Course Handbook
Course Handbook
ChatGPT Essentials
for Clinicians
Franz Wiesbauer, MD, MPH
Table of contents
Abbreviation list 3
ChatGPT basics
Deciphering the science behind LLMs 5
Identifying differences between two popular AI chatbots 9
Learning the core principles of prompting 14
Using role prompting and parameters 19
Conversing with AI 25
Unleashing the
full potential of ChatGPT
Supercharging medical and scientific writing with ChatGPT (part 1) 32
Supercharging medical and scientific writing with ChatGPT (part 2) 38
Understanding basic applications of ChatGPT (part 1) 47
Understanding basic applications of ChatGPT (part 2) 53
Navigating complex texts and contracts with ChatGPT 57
Boosting your learning experience with ChatGPT 63
Appendix
Reference list 77
ChatGPT basics
www.medmastery.com
Deciphering the science behind LLMs
How do large language models (LLMs) like ChatGPT work?
Phases of training
There are two main phases in the training:
1. Pre-training
2. Fine-tuning
Pre-training
Prediction
The model digests these words, learning the rules of language, facts about the
world, and some level of reasoning—all by predicting what comes next. This task
forces the model to learn about grammar, facts about the world, and some level
of reasoning to make accurate predictions.
For example, imagine this scenario. Let’s say it is presented with the text, “A boy
in Florida gazed up at a fruit tree and said, ‘I want to eat the _____.’” The AI model
is then asked to predict the word that comes next.
Comparison
Next comes the comparison stage. The model’s prediction is then compared to
the actual word that comes next in the dataset. So, now it knows that the
sentence, “I want to eat the _____,” actually ended in “orange.”
The model then adjusts its internal parameters, also called weights, to minimize
the loss function the next time it makes a prediction, thereby becoming
more accurate.
This entire process is iterative, just like studying for exams repeatedly until the
student masters the material. In the same way, ChatGPT keeps learning from
its misses, gradually getting better at predicting words and understanding
language.
Fine-tuning
The second main phase of its training is the fine-tuning phase. Think of this as
our AI transitioning from a medical school graduate to a specialized resident.
Here, ChatGPT hones its skills with a more tailored dataset.
Unlike the broader, more diverse dataset used in pre-training, this new dataset in
the fine-tuning phase consists of human-generated prompts and then, after the
AI generates its output, the human’s response to it.
You might ask, How are these prompts and responses created?
So, in essence, the journey of ChatGPT’s training is much like a journey through
medical education. It starts with broad learning in the pre-training phase, or
medical school, and then sharpens its skills in the fine-tuning phase, akin to
a residency. Understanding this process sheds light on the strengths and
weaknesses of large language models.
In fact, sometimes these LLMs will hallucinate. That is, they often make up stuff
and invent citations. So, as a user of ChatGPT and similar tools, it’s your job to
fact-check their output.
It’s also important to note that ChatGPT was only trained on data up to a certain
point in time. For example, when ChatGPT 4 was released, it was only trained
with data up to 2021. It needs to have internet browsing capabilities enabled if
you want to get information out of it that’s more recent.
Alternatively, another option that uses GPT is Bing’s AI. It is most similar to
ChatGPT in creative mode.
You may be wondering, what’s the difference between ChatGPT and Copilot?
Accessing Copilot
To access Copilot, take the following steps:
1. Download the Microsoft Edge browser from Microsoft’s website.
2. Open the Edge browser.
3. Click the Copilot icon on the top right. That will make the Copilot sidebar
appear. Alternatively, you can go to www.bing.com/chat to use it in
full-screen mode.
Copilot sidebar
You may notice that you’re limited in how many turns you get to have in a single
conversation with Copilot. However, you can upgrade your experience—getting
access to longer conversations—if you log into your Microsoft account when
using Copilot.
In the worst-case scenario, if you’re not already using Microsoft’s Edge browser
and you don’t want to download it, there are various third-party plugins available
(also known as browser extensions) that can give you access to Copilot in your
preferred browser (e.g., Chrome).
