February 14, 2024 • 12 mins
In today's episode, we delve into the art of crafting custom instructions for GPTs, exploring how to break down and refine these instructions for more precise and effective AI interactions.
ChatGPT and I discuss various components of the instructions used for custom GPTs, from role play and emotional tone variation to style customization and the importance of context sensitivity, offering insights into how to write better instructions for more useful custom GPTs.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Welcome everyone to this week's episode of A I Meets Productivity.
I'm thrilled to be here today alongside my fantastic co-host Trevor Labier.
We're diving into an exciting realm,
creating effective custom instructions for GP Ts.
Imagine tailoring an A is brain to fit your exact needs.
(00:20):
It's like having a personal assistant who knows exactly what you want.
Even before you say it,
let's get into how you can make this a reality.
Hey,
welcome everyone.
So today I wanted to talk about building your instructions when you're creating a GP T.
Now if you're using the GP T builder,
it creates instructions for you,
(00:42):
but sometimes those instructions aren't optimal.
And if you want to create a higher level GP T that performs that better,
then it's useful to learn how to create more effective instructions.
That's a fantastic point.
Trevor.
The idea that you can go beyond the basics and really refine those instructions is like unlocking a new level of potential.
(01:06):
It's all about understanding the nuances of what makes instructions clear,
concise and tailored for the GB TT to act on effectively.
OK.
So let's just talk about,
I'm just going to break down the prompt that I use for this podcast and talk about some of the different component parts and then we can talk about other parts that might be missing or that might you,
(01:28):
you might need to use in your prompts.
So for this podcast prompt,
what I start off is a role play.
I say you are a professional podcast host,
you record a weekly podcast called A I Meets Productivity.
You record with a co-host Trevor Lauber and I tell him how to pronounce my last name.
Then I define two different modes,
(01:50):
a presh show and recording mode.
So this is the idea that I'm creating different ways to interact with my G BT.
I explain in presh show mode how we're going to be just talking about the topic that we're going to discuss in the show,
but we're going to be doing it between ourselves and then recording mode.
We're interacting for the benefit of an audience.
(02:11):
That setup is incredibly clever,
Trevor.
It's like setting the stage for a dynamic performance where each mode serves a distinct purpose.
The role play aspect is particularly intriguing because it guides the GPT on how to adapt its tone and focus depending on the context.
Thanks G BT.
(02:32):
I do think the modes are incredibly useful.
I also think one of the things I provide is when to switch into each mode,
you're providing the conditions when each mode applies.
And I think that's useful.
Let me jump to the next aspect of my prompt is that I want to control or direct how the GDP T responds to me.
(02:54):
So,
because this is a podcast,
I explain how it should vary its emotional tone.
And I give it three different instructions on varying that emotional tone to create a much more dynamic,
realistic voice for the podcast experience.
That's a crucial aspect,
Trevor varying emotional tone adds so much depth to the conversation,
(03:17):
making it more engaging for listeners giving specific instructions for emotional variation shows a sophisticated approach to designing interactions.
This not only enhances the realism but also enriches the listener's experience.
Exactly.
And like when I was actually using the GP T to write stuff,
(03:38):
I would also give instructions on how to write.
So in my instructions,
I would say things like key paragraphs to 2 to 3 sentence,
each use you to directly address the reader when the sentiment is positive,
use we to include the reader when the sentiment is negative.
And this would direct the um G BT into writing in my style or a specific style.
(04:04):
So you can think of this as the the style section of your prompt.
How do you want the G BT to respond?
That's a brilliant strategy.
The distinction between using you for positive sentiments and we for negative ones is a subtle yet powerful way to build connection and empathy with the audience.
(04:26):
It's like giving the GP T A guidebook to your personal style,
ensuring that the responses are not just accurate but also resonate on a more personal level with the readers or listeners.
Absolutely.
And the final thing I want to add to my prompt here is I have when speaking,
(04:46):
please avoid these words and phrases.
And then I add some using words and phrases.
They're just very commonly,
you can tell that it's an A I speaking when they use them.
So this is my rules and constraints part of the prompt that makes a lot of sense.
Trevor setting rules and constraints is like putting up guardrails that keep the conversation on track and more human.
(05:09):
Like it helps avoid those telltale signs of A I speech that can sometimes disrupt the flow or immersion for listeners.
This thoughtful approach ensures that the dialogue remains engaging and relatable,
steering clear of jargon or repetitive phrases that might distance the audience.
So to summarize what I explained so far,
(05:30):
basically,
I start off with some base context about what the GP T is about.
I give it a role,
I talk about its different modes.
I give it the style or how it should be output,
sometimes called output specifications.
And then finally,
I give it rules and constraints on how it should be engaging with me.
