Speaker A
If you use AI and you don't have a note-taking system, you have to start one now.
Vicky Zhao explains why a note-taking system is essential for effective AI use, emphasizing context engineering over prompt engineering.
Key Takeaways
- Start a note-taking system to improve AI output quality and thinking.
- Context engineering is more important than prompt engineering for AI communication.
- Explicitly define inputs, sources of truth, and judgment criteria in notes for AI.
- Use frameworks or examples to help AI understand what good output looks like.
- Proper note-taking enables deeper cognitive engagement rather than outsourcing thinking.
Summary
- The shift from prompt engineering to context engineering makes note-taking crucial for AI users.
- Notes help convert tacit knowledge into explicit knowledge, improving communication and collaboration.
- Effective communication with AI requires clear context, including role, goals, audience, style, and constraints.
- Context engineering involves specifying inputs, sources of truth, and judgment frameworks for AI.
- Frameworks help articulate quality standards and improve AI output by guiding its evaluation criteria.
- Users can teach AI their own frameworks by providing examples of good and bad outputs.
- A well-maintained note system enhances AI's ability to summarize, spot patterns, and assist deeper thinking.
- Outsourcing thinking to AI without notes can lead to disengagement and lower quality output.
- Proper use of AI with notes can deepen engagement and expand mental capacity.
- Vicky demonstrates practical application using her Obsidian vault to improve AI interaction and thinking quality.
Full Transcript — Download SRT & Markdown
Speaker A
And the reason lies in the clear trend that we're moving away from prompt engineering.
Speaker A
And the increasing importance of context engineering, having used these tools for two plus years.
Speaker A
I can really say that having the right context, having the right notes to be able to set up what you actually wanted, what you want these LLMs, whether it's ChatGPT, Claude, Gemini, whoever, to be able to draw upon the quality of the output is going to be different.
Speaker A
As well as the quality of thinking you can put into it.
Speaker A
Everyone's complaining that, you know, we have to outsource thinking, it doesn't have to be that way, but if you don't have notes, then, yeah, there's no choice.
Speaker A
I've talked about this in another video that you can explore here.
Speaker A
Which is it is so important to turn tacit knowledge, knowledge that just exists in our head.
Speaker A
And we cannot articulate into explicit knowledge.
Speaker A
So that we can share with other people.
Speaker A
Which is the source of why note-taking came into place, right?
Speaker A
Before, level one was just us and our brain.
Speaker A
We have things swirling in it and because we kind of know where they are.
Speaker A
We say, okay, that's fine, right?
Speaker A
Don't need to write anything down.
Speaker A
Until the second iteration, we developed writing systems.
Speaker A
And we realized, okay, there are things that we need to unload.
Speaker A
In order for our brains to do more.
Speaker A
So that's why we started to translate this tacit knowledge into explicit knowledge.
Speaker A
Because then we can work with other people.
Speaker A
Right, so as we're talking about on this channel.
Speaker A
It really is very important to recognize that communication is not about performance.
Speaker A
Of course, for for some people who are, I don't know, motivational speakers or whatever.
Speaker A
I guess the performance side is also important.
Speaker A
But for people who are working on ideas, knowledge workers.
Speaker A
Right, why we need to communicate clearly is not to look good, but to get the tacit knowledge into explicit knowledge.
Speaker A
And to share with other people so they can understand and we can take action together.
Speaker A
With as little friction as possible.
Speaker A
Right, and vice versa.
Speaker A
When they're saying something, we can process it and understand it.
Speaker A
Which is why frameworks are really important.
Speaker A
But that is the core of communication.
Speaker A
And now we're adding on another layer of it's not just us and other people.
Speaker A
But there's also an AI that has access to other kinds of information.
Speaker A
So we can put it together.
Speaker A
And all move forward.
Speaker A
And having a note-taking system that has the important information that the LLMs can use makes everything better.
Speaker A
It then you can use it to work on the things that take you a lot of time.
Speaker A
And I'll show you an example in just a moment.
Speaker A
But just want to iterate why you must have a note-taking system and have information downloaded from your brain.
Speaker A
Onto something that LLMs can read.
Speaker A
And then use its superpower to be able to summarize, to be able to spot patterns.
Speaker A
All of those good stuff.
Speaker A
Okay.
Speaker A
So that's part number one.
Speaker A
Now, what do we actually put into the notes then?
Speaker A
What kind of notes do we need to have?
Speaker A
Well, luckily, there are just six categories.
Speaker A
We have to think about.
Speaker A
And I'm drawing straight from context engineering.
Speaker A
Because to be able to communicate with AI, we might as well use its rules.