Copilot settings
Copilot has three settings that you can choose from:
1. More precise
These are kind of like multiple personalities for the AI, and which personality you
choose depends on the type of output you need.
More precise
In this mode, Copilot is all business. It’s optimized to provide you with information
that’s accurate. It also tends to give more concise responses compared to the
other modes. Of course, being an AI, it’s not going to be perfectly accurate 100%
of the time, but this is still your best bet if you want to maximize accuracy.
Examples of the types of questions that are well suited to this mode include
What is the gross domestic product (GDP) of Japan? or What is the chemical
formula for glucose?
More creative
You could describe Copilot’s more creative mode as the artist. This mode isn’t
just about delivering facts—it’s about crafting narratives, telling jokes, and even
writing poems. lt may occasionally stray from factual accuracy but is optimized
for creating engaging and imaginative content.
Examples of the types of questions that are particularly well-suited to this mode
include Can you write me a rhyming poem to invite guests to a birthday party? or Can
you create a drawing of a landscape?
More balanced
In more balanced mode, Bing’s AI is like a diplomat. It attempts to find a middle
ground, maintaining a balance between creativity and precision in its responses.
Take a moment to ask Copilot some questions in each of its three modes so you
can see for yourself what it can do.
Accessing ChatGPT
Now it’s time to try ChatGPT! Here are the steps:
1. To create your account, go to OpenAI’s website.
2. Follow the on-screen instructions to activate your free account.
Now you can use ChatGPT to try out the functionality you’re going to learn about
in upcoming lessons.
After you’re logged into ChatGPT, ask it some of the same questions you asked
Copilot earlier and compare the responses. Keep in mind that those initial
interactions will only scratch the surface of ChatGPT’s true potential.
As you continue through this course, you’ll discover how to unlock even more of
this tool’s remarkable capabilities!
Return to table of contents.
The same applies to AI. For instance, if we ask the AI, Explain heart disease, it
might provide a broad overview that covers all types of heart diseases. This
might be too generic and not quite what we were looking for.
Another very effective strategy is to tell the AI what not to do. In other words, to
put constraints on the task. For example, you could provide ChatGPT with the
text of an article and ask it to do the following: Please provide a summary of the
Or you could ask the AI to summarize text using Hemingway’s rules for writing,
some of which include the following:
• Use short sentences.
• Use vigorous English.
• Use positive language, not negative.
• Minimize the use of adverbs in your writing.
• Use simple language; avoid complex words or phrases.
ChatGPT already has an understanding of those rules, so you don’t need to list
them—try simply instructing ChatGPT as follows: Summarize the following text
using Hemingway’s rules for writing.
Types of prompts
The manner in which you frame your prompts can have a significant impact on
the AI’s responses. Here are four important types of prompts:
1. Zero-shot
2. One-shot
3. Few-shot
4. Example-led
Any other type of singular question like What’s the job of the sinus node? or What’s
the capital of France? could also be categorized as zero-shot prompts.
One-shot
In one-shot prompts, we provide one example before posing our question, like
the example below.
Few-shot
Few-shot prompts involve giving the AI several examples before asking the
question. For instance, let’s say you’ve got a survey response that says, The
doctor was competent, and you tell the AI that this falls into the positive category.
You also have a survey response that says, The waiting time was long, and you tell
the AI that this is negative.
Example-led prompts
This approach is akin to learning how to make a variety of sandwiches—a BLT,
a club sandwich, a grilled cheese, and so on—before being asked to make a
chicken sandwich, even though you haven’t specifically seen how to make one.
The diverse examples of sandwiches you’ve seen allow you to grasp the general
concepts that will allow you to make a chicken sandwich.
Applying this to an example-led prompt, you could provide the AI with several
examples of your email writing style and then ask it to do the following:
Understanding these types of prompts will enable you to extract more relevant
and precise responses from the AI, making it a powerful tool for your work.
Return to table of contents.
