear your prompt input, the better
00:07:20
its prediction and results will be.
00:07:22
And second, AI models will only
get smarter, but it doesn't
00:07:25
mean prompt engineering is dead.
00:07:28
It is evolving from just focusing
on techniques to be more strategic.
00:07:44
It's not just the Claude model, it's
also how OpenAI trains the O1 model to
00:07:49
incorporate chain-of-thought techniques
into the models to scale the performance.
00:07:54
So we can expect the need for complex
prompting will decrease over time
00:07:58
and the models will trigger those
technical prompting frameworks
00:08:01
techniques without you even noticing.
00:08:04
And therefore, to be able to
craft better prompts, I realized it
00:08:07
really boils down to two things.
00:08:09
Clear thinking and clear communication.
00:08:12
Basically, all those technical
frameworks, techniques are all evolved
00:08:16
from these two core components.
00:08:18
Clear thinking, understand what
you really want, it means to
00:08:21
a thought planning process.
00:08:23
Clear communications, including your
certainty and uncertainty in the
00:08:26
prompt using simple direct language.
00:08:29
When you pair thoughtful planning with
effective communications, you are creating
00:08:33
a positive loop to make sure you will get
a better response from LLM every time.
00:08:38
First, begin with an end
in mind, not a formula.
00:08:41
Don't try to start with a prompting
formula, but first get super clear
00:08:45
on your end goal and the problem
you're trying to solve first.
00:08:49
Is it a proposal, an analysis, a report,
or summary, or just simply getting ideas?
00:08:54
You need to identify your current
state and your desired outcome and
00:08:57
what success looks like to you.
00:08:59
And so AI is here to help
you to fill that gap.
00:09:02
If you even don't have any ideas, you
can use the 5W1H method
00:09:06
And I would say the what and
the why is the most important.
00:09:10
The what force you to pinpoint the
real problem that you need to solve.
00:09:14
The why reveals the motivations behind
this problem to give AI more context.
00:09:19
And so to give a more meaningful response.
00:09:21
For example, you're analyzing
your company sales data.
00:09:24
So the current state is you
have monthly sales data.
00:09:27
The desired outcome is you want
to understand the sales trends and
00:09:31
success is about identifying the
growth patterns to inform strategies.
00:09:35
So to craft an effective prompts, analyze
the monthly sales data is the What,
00:09:40
and inform better sales
strategy is the Why.
00:09:43
Of course, you can also include other
elements to improve context, but
00:09:46
getting clear on your end goal and what
you want AI to achieve will set the
00:09:50
strong foundations of a good prompt.
00:09:52
The next tip is to
identify the type of tasks.
00:09:55
Not all tasks are created equal.
00:09:57
And it is so important to understand
this in the first place to
00:10:00
shape the best possible prompt.
00:10:02
Basically, there are two types of tasks.
00:10:05
Tasks that you do understand what
to do, and you know exactly how to
00:10:08
do it manually, even without AI.
00:10:10
You're just seeking AI for executions.
00:10:13
I call these tasks
“Goal & Process Clear Tasks”.
00:10:16
The second type is tasks you
don't know how to do, but you're
00:10:19
clear about the desired outcome.
00:10:21
You just need more guidance
from AI on problem solving.
00:10:25
I call them “Goal Clear Tasks”.
00:10:27
And each type of task also has
a different level of complexity.
00:10:31
So by knowing which category your
request falls into, you can tailor
00:10:34
your prompt more effectively.
00:10:36
For example, doing keyword research
for a mental health website and
00:10:39
identify the best money keywords,
you know exactly how to do it.
00:10:43
And so you can just ask AI to analyze
the data based on your criteria
00:10:47
and ask AI to execute all the steps.
00:10:50
However, for building a mobile
application for a health website with
00:10:53
specific features, you only know the
outcome, but you don't have any ideas.
00:10:56
And then instead, you need to frame
the prompt to focus on exploration,
00:11:01
brainstorming and strategy generation.
00:11:03
And you can even express you are
uncertain about the approach you're
00:11:06
inviting the AI to add as your guide.
00:11:09
And this will greatly change
how you frame your request to
00:11:12
align with your actual needs.
00:11:13
Next is to communicate without assuming
shared context or background knowledge.
00:11:18
Besides the goal, whenever I
start writing a prompt, I find it
00:11:21
helpful to ask these two questions.
00:11:23
What do I actually know and
I have not yet mentioned?
00:11:27
What background context it need
to fully understand my request?
00:11:31
For example, help me optimize
my funnel's CTR is a bad prompt.
00:11:35
It lacks all important context, like
the why context we just mentioned.
00:11:39
But this enhanced version will
generate a much better response.
00:11:43
I share all necessary context.
