99% of Beginners Don't Know the Basics of AI
By Jeff Su
Summary
## Key takeaways - **Three Types of AI Tools**: AI tools can be categorized into three types: Standalone tools like ChatGPT, integrated AI features within existing software, and custom AI solutions built for specific problems, such as John Hopkins' sepsis detection system. [00:30], [02:02] - **Surface Implied Context for AI**: When prompting AI, explicitly state implied context, just as you would naturally do in human conversation. For example, if asking for restaurant recommendations, mention dietary restrictions even if obvious to a friend. [03:43], [04:06] - **Zero-Shot vs. Few-Shot Prompting**: Zero-shot prompting uses no examples, one-shot uses one, and few-shot uses two or more. Providing more relevant examples generally leads to more relevant AI output. [04:53], [05:38] - **Chain of Thought for Complex Tasks**: Break down complex tasks into manageable steps when prompting AI, similar to writing a cover letter by first generating a hook, then the body, and finally the closing paragraph. [05:52], [06:26] - **Understand AI's Core Limitations**: Be aware of AI limitations such as biased training data, knowledge cut-off dates, and the potential for 'hallucinations' where AI generates inaccurate information, especially critical for high-stakes tasks. [07:02], [07:32]
Topics Covered
- Custom AI solutions are for everyone, not just experts.
- Always explicitly state implied context for better AI outputs.
- Break down complex tasks using Chain of Thought prompting.
- Understand AI's core limitations for critical tasks.
- Google's AI Essentials course is not for experienced users.
Full Transcript
last week I spent 5 hours and $49 to
complete Google's latest AI Essentials
course for beginners and since I need to
recoup that money to fund my addiction
to parasocial relationships with AI I
made this video to share five key
takeaways from the course the pros and
cons and to give you a definitive answer
as to whether this certificate you
receive at the end will indeed get you
laid I mean paid paid more because you
now have a new skill that was a weird
slip of tongue there takeaway number one
broadly speaking there are three types
of AI tools out there first Standalone
tools are AI powered software designed
to work independently with minimal setup
this category includes general purpose
chat Bots like chache BT Gemini Claude
and perplexity as well as specialized
apps like spico otter AI mid journey and
Gamma although they serve completely
different needs all these tools are
classified as Standalone because they
can be accessed directly through their
websites or apps and can be used without
integration with other software this is
in contrast to the second category tools
with integrated AI features which refers
to built-in enhancements within a
particular piece of software for example
after drafting a post in Google Docs I
can either copy and paste this text over
to a standalone app like chbt to improve
my writing or I can use the built-in or
integrated Gemini for workspace AI
feature to make adjustments another
example would be I can either use mid
Journey as a standalone tool to generate
images for a presentation or I can
create an image directly within Google
Slides by using again gemni for
workspace in these instances chbt and M
Journey are Standalone AI tools and
Google Docs and Google slides are tools
with integrated AI features third a
custom AI solution is an application
that's tailor made to solve a specific
problem for instance John Hopkins
University developed an AI system with
the sole objective of detecting sepsis
this custom AI solution improved
diagnostic accuracy from 2 to 5% to an
average of 40% and if you're like me and
have no technical background you might
think oh custom AI Solutions are
extremely Technical and I'll probably
never have to use them in the workplace
in reality the opposite is true because
well-designed custom AI Solutions should
require little to no technical knoow for
example when I was in the sales team I
managed over 200 clients every quarter
and Performing research on every single
one was obviously timec consuming
nowadays there are custom AI solutions
that can ingest all the information
about those 200 clients taking into
account factors like seasonality
historical data and Industry Trends and
rank those clients by How likely they
are to need assistance helping the
salesperson prioritize their time by the
way if you're thinking of buying
Google's AI Essentials course don't
because I only found out after I paid
for it you get the AI Essentials course
for free if you enroll in the Google
project management certification on
corsera who is kindly sponsoring this
portion of the video regular viewers
know that I have a full-time job and
project management obviously plays a
large part in my day-to-day to be
transparent though I'm mainly
self-taught since there wasn't a go-to
course back in the day but I recently
started the project management
certification on corsera since it's now
like the golden standard for project
management project management is
literally applicable to all Industries
and roles so if you want to be more
organized in the workplace click the
link below to enroll in the Google
project management certification and
unlock the AI Essentials course for free
thank you corsera for sponsoring this
part of the video back to my takeaways
the second learning from the course is a
prompt engineering tip and it's to
surface the implied context to
illustrate this imagine your vegetarian
friend asks you for restaurant
recommendations you instinctively reply
with vegetarian friendly options even if
your friend doesn't explicitly say hey
idiot make sure it's vegetarian here
here the fact that your friend is a
vegetarian is implied context and needs
to be explicitly stated when
communicating with AI tools like Chachi
BT and Google Gemini another example
might be you're preparing to negotiate a
raise with your boss you know in your
head that last year you received a 10%
increase this year you're the highest