Imagine you’re a director, and the AI is your actor. You can provide a role to the AI,
and it will respond based on that role. These roles can vary significantly, ranging
from a historian discussing the impact of past pandemics to a technical expert
explaining a complex medical device.
By assigning a role to the AI, we’re able to harness its vast training data to
present the material in a way that’s specific to that role’s domain. It’s like having
an expert consultant right at your fingertips!
Some roles commonly used with ChatGPT are things like writer, editor, and
university professor.
Assigning parameters
You can also add parameters to your role prompts. Parameters provide
guidance that help the AI understand the context of what's expected
in its response, and I use them almost all the time in my prompts.
I usually write something like Act as: writer and Type of text: email.
Output language
Next, you might want to tell the AI what output language to use. In this case, I
chose the output language to be English US.
Reader comprehension
Another super useful parameter that I use all the time is reader comprehension.
Continuing with our previous example, you could specify something like
reader comprehension: 5 years old (or 10 years old or 25 years old).
As you can imagine, choosing 5 years old will result in the AI coming up with
analogies that put the information into very simple terms. You can try it out. Ask
ChatGPT to Explain special relativity to a 5-year-old and see what it does.
Moving on with the same prompt we used earlier, you can further specify
that reader comprehension parameter by adding something like the
following examples:
• Advanced, assume extensive prior knowledge
• Intermediate, assume some prior knowledge
• Beginner, assume no prior knowledge
Task
Of course, you obviously want to tell the AI what it should do—in other words,
what the task is. For example, you could write Task: Write an email introducing
Medmastery to a professor.
Output format
Another thing you can do is define the type of output you want to get, with things
like output: numbered list or output: table.
If you want the AI to work with some of your own text, then you could tell it, use
my text, and then insert your text.
That’s pretty sweet I would say. You’ll notice that it didn’t use our exact wording,
“Yo buddy, Medmastery is a pretty damn sweet app!” But it did incorporate the
spirit of it in its output in a way that would be suitable for a student’s email to
their professor.
Context
Here’s another thing you can do. You can provide ChatGPT with more context.
That’s pretty amazing, don’t you think? But we’re not done yet! Coming up next is
a highly useful prompting technique that I think you’ll use a lot.
For example, let’s say we wanted to write a blog post about how to best learn
about electrocardiography (ECG). We might ask ChatGPT to create the outline,
as you see below.
Now this isn’t perfect, but if I were working with this text, I could now use it as a
first draft and either ask ChatGPT to rewrite it based on my feedback, or I could
refine this text myself. Then I could ask it to write part two of the blog post:
You’ll notice that the last two prompts include a reminder for ChatGPT about
what it’s supposed to cover in each section. That’s sometimes necessary
because when you’re having a conversation with ChatGPT, there’s a limit on how
much of your past conversation that it can remember. So, if it’s been generating
lengthy pieces of text for you, it may forget important details that came from
earlier in your conversation unless you remind it.
Anyhow, I’d basically enter into a back-and-forth dialogue with ChatGPT, and in
my experience, this iterative approach yields the best outcomes.
Let’s say you want to know how to come up with a great website for your private
practice. Instead of asking ChatGPT, I’m a doctor; how can I create a great
website? you could ask, I’m a doctor; what are the first three steps I need to take
to create a great website for my private practice? That will provide you with a
much more actionable and useful answer. Then, if needed, you can build on that,
asking it to provide you with more details on step 1, step 2, and finally, step 3.
Unleashing the
full potential of
ChatGPT
www.medmastery.com
Supercharging medical and scientific writing
with ChatGPT (part 1)
As already mentioned, when you want to use ChatGPT to write a longer
document, you should always begin with an outline. Now, you’re not going to nail
the outline in the first attempt. You should iterate on it; enhance it with ChatGPT
until you’re pleased with the structure.
Then, the ideal approach is to let ChatGPT draft the first section of your outline,
and, you guessed it, iterate on that too. Once you are content with the first
section, you can ask ChatGPT to write the next section, building upon the first.