Speaker A
And its way of thinking.
Speaker A
So we mentioned three already that mostly are covered in prompt engineering.
Speaker A
Which is we have to give it a role.
Speaker A
And then give it some general idea about the goal and the audience.
Speaker A
That is going to use that output.
Speaker A
Then also we want to give it some constraints around style or formats.
Speaker A
These kinds of things.
Speaker A
Most people know this part already.
Speaker A
Now, the next three things are more covered in context engineering.
Speaker A
And also notes that we want to have.
Speaker A
Which are about the inputs.
Speaker A
The source of truth.
Speaker A
As well as judgment.
Speaker A
Okay.
Speaker A
So by input, I mean for the LLM.
Speaker A
You know, what's the data?
Speaker A
Just in general that it can use.
Speaker A
So for example, for me, everything in my vault, you know, in my Obsidian vault, it can use.
Speaker A
Or it maybe is, okay, there maybe there's some personal things or sensitive things or confidential things.
Speaker A
I don't want it to use.
Speaker A
Then I can also say, okay, use everything in my vault.
Speaker A
Except for these tags.
Speaker A
Right, you want to be able to tell it explicitly.
Speaker A
What's the input it can use.
Speaker A
Number two is the source of truth.
Speaker A
So what kind of documents outrank its generic information?
Speaker A
Right, so for example, when you have a do not use, it means, yes, you have these things in your general information.
Speaker A
But I don't want you to use them.
Speaker A
That's a source of truth.
Speaker A
So that it knows, okay, I'm not going to rely on those.
Speaker A
I'm going to focus specifically on these documents.
Speaker A
Then the third one is judgment.
Speaker A
So how does LLMs judge what is good, what is bad?
Speaker A
Right, this is especially where we need to take tacit knowledge into explicit knowledge and luckily there are frameworks to do this.
Speaker A
And frameworks help us better articulate what's good and what's bad.
Speaker A
So that's the type of notes.
Speaker A
That we're going to take.
Speaker A
A quick note on frameworks.
Speaker A
I think if you've used AI for a while, then you probably noticed.
Speaker A
It has several frameworks that it keeps on using.
Speaker A
Right, for example, for formatting, it uses the heading options so that as it goes through, it organizes things by headings.
Speaker A
Or it likes to say a lot, like it is not this.
Speaker A
It is that.
Speaker A
Right, a great little via negativa there.
Speaker A
But the problem with its frameworks is it keeps on using the same ones.
Speaker A
And it keeps on using very generic ones when it produces output.
Speaker A
But to be able for you to judge what's good or what's bad.
Speaker A
A lot of the time, a simple way of doing that is to say, okay, write me a sales letter.
Speaker A
Right, and mostly the LLM might use AIDA.
Speaker A
Like something that's very basic in copywriting.
Speaker A
But if you have a different framework, you would like it to focus on.
Speaker A
Then just share that one.
Speaker A
Right, and this way you are inherently building what's good into your prompt and your context.
Speaker A
So that you don't have to worry about how does it judge, I don't know, like intuitively I look at it.
Speaker A
It's pretty good.
Speaker A
Right, instead.
Speaker A
What you can do is give it a framework.
Speaker A
Or if really you don't have any frameworks, but you have examples of what's good and what's bad.
Speaker A
It would be great to have.
Speaker A
You know, the good and the bad so that you can get your LLM.
Speaker A
To figure out what is the what is the framework underneath.
Speaker A
And the other thing about LLMs is they're great at distilling.
Speaker A
Right, as long as you give it information, it's pretty good at distilling.
Speaker A
So what you can do is if you have 10 examples of things that you think are great and 10 examples of things that you think are terrible.
Speaker A
Get the LLM to create that framework for you.
Speaker A
Right, help me articulate what makes these examples great and what are the patterns of the bad ones.
Speaker A
Right, then you've got your own framework that you can use next time.
Speaker A
Consciously.
Speaker A
So that is a great way of injecting framework in a way that helps you get the outcome that you want.
Speaker A
All right, I'll give you an example from my actual Obsidian vault.
Speaker A
So that you can see how not only the quality of the output changes.
Speaker A
Also the quality of the thinking changes.
Speaker A
And I really believe that, yes, working with AI can absolutely make us dumb.
Speaker A
And studies have shown, right, if we just outsource our thinking.
Speaker A
Then, yes, as we go on, quality of the output drops.
Speaker A
Also we are disengaged from the thinking experience.
Speaker A
But that's not the only way to use LLMs.
Speaker A
And studies have also found.
Speaker A
Depending on how you use it, right, you can deeply engage it.
