Unleash ChatGPT Potential with the SCRIBE Method

Summary

This video script delves into the Scribe method for structuring prompts to enhance interactions with AI, particularly in the context of live courses focusing on practical applications and advanced prompt engineering. It explains the Scribe acronym—Specify, Role, Context, Responsibility, Instructions, Bentor, and Evaluate—offering a detailed guide to crafting effective AI prompts. Additionally, the script highlights the significance of markdown for clarity and structure in prompts, providing examples and best practices. It concludes by illustrating the application of the Scribe method, including a unique 'criticize me' step, to improve AI-generated content, and introduces an advanced prompting course for further learning.

Takeaways

• 📝 The Scribe method for prompt engineering is essential for structuring prompts effectively, emphasizing Specify, Role, Context, Responsibility, Instructions, Bentor, and Evaluate.
• 📈 Live courses focus on practical application, hands-on projects, and feedback, with a follow-up course named 'ChatGPT and Beyond: Advanced Prompt Engineering' starting on February 19, 2024.
• 📚 The Scribe method encourages specifying a role for the AI, such as acting as a blog writer or social media manager, to guide its response direction.
• 🔧 Context is crucial in prompting; sharing relevant background information leads to more appropriate AI responses.
• 📌 Instructions should be detailed and broken down into step-by-step guidance to achieve desired outcomes.
• 💬 Engaging in a follow-up conversation with the AI to refine initial responses and provide feedback enhances the quality of results.
• 📄 Evaluating AI's responses for accuracy is important, including asking the AI to assess its own effectiveness.
• 🏆 Markdown is used in prompts to add formatting elements, helping communicate with the AI in a universally understood language.
• 🚀 Advanced prompting techniques, like 'criticize me mode,' involve instructing the AI to critique its own output to improve quality.
• 📝 Positive language in prompts (e.g., specifying what to do rather than what not to do) can significantly enhance interaction with the AI.

Q & A

What is the start date for the second cohort of the course?

The second cohort of the course starts on January 29.

What role will the speaker play in the upcoming cohort?

The speaker will help learners as a technical adviser in the upcoming cohort.

What is the 'Scribe' method in the context of this course?

The 'Scribe' method provides a framework for structuring prompts, including specifying a role, context, responsibility, instructions, banter, and evaluation.

What is the name of the follow-up course mentioned, and when does it start?

The follow-up course is called 'ChatGPT and Beyond: Advanced Prompt Engineering', and it starts on February 19, 2024.

Why is domain expertise important in prompting?

Domain expertise is crucial for using specific language to condition the AI model and steer its responses in the right direction.

How does the Scribe method suggest using roles with large language models?

The Scribe method suggests giving large language models a specific role, like acting as a blog writer or social media manager, and specifying the style or tone for the model to inherit.

What is Markdown, and why is it used in prompts?

Markdown is a lightweight markup language used to add formatting elements to plain text documents, helping communicate universally understood terms to the AI model.

How can headings in Markdown enhance structuring of prompts?

Headings in Markdown, denoted by hashtags, help establish structural hierarchy in prompts, with different levels of headings indicating the importance and organization of content.

What is the 'criticize me mode' mentioned in the script?

The 'criticize me mode' is an instructional approach where the AI is asked to critique its own work, aiming to improve the quality of its outputs by identifying potential flaws.

How does the script suggest improving the effectiveness of AI prompts?

The script suggests using positive, affirmative language in prompts rather than negative language to more effectively guide the AI's responses.

Outlines

00:00

📚 Introducing the Scribe Method in Prompt Engineering

The speaker discusses the preparation for the second cohort of an AI and prompt engineering course starting on January 29, highlighting the role as a technical adviser and the students' fondness for the scribe prompting method. Despite this method not being explicitly shared before, it's emphasized for structuring prompts effectively. The upcoming course, 'ChatGPT and Beyond: Advanced Prompt Engineering,' is mentioned, starting on February 19, 2024, focusing on advanced topics like meta multi-agent prompts. The scribe method, which stands for Specify, Context, Responsibility, Instructions, Bentor, and Evaluate, is detailed as a framework to improve interaction with AI by defining roles, providing context, outlining tasks, giving detailed instructions, and engaging in iterative refinement to enhance output quality. Additionally, the importance of domain expertise, the use of specific language, and a reference guide for different contexts and prompts are discussed. Markdown's role in structuring prompts for better AI understanding and interaction is explained, highlighting its significance in the AI prompting process.