Topic sentences
The first point of order is topic sentences. They’re the backbone of your
paragraphs, guiding your reader through your arguments, points, and ideas.
You can start with a paragraph you’ve written, then run it through ChatGPT
with the prompt: Please write a topic sentence for this paragraph. Ask ChatGPT
to generate three variations of this topic sentence. This way, you have more
options to choose from.
Let’s put this into practice. Let’s take the paragraph below and feed it into
ChatGPT with our prompt.
Out of these, I personally would use a combination of options one and three.
Transition sentences
These are the glue that binds your paragraphs together, ensuring a smooth flow
of ideas from one point to the next.
To use ChatGPT for this task, you’ll need to identify two paragraphs you wish
to connect with a transition sentence. Now we run these through ChatGPT with
the prompt: Please write a transition sentence to connect these two paragraphs.
With this, you’re asking ChatGPT to create a bridge that will seamlessly connect
your ideas.
Now, using our prompt, let’s see what ChatGPT can generate for a transition
sentence to connect them. And there we have it:
Transition sentences are the lubricant for your essay’s engine, ensuring a
smooth ride for your reader. And with ChatGPT, creating them can be a breeze.
Introduction
Let’s delve into how you can use ChatGPT to formulate compelling introductions.
But as many experienced and novice academic writers would tell you, crafting a
captivating introduction is often a herculean task. Thankfully, with ChatGPT, the
seemingly impossible becomes a lot more manageable.
As you can see, ChatGPT restructured our draft into a more compelling and
engaging introduction. But again, it’s essential to remember that ChatGPT is an
AI assistant. Take its output as a suggestion, not the final word. It’s up to you to
mold it, to tweak it to fit your personal writing style.
In the next lesson, we’ll tackle the methods section of your paper.
Methods section
Let’s take a moment to focus on a part of the paper that’s crucial yet often
considered tedious to write: the methods section.
If you’ve ever found yourself dragging your feet when it comes to writing up your
research methodology, you’re not alone. But with ChatGPT, this task can become
a whole lot easier.
Step 1: Brainstorm
Here’s a simple trick. Start by brainstorming and jot down all these details in
bullet-point format.
Now I’ll show you an example of this action. Below is both the prompt and the
output from ChatGPT.
Remember, ChatGPT is a tool to assist you, and it’s important to put your touch on
the final output. This approach makes the process of writing a methods section
a breeze, saving you time and energy and ensuring you have a comprehensible
and complete description of your methodology.
With ChatGPT, you can input your claim or argument and use that to generate
counterarguments. Use the prompt, Please give me a counterargument for
this claim.
Let’s put it to the test. Below is a claim from our example paper:
But, remember to use your own critical thinking in using these counterarguments.
After all, ChatGPT is your research assistant, not your supervisor.
Writing a conclusion that effectively summarizes your work and leaves a lasting
impact on the reader can indeed be challenging. But with ChatGPT, this task
becomes significantly more manageable.
Similarly to how we crafted our introduction, you should start by writing a draft of
your conclusion. Once you have this, run it through ChatGPT with the following
prompt: Please rewrite this paragraph as a conclusion. You should ask for three
variations to have a wider range of choices.
Remember that you can ask ChatGPT for more than one version of your
conclusion and select the one that best fits your work and personal style.
Simply give ChatGPT the conclusion you liked, along with your own work’s
main points or arguments. Then ask it to write a similarly structured conclusion
for you. Remember to specify that you’re looking for a conclusion structured
similarly to the example you provided.
This is an incredibly powerful technique to learn from work you admire and apply
that same brilliance to your own writing, while still retaining your unique voice
and perspective.
As always, tailor the final output to your style and the specific requirements of
your paper.
How can you make it more engaging, or even humorous? Well, with the power of
ChatGPT, it’s as simple as using the right prompt.
The key is to specify the style you want the text to be in. You can use the
following prompt: Change the writing style of the text below to [desired style]. Then
paste your text after the prompt.