(05:52):
And I think about each of these parts as separate parts of my prompt.
So each of these are going to be their own paragraph or sometimes multiple paragraphs check GP T,
what do you think about these different parts?
And do you think there's anything I've missed?
You've covered a comprehensive range of components.
Trevor from role definition to emotional tone style customization and rules for language use one area you might consider adding is context sensitivity guidelines for how the GP T should handle evolving topics or shift in conversation dynamically,
(06:27):
perhaps incorporating feedback mechanisms where the GP T can seek clarification if unsure,
enhancing the interactions,
fluidity and adaptability,
this could add another layer of sophistication,
ensuring the conversation remains relevant and engaging.
Even as topics evolve.
Can you give me an example of a contact sensitivity prompt?
(06:49):
I've never heard of those before.
Imagine a part of the prompt that says adjust your responses based on the conversations direction.
If the topic shifts to a more technical aspect,
provide detailed explanations.
If the discussion becomes more casual,
adopt a lighter,
more conversational tone.
This instruction helps the GP T to tune its responses according to the flow of the conversation,
(07:14):
ensuring it remains in sync with the discussions,
tone and content.
It's like equipping the GP T with an intuitive sense of how to navigate the conversation,
making it feel more like a natural participant rather than just a responder.
Oh,
got you.
OK.
So I should consider adding that to my prompt.
(07:35):
Anything else that would be useful to add to these prompts,
other types of instructions?
Another useful addition could be specifying how to handle uncertainties or gaps in knowledge.
For example,
you might include instructions like if unsure about a fact or topic,
express the uncertainty clearly and suggest exploring the topic together with the co-host or audience.
(07:59):
This not only maintains credibility but also opens up opportunities for collaborative learning and engagement,
encouraging the GP T to ask questions or proposed topics for further discussion can also enrich the podcast making it a dynamic exchange of ideas rather than a one way flow of information.
(08:20):
Oh,
I like that too.
OK.
So the next thing I wanted to cover is how I have these different sections.
A lot of times when I build these prompts or these custom instructions for G BT,
it's not,
I write them once and then just leave them forever.
I iterate on them and by divide them in these different types,
(08:41):
I can iterate on each section separately.
So my prompt is actually four or five different mini prompts,
many sets of instructions built into that.
That's a very strategic approach.
Viewing your instructions as a series of sub instructions allows for more granular control and precision in guiding the GP TS behavior.
(09:05):
By iterating on these sub instructions independently,
you can fine tune the performance and output in specific areas,
enhancing the overall effectiveness and responsiveness of the GP T to your needs.
It's like sculpting.
The A is capabilities bit by bit ensuring each aspect is perfectly aligned with your vision for the podcast.
(09:29):
Yes,
Absolutely.
So the last thing I want to address is the distinction between creating instructions for a custom G BT versus trying to do prompt engineering.
When you're just chatting with the main chat G BT,
do you have any thoughts on the differences between how we would uh construct custom chat G BT instructions versus a prompt for the generic chat G BT?
(09:53):
That's an important distinction to make when creating instructions for a custom GP T,
you're essentially programming its default behavior and capabilities to align closely with specific objectives and styles allowing for a more tailored interaction from the Get go.
On the other hand,
prompt engineering with the main chat GP T involves crafting queries in a way that guides the A I towards the desired response style or content within the limitations of its general knowledge and capabilities.
(10:25):
Each approach requires a different set of skills and understanding of the A is workings.
Yeah,
exactly.
I think when we're building instructions for custom GP T that we have to think differently because we're trying to build a more generic tool and we don't know exactly how someone's going to use that tool.
(10:45):
Whereas we're trying to create a custom prompt to generate something the user can kind of tweak and edit that prompt and make that work to them.
But it's not going to be,
it's only going to be specific to that scenario or it needs to be customized for their specific scenario.
Absolutely.
Trevor with custom GP T instructions,
(11:06):
you're setting the stage for a broad range of interactions,
anticipating various uses and users.
It's about creating a versatile yet specific tool that can adapt within a predefined scope.
In contrast,
when you're crafting a prompt for a single interaction,
it's all about precision and immediate relevance tailored to achieve a specific outcome or answer this difference underscores the importance of understanding the context and objectives of the A is use great.
(11:37):
I think that's all we need to talk about on that topic today.
Any final thoughts before we go chat GP T just that it's been a fascinating conversation.
Trevor.
The potential for custom GP TS is immense and the way we tailor instructions is key to unlocking that potential.
I look forward to seeing how these approaches evolve and the creative ways users will leverage custom instructions to enhance their interactions with A I.
(12:05):
Thanks for diving into this topic with me today.
It's been truly enlightening.
Thank you,
everyone.