Speaker A
Actually think deeper and access things that you didn't have the mental capacity to do before.
Speaker A
And really what is holding us back from being more engaged in the thinking process.
Speaker A
Is have we downloaded the context out from our brain and into the LLM?
Speaker A
So when we have a conversation, it's an intelligent one.
Speaker A
It's one that's based on the information that we want it to work with.
Speaker A
Right, so if you feel like, okay, I'm just getting dumber by the day using ChatGPT or whoever.
Speaker A
Then you have to also ask yourself, am I also just consuming and not processing?
Speaker A
Outputting documents.
Speaker A
That I can feed my LLM.
Speaker A
Because if you're not doing that process, right, then there is not much context to share.
Speaker A
Right, because it's difficult, not every single time you're there.
Speaker A
You want to start typing out the context.
Speaker A
It's very time-consuming.
Speaker A
There's a lot of friction.
Speaker A
Which is why taking notes as a habit then help you just drag and drop the context into the LLM.
Speaker A
And your conversation with it.
Speaker A
Right, so if you feel like, okay, this is what I'm missing.
Speaker A
Then let me just show you how significantly things can change.
Speaker A
For the example, I'm going to use a problem a lot of us on this channel face.
Speaker A
Which is I'm multi-passionate.
Speaker A
I'm interested in a bunch of stuff.
Speaker A
But I feel so scattered all the time.
Speaker A
All I know is I don't want that singular track career or, you know, life goals.
Speaker A
That other people have.
Speaker A
By the same time, I don't know how do I focus in order to get to where I want to go.
Speaker A
Like there must be something that can tie everything together.
Speaker A
So, let's talk about that and let me show you the difference between talking to.
Speaker A
I'm going to use ChatGPT here in incognito.
Speaker A
Versus using your own context.
Speaker A
So I'm going to share my Obsidian vault.
Speaker A
In a moment.
Speaker A
But let's give this a try.
Speaker A
Okay.
Speaker A
So what we can do with ChatGPT is I'm interested in creativity and frameworks.
Speaker A
The two seems contradictory.
Speaker A
Help me see how to combine these two interests together into a career that makes sense.
Speaker A
Let's see what it says.
Speaker A
I haven't really shared anything.
Speaker A
But let's see what it says.
Speaker A
Blah, blah, blah.
Speaker A
Blah, blah, blah.
Speaker A
Where is the.
Speaker A
Design thinking, product strategy.
Speaker A
Okay.
Speaker A
Creative direction with systems.
Speaker A
Creative director.
Speaker A
Brand strategist.
Speaker A
I mean.
Speaker A
I guess, but, you know, for example, the career I'm thinking about is not like these roles.
Speaker A
But of course, I didn't tell it.
Speaker A
So, okay, makes sense.
Speaker A
It's saying something like this.
Speaker A
Yeah.
Speaker A
Okay, so I'm kind of in this category, the framework builder, thought leader kind of.
Speaker A
Learning design.
Speaker A
I mean, these are things I'm interested in.
Speaker A
But just seeing this list doesn't really help me say, okay.
Speaker A
So how do I know which one's for me?
Speaker A
And I guess they all sound good.
Speaker A
But how do I prioritize them?
Speaker A
Let me ask it.
Speaker A
How do I prioritize this?
Speaker A
Okay.
Speaker A
Great question, this is the right place to feel stuck.
Speaker A
Prioritization is where creativity.
Speaker A
Frameworks either click.
Speaker A
Or spiral.
Speaker A
Let's make it simple and actionable.
Speaker A
I'll give you one clear prioritization model.
Speaker A
Not a buffet.
Speaker A
The rule of thumb, anchor this.
Speaker A
Prioritize the intersection, not the options.
Speaker A
You're not choosing between.
Speaker A
Creativity versus structure.
Speaker A
Learning versus doing.
Speaker A
Ask.
Speaker A
Am I turning intuition into something legible?
Speaker A
Good signs.
Speaker A
Explaining why something works.
Speaker A
If it's only.
Speaker A
Okay.
Speaker A
Let's say something like this.
Speaker A
And see what kind of recommendation it will come up with.
Speaker A
Okay.
Speaker A
So five minutes later, here is what it shared.
Speaker A
I really recognize a lot of these books.
Speaker A
But.
Speaker A
But.
Speaker A
These, these are good.
Speaker A
I mean, definitely know some of these.
Speaker A
The recommendations.
Speaker A
I've written them.
Speaker A
But, um, all right.
Speaker A
That's good to know.
Speaker A
I would now come back to the question of the career.
Speaker A
Okay, let's say something like this.
Speaker A
And see what kind of recommendation it will come up with.