For instance, you can use the prompt: Rewrite this text as if I were a beginner.
Below is an example. By feeding the text into ChatGPT with this prompt, you can
transform a complex explanation into a simpler one, making it easier for
beginners to understand.
Begin with the prompt: Edit the following text. It’s for my [type of text]. Then paste
your text after that prompt. Now watch ChatGPT go to work. Its true power lies
in customization.
You can detail the degree of changes needed, guide the AI on the type of edits,
and even get feedback to improve your writing skills. ChatGPT can also tailor
your text’s style and tone to fit your needs, even mimicking the voice of a famous
person or translating your work.
For the scriptwriters and novelists out there, ChatGPT can find weak points in
your plot. Just ask it the following: Show me all the weak plot points in this. It’s like
having a reliable second set of eyes for your work.
With ChatGPT, the possibilities for editing are truly limitless, making it a potent
tool for writers.
Suppose you’re working in a clinic and want to streamline the patient workflow.
It might feel like a daunting task, but ChatGPT can ease this process.
Regarding language, we’ll stick with English US. For the reader comprehension
level, let’s assume this is aimed at a 25-year-old with limited experience in hospital
administration and logistics. Next, we determine the format of the output. In this
case, we want it to be a bullet-point list.
Finally, specify the topic: Patient flow in the hospital, patient logistics. The task
would then be: Based on these, can you provide five ideas to make the patient flow
in my clinic more effective?
With these elements in place, ChatGPT can now generate a list of five
professional, friendly, and comprehensive ideas to streamline the patient flow in
your clinic.
Data structuring
Imagine this scenario. You receive an email, extensively detailing a completed
construction project.
That’s exactly what happened in the example below regarding a recently finished
building project. It summarizes the materials used and their respective quantities.
To structure this data, you can feed the content of this email into ChatGPT with a
specific prompt asking it to structure the information into a table. For instance,
you might ask the following:
Then ChatGPT will output a structured table containing all the material types,
their uses, and their corresponding quantities, as described in the email.
Isn’t that fantastic? All the essential information from the lengthy email is now
compiled into an easy-to-read and digestible format.
In our emoji-filled world, pinpointing the one that encapsulates your emotions or
message can be challenging. That’s where ChatGPT comes to the rescue!
Imagine this. You’re writing an email for an invite to a surprise birthday bash this
Saturday, and you want an emoji to add a dash of fun to the subject line. In
ChatGPT, you can type in your subject line followed by the prompt, Suggest 10
emojis for this, and ChatGPT will spring into action.
This way, ChatGPT acts as your emoji concierge, adding a playful and personal
touch to your text. Obviously, you can always refine it or drill down into a specific
idea as much as you want to.
Why do they always have to be written in such a complex way, full of confusing
jargon and seemingly bloated phrases? Well, the good news is that ChatGPT is
about to become your new best friend when it comes to deciphering contracts.
This is where ChatGPT comes in. Let’s say you’re not sure what clause
three means: “Lessor shall maintain the common areas in a clean and
sanitary condition.”
Use the prompt, What does this mean? and then paste in the text that you
found difficult.
But it doesn’t stop there. With ChatGPT, you can dig deeper. By asking follow-up
questions, you can improve your understanding even further. For example, you
might ask, What can the tenant do if the lessor fails to maintain the common areas?
Below is the answer.
Let’s go back to the lease scenario. While many parts of the lease may
seem standard, there might be clauses inserted that put the tenant at
a disadvantage.
As you may recall, clause five of the example lease we saw earlier said, “Lessee
shall be responsible for all repairs, major or minor, to the leased premises.” It
sounds like a hefty responsibility for a tenant, doesn’t it?
First, in some jurisdictions, it may be required by law that the landlord takes
responsibility for certain repairs, especially those that are essential for the
property to be habitable. Second, If the property was in poor condition at the
beginning of the lease, it may be considered unfair to hold the tenant responsible
for all repairs, major and minor.