05:02

🔧 Utilizing Markdown for Structured AI Prompts

The second paragraph elaborates on the practical application of markdown in structuring AI prompts, with a focus on enabling markdown detection in tools and its impact on organizing content. The use of headings to define roles and context, and how markdown's structure is recognized across different platforms like Obsidian, is demonstrated. The segment further expands on structuring detailed instructions for AI, including specifying roles, providing context, detailing responsibilities, and breaking tasks into step-by-step instructions. A unique 'criticize me' mode is introduced, encouraging the AI to critically evaluate its work, thereby enhancing the refinement process. The effectiveness of positive language in prompts over negative instructions is highlighted, showcasing how specific prompting can lead to improved AI interactions. Finally, the segment mentions a successful prompt developed by a team member, which enhances the ChatGPT experience, illustrating the potential of advanced prompting techniques in maximizing AI utility.

Keywords

💡Scribe method

The Scribe method is a structured approach to crafting prompts for large language models, emphasizing the importance of specifying a role, context, responsibility, instructions, bentor, and evaluation. This method is aimed at improving the effectiveness of interactions with AI by providing a clear framework for users to communicate their needs. In the video script, it is highlighted as a helpful framework in structuring prompts to guide the AI in generating relevant and accurate responses. For instance, acting as a blog writer with a specified style or tone is an example of applying the Scribe method.

💡Prompting

Prompting refers to the process of structuring inputs or questions to an AI model to elicit specific responses. In the video script, it's emphasized as a crucial skill, especially when utilizing large language models like GPT for various tasks. The effectiveness of prompting lies in its ability to guide the AI's response in a direction that is useful for the user, with the Scribe method being presented as a powerful strategy for achieving this.

💡Markdown

Markdown is introduced as a lightweight markup language that can be used to format text documents. In the context of the video, markdown is used in prompts to communicate with AI in universally understood terms, improving the clarity and structure of requests. For example, using hashtags to denote headings helps the AI understand the hierarchy and significance of various parts of the prompt, thereby facilitating better-structured and more relevant responses.

💡Large language models

Large language models (LLMs) like GPT are AI systems trained to predict and generate text based on the input they receive. The video script discusses the importance of structuring prompts effectively to guide these models in generating useful and accurate responses. LLMs' ability to process and produce human-like text makes them versatile tools for a wide range of applications, from writing assistance to data analysis.

💡Role prompting

Role prompting is a technique highlighted in the Scribe method where the user assigns a specific role for the AI, such as 'Blog writer' or 'Social media manager'. This specificity helps the AI to understand the context and desired tone or style of the output, leading to more targeted and appropriate responses. It's a practical example of how giving AI a clear context can significantly improve the relevance and quality of its outputs.

💡Meta multi-agent prompts

Meta multi-agent prompts refer to advanced prompting techniques that involve creating prompts that simulate the interaction between multiple agents or perspectives within a single AI session. This concept, mentioned as part of an advanced course, illustrates the progression from basic to more sophisticated methods of interacting with AI, enabling users to tackle complex scenarios and generate nuanced responses.

💡Feedback loop

The feedback loop is a critical component of the Scribe method, involving engaging with the AI in a conversation to refine its responses. This iterative process of providing feedback and asking clarifying questions ensures that the AI's outputs become more accurate and relevant over time. It's a practical approach to collaborative work with AI, leading to higher quality results.

💡Evaluate

Evaluation is the final step in the Scribe method, where the user reviews the AI's responses for accuracy and effectiveness. This step may include asking the AI to assess its own performance, thereby helping to reduce errors and improve future responses. In the video, it's emphasized as a crucial part of ensuring the reliability and utility of AI-generated content.