Speaker A
Okay, here we go.
Speaker A
Um, it's mentioned.
Speaker A
Okay, so first of all, the positioning.
Speaker A
It knows what I'm doing.
Speaker A
I'm glad this showed up.
Speaker A
As in, yeah, great.
Speaker A
Um, the gap you could fill, productivity influencer, tactics without theory.
Speaker A
Academic, theory without practice.
Speaker A
AI tool builders.
Speaker A
Tools without methodology.
Speaker A
Corporate L&D.
Speaker A
Generic, not practitioner-led.
Speaker A
Business authors.
Speaker A
Frameworks without depth.
Speaker A
Yeah, these are great, the gap is someone who can articulate how knowledge work actually works.
Speaker A
With philosophical depth, practical frameworks, and AI fluency.
Speaker A
And translate that into teachable, applicable methods.
Speaker A
Honestly, this is how I feel.
Speaker A
I mean, I never really mentioned this in my notes.
Speaker A
So it's kind of crazy that it articulated this.
Speaker A
Um, building intellectual property through writing.
Speaker A
I am writing a book currently.
Speaker A
I don't know about the book to speaking to corporate workshop to advisory type.
Speaker A
Revenue stream.
Speaker A
Yes, I know.
Speaker A
These are awesome people do it.
Speaker A
Okay.
Speaker A
The educator builder.
Speaker A
Okay.
Speaker A
The applied researcher.
Speaker A
Work inside AI companies.
Speaker A
I don't really want to do that.
Speaker A
Hybrid phase one.
Speaker A
Now, 18 months.
Speaker A
Deepen your methodology.
Speaker A
Continue content but shift towards original frameworks.
Speaker A
Not just explaining others.
Speaker A
Run small paid workshops to test what resonates.
Speaker A
Document everything, this becomes your book material.
Speaker A
Phase two.
Speaker A
18 months to three years.
Speaker A
Productize.
Speaker A
Launch a signature course cohort on your methodology.
Speaker A
Or write the book and use it to unlock speaking corporate.
Speaker A
Build a small team to scale delivery.
Speaker A
Phase three.
Speaker A
Three plus years.
Speaker A
Choose your leverage.
Speaker A
License your methodology to organizations.
Speaker A
Build advise AI tools for knowledge work.
Speaker A
Or go deep on research, academic partnerships, think tank, etc.
Speaker A
Clarifying questions.
Speaker A
Before I go deeper, it would help to know.
Speaker A
What's your current revenue model, YouTube ads, sponsorships, courses, consulting?
Speaker A
Do you want to scale yourself or scale a team?
Speaker A
Is being known important or would you take impact without visibility?
Speaker A
How do you feel about corporate clients versus individual learners?
Speaker A
These will sharpen which path fits you.
Speaker A
Worked for 39 seconds.
Speaker A
Okay, before I continue.
Speaker A
I just realized now you're just indulging me.
Speaker A
In my me figuring out my future plans.
Speaker A
So if you're not interested in that, honestly, you can stop right here.
Speaker A
Because I'm just going to continue doing this.
Speaker A
For the next little bit.
Speaker A
But the point is bringing it back for having a note-taking system.
Speaker A
Right, what really continued to surprise me is how good LLMs are at taking information.
Speaker A
And being able to pull out some really great ideas.
Speaker A
And being able to identify, especially with connected notes.
Speaker A
The core notes that matter to you.
Speaker A
So what it's able to do is without me needing to review all of my notes.
Speaker A
It's able to pick out things like this.
Speaker A
So if you are working on something and you want to build on top of it, you want the ideas to compound and not just build sand castles.
Speaker A
Then you have to document it.
Speaker A
It has to be explicit.
Speaker A
So that AI can work with it.
Speaker A
And help you see the things that you as a human being have, you know, natural weaknesses.
Speaker A
Like digesting a lot of information in a short amount of time.
Speaker A
Right, so being able to engage in that and have your curiosity and your questioning mind.
Speaker A
And your judgment mind and your editorial mind come in.
Speaker A
Instead of just say, hey, you know, give me something.
Speaker A
And then now, oh, this is not like right.
Speaker A
So let me move on and have that task sit in its current state.
Speaker A
Is going to be so much more different than if you had these contexts.
Speaker A
You can go deeper.
Speaker A
So with that, I hope you enjoyed this video.
Speaker A
If you have any questions.
Speaker A
Let me know down below, I will link the Claude code setup information down there as well.
Speaker A
All right, I'll see you in the next video.
Topics:AInote-takingcontext engineeringprompt engineeringlarge language modelsLLMknowledge managementcommunicationframeworksObsidian