Of course, you can go through the same process with any other complex
document. You could take any section of the document and ask ChatGPT to
summarize it as if you were 16 years old, for example. That’ll make it much
easier to understand, without oversimplifying it.
Explain it to a child
The first technique allows you to ask the AI to explain complex medical or
scientific concepts in a simplified manner, which is similar to how you would
explain a complex topic to a child or a preteen. You can choose any age.
We call this technique, Explain it to a child (e.g., 5, 10, 12, 15, 20, or any age). Each
explanation will be slightly different, and I encourage you to play around with it.
Here’s an example of how you can do this. If you’re finding a concept like
epigenetics overwhelming, you could type: Explain epigenetics like I’m 12.
ChatGPT will then generate a response that simplifies the concept.
Step-by-step explanations
The second technique is step-by-step explanations. This technique is helpful
when you want to understand a complex process or concept in more detail.
For example, to understand cellular respiration, type: Can you explain the steps of
cellular respiration?
Dialogue-based learning
The third technique involves you and ChatGPT engaging in a back-and-forth
conversation and is known as dialogue-based learning.
This method allows you to tailor your learning experience and dig deeper into
topics of interest.
Flashcard generation
Finally, the fourth technique is flashcard generation. This involves using ChatGPT
to create flashcards.
In this example, we’re going to use the 2014 American Heart Association /
American College of Cardiology guideline for the management of patients with
non-ST-elevation acute coronary syndromes.
However, it’s important to note that these guidelines are 83 pages long, and
feeding all of them into ChatGPT at once is not an option. That’s because
ChatGPT has a limit on how much context it can take into account in a given
conversation.
Or, to put it another way, it’s kind of like ChatGPT has a limit on how much it can
remember at one time. In fact, if you submit too much text to ChatGPT, you’ll get
For this example, we’ll use the prompt, Please create five United States Medical
Licensing Examination (USMLE)-type questions based on the following guidelines on
the diagnosis of acute coronary syndromes.
Then we’ll paste about 2000 words of text from section three of the guidelines
into the prompt window. (Keep in mind that ChatGPT is updated regularly, so the
limit on your input length may change over time.)
ChatGPT will come up with questions and answers like the ones below.
You may need to copy the questions it previously created into your prompt,
too. That type of reminder is helpful to keep ChatGPT on track, particularly with
longer conversations.
Here’s a prompt you can use: Now please output this information in a CSV format
that can be easily imported into Anki. On the front of the card, write the question
together with the possible answers, on the back of the card write the correct answer.
Make sure that on the front side of the cards, each answer appears on a new line.
Make sure the fronts and backs are separated by a semicolon. Also provide clear,
step-by-step instructions on how to import the CSV output into Anki.
Import the CSV file you created by selecting the file and clicking Open.
Below is what the front of a card looks like. If you’d like to adjust any of the
formatting, you can do that by clicking the Edit button.
When you click Show Answer, the answer is now at the bottom.
As you can imagine, this technique can help streamline your studying, making
it more efficient and effective. I once helped my son create an Anki deck for a
geography test. He had to study 10–15 pages in his book.
We used the camera of my iPhone, which comes with OCR capabilities meaning
that you can basically hold the camera over a piece of text and copy-paste
that text. I then went through the same process as the one I just showed you,
creating flashcards using ChatGPT, which made it much easier for him to study
for the test.
If you want to learn more about how you can use Anki at its full capabilities to
supercharge your learning, be sure to check out our Anki Essentials Course!
www.medmastery.com
Reference list
https://Bing.com/chat
https://openai.com/blog/chatgpt
https://www.nejm.org/
Amsterdam, EA, Wenger, NK, Brindis, RG, et al. 2014. 2014 AHA/ACC Guideline for the
Management of Patients with Non-ST-Elevation Acute Coronary Syndromes: A report
of the American College of Cardiology/American Heart Association Task Force on
Practice Guidelines. J Am Coll Cardiol. 64: e139–e228. PMID: 25260718