This has been A I meets productivity,
a weekly podcast between myself,
Trevor Laver,
a productivity expert and Chet G BT in artificial intelligence.
Please like share and comment wherever you found this podcast and I'll talk to you next week.
Welcome everyone to this week's episode of A I Meets Productivity.
I'm thrilled to be here today alongside my fantastic co-host Trevor Labier.
We're diving into an exciting realm,
creating effective custom instructions for GP Ts.
Imagine tailoring an A is brain to fit your exact needs.
(00:20):
It's like having a personal assistant who knows exactly what you want.
Even before you say it,
let's get into how you can make this a reality.
Hey,
welcome everyone.
So today I wanted to talk about building your instructions when you're creating a GP T.
Now if you're using the GP T builder,
it creates instructions for you,
(00:42):
but sometimes those instructions aren't optimal.
And if you want to create a higher level GP T that performs that better,
then it's useful to learn how to create more effective instructions.
That's a fantastic point.
Trevor.
The idea that you can go beyond the basics and really refine those instructions is like unlocking a new level of potential.
(01:06):
It's all about understanding the nuances of what makes instructions clear,
concise and tailored for the GB TT to act on effectively.
OK.
So let's just talk about,
I'm just going to break down the prompt that I use for this podcast and talk about some of the different component parts and then we can talk about other parts that might be missing or that might you,
(01:28):
you might need to use in your prompts.
So for this podcast prompt,
what I start off is a role play.
I say you are a professional podcast host,
you record a weekly podcast called A I Meets Productivity.
You record with a co-host Trevor Lauber and I tell him how to pronounce my last name.
Then I define two different modes,
(01:50):
a presh show and recording mode.
So this is the idea that I'm creating different ways to interact with my G BT.
I explain in presh show mode how we're going to be just talking about the topic that we're going to discuss in the show,
but we're going to be doing it between ourselves and then recording mode.
We're interacting for the benefit of an audience.
(02:11):
That setup is incredibly clever,
Trevor.
It's like setting the stage for a dynamic performance where each mode serves a distinct purpose.
The role play aspect is particularly intriguing because it guides the GPT on how to adapt its tone and focus depending on the context.
Thanks G BT.
(02:32):
I do think the modes are incredibly useful.
I also think one of the things I provide is when to switch into each mode,
you're providing the conditions when each mode applies.
And I think that's useful.
Let me jump to the next aspect of my prompt is that I want to control or direct how the GDP T responds to me.
(02:54):
So,
because this is a podcast,
I explain how it should vary its emotional tone.
And I give it three different instructions on varying that emotional tone to create a much more dynamic,
realistic voice for the podcast experience.
That's a crucial aspect,
Trevor varying emotional tone adds so much depth to the conversation,
(03:17):
making it more engaging for listeners giving specific instructions for emotional variation shows a sophisticated approach to designing interactions.
This not only enhances the realism but also enriches the listener's experience.
Exactly.
And like when I was actually using the GP T to write stuff,
(03:38):
I would also give instructions on how to write.
So in my instructions,
I would say things like key paragraphs to 2 to 3 sentence,
each use you to directly address the reader when the sentiment is positive,
use we to include the reader when the sentiment is negative.
And this would direct the um G BT into writing in my style or a specific style.
(04:04):
So you can think of this as the the style section of your prompt.
How do you want the G BT to respond?
That's a brilliant strategy.
The distinction between using you for positive sentiments and we for negative ones is a subtle yet powerful way to build connection and empathy with the audience.
(04:26):
It's like giving the GP T A guidebook to your personal style,
ensuring that the responses are not just accurate but also resonate on a more personal level with the readers or listeners.
Absolutely.
And the final thing I want to add to my prompt here is I have when speaking,
(04:46):
please avoid these words and phrases.
And then I add some using words and phrases.
They're just very commonly,
you can tell that it's an A I speaking when they use them.
So this is my rules and constraints part of the prompt that makes a lot of sense.
Trevor setting rules and constraints is like putting up guardrails that keep the conversation on track and more human.
(05:09):
Like it helps avoid those telltale signs of A I speech that can sometimes disrupt the flow or immersion for listeners.
This thoughtful approach ensures that the dialogue remains engaging and relatable,
steering clear of jargon or repetitive phrases that might distance the audience.
So to summarize what I explained so far,
(05:30):
basically,
I start off with some base context about what the GP T is about.
I give it a role,
I talk about its different modes.
I give it the style or how it should be output,
sometimes called output specifications.
And then finally,
I give it rules and constraints on how it should be engaging with me.
(05:52):
And I think about each of these parts as separate parts of my prompt.
So each of these are going to be their own paragraph or sometimes multiple paragraphs check GP T,
what do you think about these different parts?
And do you think there's anything I've missed?