💡Structural hierarchy

Structural hierarchy in prompts, facilitated by markdown formatting, allows users to organize their instructions in a clear and hierarchical manner. By using different levels of headings (from H1 to H6), users can delineate main topics, subtopics, and detailed instructions. This helps the AI understand the relative importance and relationships between various parts of the prompt, leading to more organized and focused responses.

💡CHPT Master reference guide

The CHPT Master reference guide is mentioned as a resource compiled to assist users in crafting effective prompts by providing useful terms, contexts, and role-prompting personas. This guide embodies the practical application of the Scribe method, offering users a tool to navigate areas outside their domain expertise and leverage AI capabilities more effectively.

Highlights

Preparation for second cohort starting on January 29 with technical adviser assistance.

Introduction of the Scribe method for structuring prompts.

Live courses focused on practical applications, hands-on projects, and feedback.

Announcement of follow-up course 'ChatGPT and Beyond: Advanced Prompt Engineering' starting on February 19, 2024.

Emphasis on the importance of specifying a role for effective prompting.

Highlighting the necessity of domain expertise in prompting.

Introduction of a Master reference guide with useful terms and prompts.

Explanation of the significance of context in generating appropriate AI responses.

The importance of clear instructions and step-by-step task breakdown.

The role of iterative feedback in refining AI responses.

The value of evaluating AI's responses for accuracy and effectiveness.

Use of markdown in prompts to structure and clarify requests.

Example of structuring prompts with markdown for clarity.

Introduction of 'criticize me' mode to enhance prompt feedback.

Technique to improve AI interactions by using positive affirmative language.

Introduction of a prompt that transforms ChatGPT into an AutoGPT experience.