00:11:44
It's a B2B software, it explains
funnel structure, it shares current
00:11:48
metrics, it includes what success looks
like and makes the goal much clearer.
00:11:53
Now you may wonder how
much context is enough.
00:11:55
There is no absolute rules,
but generally three principles.
00:11:59
First, include details that
are not common sense to AI.
00:12:02
This is usually about more specific
details about your project or tasks.
00:12:05
Just ask yourself, "Is
this common knowledge?"
00:12:08
If no, you better include some
background in your prompt.
00:12:11
And second, if any information in
your prompt would confuse AI, just
00:12:15
ask yourself if your friend who has
no background information would get
00:12:19
confused by this piece of information.
00:12:22
If yes, clarify it or remove it.
00:12:25
And lastly, start simple
and add if needed.
00:12:28
Keep iterating based on the AI
response until you're satisfied and
00:12:32
add some constraints, examples to
shape the kind of response you need.
00:12:36
So the key is more context is
only better when they're relevant
00:12:40
and you should avoid overloading.
00:12:42
And that leads into the next,
identify blind spot in your prompt.
00:12:45
Sometimes you may not fully know
you have given enough context to AI.
00:12:50
So you can actually ask
AI what it needs from you.
00:12:53
And so to see it more as a thinking
partner, just as if you're working
00:12:57
on a project with a teammate.
00:12:59
For example, explicitly ask in your prompt
to encourage it to ask for additional
00:13:04
Or even more direct, "What
information it needs from you
00:13:07
in order to solve the task?"
00:13:09
these not only improve the response
quality, but also trains you to
00:13:12
think more about the information
gap in your own reasoning.
00:13:16
Another method you can use is to ask it
to identify any contradictions or gaps.
00:13:21
You can even set it in
your custom instructions.
00:13:23
So whenever it encounters any
contradictions, it should highlight them
00:13:27
before it generate the response to you.
00:13:29
This way, you can try best to
minimize the information gap and
00:13:32
provide the most necessary context
to AI for the best possible response.
00:13:37
I even encourage you to direct AI
to expose blind spots in your every
00:13:42
prompt and do it for some time.
00:13:44
And you will start understanding
the thought process and naturally
00:13:47
become more sensitive when a response
might not be up to your standard.
00:13:51
The next tip is to apply the 80 20
rules, 80 percent of your desired
00:13:55
results can come from just 20
percent of your prompting efforts.
00:13:59
So do not over engineer
in the first place.
00:14:01
Start simple.
00:14:03
Give the AI a straightforward,
decent prompt and see what you get.
00:14:06
If you're not satisfied with the
output, keep refining and iterate.
00:14:10
Also, use concrete,
clear natural language.
00:14:13
I know they may sound too obvious, but
this is one of the top mistakes people
00:14:18
make when approaching AI prompting.
00:14:19
AI models are being trained on
00:14:22
human communication patterns.
00:14:23
So you don't need to use the exact
wordings, but more importantly is to
00:14:27
understand the thinking process behind.
00:14:29
Avoid vague instructions
and aim for specifics.
00:14:32
Instead of saying, "Give me some
business ideas", try saying,
00:14:36
"Propose three new distinct business
concepts targeting Canadian parents."
00:14:40
A tip I personally use is to confirm
with the model its understanding about
00:14:44
the task by asking "Do you understand?"
00:14:47
Or ask it to recap again the task to
you and find if there's any discrepancy.
00:14:52
So you can make sure there's no
misunderstanding before you move on.
00:14:56
When you use natural, clear, direct
language, you're working with the model
00:14:59
in a way they were designed to interact.
00:15:02
And with clear thinking, most of
the time, it's already sufficient
00:15:05
to generate a great response.
00:15:07
When AI is widely adopted, one
of the most valuable skills, and
00:15:11
I'm talking about future proofing
skills, is not following formulas.
00:15:15
It's actually thinking critically.
00:15:17
The book, "The 5 Elements of Effective
Thinking" lays down the core five
00:15:21
elements of improving thinking skills.
00:15:24
And I found it applies
perfectly in how we approach AI,
00:15:28
prompting, all the tech stuff.
00:15:29
Understand deeply, learn from
mistake, keep asking question,
00:15:33
understand how different ideas
interconnected, embrace change.
00:15:37
I've create a bonus videos about
advanced techniques anyone can
00:15:41
use to think smarter with AI.
00:15:42
And I share practical methods to use AI
to enhance your own thinking process.
00:15:47
I have put it in a community.
00:15:48
You can find the link in
the description to join.
00:15:51
And if you want more inspiration
about prompting, also check out my
00:15:54
other video about smart prompting
specifically for AI Search Engines.
00:15:59
I will see you next time!