performer on the team and the industry
average is a 12% increase so you decide
to ask for 15% raise If You Leave Out
All That implied context when
brainstorming negotiation techniques
with an AI tool you will receive a lower
quality AKA more generic output if you
want to dive deeper into this I have an
entire video on how to write the perfect
prompt so I'll leave a link to that down
below and if you want to make a copy of
my five favorite prompts for
productivity I'll leave a link to my
completely free workspace toolkit as
well takeaway number three know when to
use zero shot and few shot prompting in
a nutshell the word shot simply means
examples zero shot means you use a
prompt with no examples one shot means
you include one example few shot means
you include two or more examples for
instance a zero shot example might look
something like write me a pickup line
for Bumble which is a completely
hypothetical scenario I would never
condone much less participate in a
one-hot prompt would be write me a
pickup line for Bumble reference this
pickup line my friend used that worked
well for him and you include an example
of what your friend wrote A F shop
prompt would look the same as one shot
with just two or more examples of
successful pickup lines and as you can
imagine the more relevant examples you
provide the AI tool the more relevant
the output and for the record if my H
future wife is watching this I've
obviously never done this myself this is
just an example for this video as a
matter of fact I don't I don't even use
dating apps takeaway number four use
Chain of Thought prompting for complex
tasks I've talked about this concept in
previous videos but I really like this
simple and straightforward definition
from Google's course when you divide a
single task into more manageable steps
you help the large language model
produce accurate and consistent results
a relatable real word example would be
writing a cover letter option one you
share your current resume and the job
description with the chatbot and just
prompt it to write you a cover letter
option two with Chain of Thought
prompting you break the large task
writing an entire cover letter into
manageable steps with step one being
based on my resume and the job
description write an attention Gra
having hook for my cover letter then
after making some minor tweaks to the
hook paragraph step two would be pasting
that back into the chatbot and asking it
for the body paragraph rinse and repeat
for the closing paragraph if you want
proof that this works I have an entire
video where I show job Seekers how to
use Chain of Thought prompting to not
only write cover letters but to also
improve their resumés I'll leave a link
down below takeaway number five is a
topic that many of us including myself
tend to overlook when using AI tools and
it's to understand the limitations of AI
broadly speaking there are three main
limitations first the underlying data
that is used to train AI models may be
biased if a text to image model is only
trained on minimalistic Graphics it
might not be able to produce flashy and
bold designs second there simply wasn't
enough information in the source
training data on a given topic many AI
models have a cutoff date so if you ask
for something that happened recently it
won't have enough data on that topic to
give you an accurate accurate answer
finally hallucinations are AI outputs
that are straight up factually
inaccurate sometimes this is a feature
and not a bug when you're brainstorming
new ideas for example other times
hallucinations perpetuate false
information so you definitely want to
double check your answer for what I call
high stake tasks for example what sort
of supplement or vitamin to take given
your health goals when it comes to pros
and cons I'm going to start with who
this course is not for this course is
not for you if you're already using AI
tools like chpt and Google Gemini as
part of your daily workflow and you're
looking to dive deeper into specific AI
use cases and that's because although
they do a great job explaining complex
topics the examples they give in the
course are pretty vague for instance in
one of the lessons they gave an example
where a company uses AI to decrease
customer service response times and that
was it that was entire example it would
have been nice to dive a bit deeper and
talk about whether this company used a
standalone AI tool or custom AI solution
how they trained the workers to use the
AI system how they grounded the data so
that the AI system didn't hallucinate
since this was a customer service um
application right so stuff like that
that being said this is an excellent
beginner level course with three huge
advantages first you're learning from
Google employees who are established
experts in the field of AI they know
what they're talking about and they're
not just some random person on the
internet who enjoys making crass jokes
second as a visual learner I'm amazed at
how they're able to use Simple Graphics
to explain complex topics for example
they analogize AI tools and AI models to
a car and its engine the model sleng
engine provides the underlying
capabilities while the tool like the car
helps you in completing the task third
the interactive elements are
surprisingly helpful the activities or
homework are well-designed in that they
actually help you learn a key concept
from that lesson the graded assignments
are also not that easy they're mainly
multiple choice but you actually have to
pay attention to get 80% and pass the
quiz last but not least the course
provides a curated list of AI tools for
beginners to explore and includes a
glossery of common AI terms that are now
prevalent in our daily lives to sum up
this course is great for beginners
visual Learners and can use the legit
certificate you get the end to attract
prospective employers and or Partners um
if you this video helpful you might want
to check out my summary of Google's free
AI course that's a bit more conceptual
but equally important in the meantime
though as
usual have a great one
Loading video analysis...