Subcategories

• Tutorials

  Step-by-step guide for building mini-projects

  119
  Topics
  157
  Posts

  

  Introduction

   

  Welcome to this tutorial on building an AI-powered calendar assistant! This project is a great example of how AI tools can be extended with new abilities — allowing them to solve more complex, real-world problems.

  🧠 What’s This All About?

   
  Large language models (LLMs), cannot remember information across different sessions — they can only “remember” the messages within the current chat session. That’s a big limitation because most apps need to store and recall information over time. Think about a calendar app: it needs to store all your events so you can see them later!

  To work around this limitation, we’ll use Google Sheets to store data. Then we’ll teach the AI how to interact with that data to manage a user’s calendar.

  By the end of this project, your app will look like this:

  calendar.gif

   

  This tutorial is divided into two main parts:

  Part 1: Set up a basic AI chat app Part 2: Teach the AI how to manage the user’s calendar using Google Sheets

   
   
   
   
   

  🧩 Part 1 - Setting up the chat app

   

   
   

  Step 1 - Remix the Starter Project

   

  Start by remixing this template, which gives you a basic chat app to build on. We’ll add new logic to it step by step.

  play.creaticode.com/projects/682dad154dba4796824af2a6

   
   

  Step 2 - Prepare the Google Sheet

   

  We’ll use a Google Sheet to store all calendar events. This approach is free, easy to use, and simple to review.

  Here’s what you need to do (or ask your teacher/guardian to help):

  Create a new Google Sheet Set its sharing settings so that “anyone with the link” can edit Copy the sheet’s link

  6fe532c8-c01a-4dbd-ba03-b15462392d8a-image.png

  If you can’t create your own Google Sheet, you can use this pre-made one:
  https://docs.google.com/spreadsheets/d/1tH9J09laZNl94Qb1smO5LSmDwD_yzungb9d4lmkobxE/edit?usp=sharing

   
  Then, in your app, store the sheet URL in a variable called Sheet URL. You’ll use this later to tell the AI where the calendar lives.

  eae1657b-7a9d-4cf2-a745-4b3637f32d2e-image.png

   
   

  Step 3 - Update the System Prompt

   

  Next, we’ll give the AI a new role by updating its System Instruction.

  This tells the AI what it’s supposed to do. For now, keep it simple and clear:

  197c4ae4-ba5d-4133-9953-3b190e6ec10a-image.png

   
   

  Step 4 - Customize the Welcome Message

   

  Let’s make the welcome message match the AI’s new job.

  667f26d2-e8a1-4297-bed5-0db881b29b55-image.png

   
  📌 Tip: You could make the AI generate a different welcome message every time, but that would slow things down.

   
   

  Step 5 - Add a Tab for Each User

   

  Because different users might use this app, we need a way to keep their data separate.

  The solution: whenever a new user joins, we’ll create a new tab in the Google Sheet just for them. The tab name will be the user’s unique ID.

  ae212509-2bdf-48aa-b854-bd4d886f5597-image.png

   
  📝 If the tab already exists, this block will do nothing — which is exactly what we want.

   
   

  Step 6 - Get the Current Day of the Week

   

  Here’s an important thing to remember: AI models do not know the current date or time on their own. You can test that like this:

  35c1e1ac-c623-4b17-90ee-932e5b1eabfa-image.png

   
  That’s a problem when managing calendars! We’ll fix this by manually sending the current date and time to the AI.

  Let’s start with the day of the week:

  First, create a new list “weekdays”:
  6298781c-e323-43ba-afd8-21448f235fbe-image.png

  Then, use the current [day of week] block to get today’s number, and use it to pick the correct item from your list:
  82c8a6e9-d0d6-402e-9fa9-925f23ce4cbe-image.png

   
  📅 For example, if today is Wednesday, the “current day of the week” block will return 4, and item 4 in your list will be "Wednesday".