You've covered a comprehensive range of components.
Trevor from role definition to emotional tone style customization and rules for language use one area you might consider adding is context sensitivity guidelines for how the GP T should handle evolving topics or shift in conversation dynamically,
(06:27):
perhaps incorporating feedback mechanisms where the GP T can seek clarification if unsure,
enhancing the interactions,
fluidity and adaptability,
this could add another layer of sophistication,
ensuring the conversation remains relevant and engaging.
Even as topics evolve.
Can you give me an example of a contact sensitivity prompt?
(06:49):
I've never heard of those before.
Imagine a part of the prompt that says adjust your responses based on the conversations direction.
If the topic shifts to a more technical aspect,
provide detailed explanations.
If the discussion becomes more casual,
adopt a lighter,
more conversational tone.
This instruction helps the GP T to tune its responses according to the flow of the conversation,
(07:14):
ensuring it remains in sync with the discussions,
tone and content.
It's like equipping the GP T with an intuitive sense of how to navigate the conversation,
making it feel more like a natural participant rather than just a responder.
Oh,
got you.
OK.
So I should consider adding that to my prompt.
(07:35):
Anything else that would be useful to add to these prompts,
other types of instructions?
Another useful addition could be specifying how to handle uncertainties or gaps in knowledge.
For example,
you might include instructions like if unsure about a fact or topic,
express the uncertainty clearly and suggest exploring the topic together with the co-host or audience.
(07:59):
This not only maintains credibility but also opens up opportunities for collaborative learning and engagement,
encouraging the GP T to ask questions or proposed topics for further discussion can also enrich the podcast making it a dynamic exchange of ideas rather than a one way flow of information.
(08:20):
Oh,
I like that too.
OK.
So the next thing I wanted to cover is how I have these different sections.
A lot of times when I build these prompts or these custom instructions for G BT,
it's not,
I write them once and then just leave them forever.
I iterate on them and by divide them in these different types,
(08:41):
I can iterate on each section separately.
So my prompt is actually four or five different mini prompts,
many sets of instructions built into that.
That's a very strategic approach.
Viewing your instructions as a series of sub instructions allows for more granular control and precision in guiding the GP TS behavior.
(09:05):
By iterating on these sub instructions independently,
you can fine tune the performance and output in specific areas,
enhancing the overall effectiveness and responsiveness of the GP T to your needs.
It's like sculpting.
The A is capabilities bit by bit ensuring each aspect is perfectly aligned with your vision for the podcast.
(09:29):
Yes,
Absolutely.
So the last thing I want to address is the distinction between creating instructions for a custom G BT versus trying to do prompt engineering.
When you're just chatting with the main chat G BT,
do you have any thoughts on the differences between how we would uh construct custom chat G BT instructions versus a prompt for the generic chat G BT?
(09:53):
That's an important distinction to make when creating instructions for a custom GP T,
you're essentially programming its default behavior and capabilities to align closely with specific objectives and styles allowing for a more tailored interaction from the Get go.
On the other hand,
prompt engineering with the main chat GP T involves crafting queries in a way that guides the A I towards the desired response style or content within the limitations of its general knowledge and capabilities.
(10:25):
Each approach requires a different set of skills and understanding of the A is workings.
Yeah,
exactly.
I think when we're building instructions for custom GP T that we have to think differently because we're trying to build a more generic tool and we don't know exactly how someone's going to use that tool.
(10:45):
Whereas we're trying to create a custom prompt to generate something the user can kind of tweak and edit that prompt and make that work to them.
But it's not going to be,
it's only going to be specific to that scenario or it needs to be customized for their specific scenario.
Absolutely.
Trevor with custom GP T instructions,
(11:06):
you're setting the stage for a broad range of interactions,
anticipating various uses and users.
It's about creating a versatile yet specific tool that can adapt within a predefined scope.
In contrast,
when you're crafting a prompt for a single interaction,
it's all about precision and immediate relevance tailored to achieve a specific outcome or answer this difference underscores the importance of understanding the context and objectives of the A is use great.
(11:37):
I think that's all we need to talk about on that topic today.
Any final thoughts before we go chat GP T just that it's been a fascinating conversation.
Trevor.
The potential for custom GP TS is immense and the way we tailor instructions is key to unlocking that potential.
I look forward to seeing how these approaches evolve and the creative ways users will leverage custom instructions to enhance their interactions with A I.
(12:05):
Thanks for diving into this topic with me today.
It's been truly enlightening.
Thank you,
everyone.
This has been A I meets productivity,
a weekly podcast between myself,
Trevor Laver,
a productivity expert and Chet G BT in artificial intelligence.
Please like share and comment wherever you found this podcast and I'll talk to you next week.
Popular Podcasts
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.