Transcripts

I was going through our a anbt for

everyone course material in preparation

for second cohort starting on January 29

where I will come to help Learners as a

technical adviser you check this one off

your list okay this is done while

reviewing testimonials and past student

projects I remembered how much students

love scribe prompting method I also

realized that I have never explicitly

shared or explained scribe method first

of all prompting is all about Str

structure and V scribe method provides a

helpful framework in structuring your

prompts our live courses are all about

practical applications Hands-On projects

and feedback if by the time you're

watching this video and enrollment has

closed don't worry about it if you're

interested there's going to be follow-up

course called Chad GPT and Beyond

Advanced prompt engineering which will

start on 19th of February 2024 in that

course we go as deep as creating meta

multi-agent prompts but for now I want

to share with you the Scribe method as

well as talk to you about importance of

markdown scribe stands for specify a

role context responsibility instructions

bentor and evaluate let's dig just one

by

one it's very useful to give large

language modu a specific role like for

example act as a Blog writer as act as a

social media manager and so on here you

can also add the preferred style or tone

that you want large language model to

inherit large language models are

predicting one word at a time based on

the input the specific role acts as a

key for a large language model to look

which doors it should go to which would

match this context and here's one

universal truth when it comes to

prompting you as a user have to have

domain expertise and use specific

language to condition the AI model and

steer its responses to the right

direction but if you want to use AI to

help you with task which are outside of

your domain expertise we put together

chpt Master reference guide with useful

terms for different context and prompts

for example how to write like a human

different marketing styles are a bunch

of role prompting personas you can find

the link in the description box below

and check it out for yourself okay let's

talk about context share any relevant

background context details and even

examples to help large language model to

generate an appropriate response more

specific and the more context you

provide the better response you might

expect you need to clearly outline the

task that you want large language model

to perform be as specific as possible in

describing what you want and most

importantly what success would look

like provide large language model with

detailed instructions that you wanted to

follow in order to achieve that outcome

and break the complicated task into step

by step instructions to help the large

language model to guide you and for each

of these you can use CHT to help you for

example breaking a task just literally

tell what's the big task and ask it to

break it in stepbystep instructions copy

those and input of them in here that's

it you have to engage with large

language model in a follow-up

conversation to refine its initial

responses ask clarifying questions or

provide feedback this what we call enter

back and forth conversation and like

more collaborative work leads to the

higher quality

results and always always evaluate

review their responses for accuracy you

can ask AI to even evaluate its

Effectiveness and accuracy this kind of

metal level analysis helps the model to

strengthen its generation abilities and

reduce hallucinations before I show you

scribe in action and more advanced

features I want to explain why we like

to use markdown in our prompts and what

markdown is in the first place markdown

is a lightweight markup language that

you can use to add formatting elements

to the plain text documents when it

comes to prompting this markdown

language allows us to communicate to AI

universally understood terms plus AI

model was already trained on this terms

and it knows them for example instead of

writing title you can use what is known

as headings in markdown heading one is

expressed with one hashtag this way a

model understands that a hashtag n

combination of word is a title or

heading sentences in your PR which has

hashtag in front of them will act as a

titles and the text underneath it as

part of one block now if you want to

establish structural hierarchy or

substructures you can use heading two

which are two hashtags and we can keep

breaking down the structure down to six

hashtags or heading six in Google Docs

you can visually see how this works

first thing to enable markdown you need

to go to tools preferences and then

automatically detect markdown you need

to check it and say yes and now we are

going to do a hashtag it automatically

goes to heading one you can see it here

let's say specify a roll okay so next we

are going to be like okay this is our

role and space again and then another

hashtag will tell context okay here

we're doing some context so now these

two are H1 so this is the initial

structure but if I

did two hashtags and

said Ru two it becomes a substructure of

the first one but not another rule by

itself now if we if I copy all of this

and I go to the tool like obsidian which

reads markdown you will see that this

text is automatically expressed with a

hashtag and it has different weights and

this is what I mean by universally

understood important to remember that

headings in markdown don't work if you

don't have space between hashtag and

your text so keep that in mind you can

also Define banter and instruct model to

evaluate the results and here I like to

take it one step further and use my

criticize me mode first of all we

specify a all we give context we

describe responsibilities and here we

now breaking up instructions the

instruction two is telling what it's

supposed to do so we are supposed to

craft an email in this case three as you

can see has B here we're instructing CH

what it's supposed to do once it crafted

an email first it has to ask for my

feedback and then ref find the email

content as needed now four is evaluate

stage and I'm going to tweak it a little

bit with you so once salesperson is

satisfied with the format you will

finalize the email and encourage them to

send it to the prospect you will then

ask feedback on the process to improve

your future email crafting five is

really optional but we like to Define

how chbt should interact with us like

what it should say at the beginning of a

chat in this case it says hello I'm your

personal expert salesperson prospecting

assistant not only it's nice start of a

conversation but it also confirms that

Chach B understood so first thing it

followed the instructions of how it

should interact with us second step is

gathering all the necessary information

so it's asking for product details

target audience brand background

previous successful emails I could just

continue until we get to the final email

but I want to show show you the critic

mode so I tweaked a little bit our

scribe method and included one more step

in instructions it is something like

this but you can adjust it however you

want once you have presented initial

draft you need to act as a harsh critic

your job is to critique the email and

convince me why it is bad and why

receiver would not like it the first

step scrap method does is asking for

details in this case because I didn't

want it to come up with all the detail

by myself I told it to come up with all

the details itself and it did and then

we have an email and we have a critiques

and here's like my first impression when

I saw this email I was like it's too

long and here you go in critique it says

the email is a bit too long which might

lead to loss of Interest especially for

busy Prospect and here's a bonus tip

which very few people know or use

actively we all been frustrated with ch

not giving us what we want right right I

see him hide him in the crowd more

effectively hidden in the whole point of

where's Waldo is to find him

interestingly enough telling AI model

what not to do is not as effective as

specifically instructing what to do

basically use positive affirmative

language instead of negative instead of

saying blog title should not exceed more

than 50 characters try say block title

should always be shorter than 50

characters small tweaks like this can

dramatically improve your experience

with AI and just imagine what learning

Advanced prompting can do for you take

for example Joe our genius brain behind

all our courses and prompts he has

developed this prompt which turns chpt

into more like Auto GPT experience and

now this prompt called Professor synip

is ranking as one of the top gpts on GPT

store