   
   

  Step 7 - Get the Full Date and Time

   

  We can also get the full date and time using the current block and combine them into a single string:

  6dfcfd51-db40-401c-a5d2-ad253e5d4d66-image.png

   
   

  Step 8 - Send Time Info with Every User Message

   

  Now we’ll attach the time information to each message we send to the AI model.

  bd8c0581-61a3-4da0-bc40-1d5eae91b171-image.png

   
  For example, if the user says “hi”, then what we actually send to the AI will be:

  hi (current time: Wednesday 05/21/2025 13:23:43 )

  That way, the AI will know exactly when the message was sent!

  You can now test it to make sure the AI sees the correct time:

  392697d0-3618-40dd-a401-0d1d5c871753-image.png

   
   
   
   
   
   
   

  🧩 Part 2 - Managing the Calendar with Google Sheets

   

   
   

  Step 1 -  Plan the Workflow for Adding a New Event

   

  Our entire calendar lives in Google Sheets, yet the AI can’t touch that Sheet directly — it needs help from our helper code. Here’s a sample back‑and‑forth among the user, the AI, and the helper:

  User -> AI: I will have a haircut tomorrow at 4pm AI -> Helper: add a new event for [date] [time] Helper adds it to the Google Sheet Helper -> AI: event added successfully AI -> User: I have added the haircut to your calendar.

   
  From the user’s perspective, the Sheet and helper code are invisible—the AI simply “gets it done.”

  Another way to think about it: the user speaks natural language, the helper speaks code, so the AI acts as the translator between them.

   
   

  Step 2 - Define the “addEvent” Command

   

  To teach the AI to use our helper, we need to define a new command “addEvent” in the system instruction:

  You are an AI assistant that helps manage the user's calendar. ## Commands You can use the following tools to manage events through an external service: * To add a new event to the calendar, you can return this command, which will return its eventID: addEvent | [date] | [time] | [event description] ## Instructions * If you are using a command, then only return the one line of command and nothing else.

  Notes:

  We separated Commands and Instructions with ## headers so the AI can “see” the sections clearly.

  The addEvent syntax uses pipes (|) — easy for us to split into pieces later.

  The instruction forces the AI to output only the command line when needed — no extra chit‑chat.

  Testing now shows the AI emitting addEvent lines automatically:

  324814be-e4dd-4867-9c6b-9b5b5925ad71-image.png

   
   

  Step 3 - Standardize Date and Time Formats

   

  The AI currently formats dates like 05/01/2025, which sorts poorly—05/01/2025 appears “larger” than 02/01/2026. Times such as “3 p.m.” sort badly, too.

  Add two lines to the Instructions section to enforce sortable formats:

  You are an AI assistant that helps manage the user's calendar. ## Commands You can use the following tools to manage events through an external service: * To add a new event to the calendar, you can return this command, which will return its eventID: addEvent | [date] | [time] | [event description] ## Instructions * If you are using a command, then only return the one line of command and nothing else. * Use yyyy.mm.dd date format, such as "2025.02.21" * Use hh:mm time format, such as "13:30"

  Re‑run the app and confirm:

  d1eb5487-12ab-4d69-a973-1cfc3adb3217-image.png

   
   

  Step 4 - Intercept the “addEvent” command

   

  Whenever we receive AI output, we’ll check if it begins with "addEvent " (note the trailing space). If so, we execute helper code in “handle add event”.

  ee6922cd-072a-4a8b-a75b-38d00263f53b-image.png

   
   

  Step 5 - Extract the Parameters

   

  The addEvent line has four parts separated by three pipes. Use the “part of” block to pick them apart and trim whitespace:

  b904221b-f61a-44b0-a897-5ae7264f3e73-image.png

   
   

  Step 6 - Read Data from Google Sheets

   

  Our add‑event logic is:

  Read the user’s tab into a table called events. Add the new event. Write the entire table back.

  To start, we will read the sheet using a “read data” block:

  c9234755-1b23-4d7b-ab29-747080b966df-image.png

   
  In this block, we will use the “read from Google Sheet” block to read the first 10000 rows from this user’s tab, assuming there are less than 10000 events on the calendar:

  333550a9-d29b-4284-8fcc-67a38af3aa7f-image.png

   
  Run this block and remove all the widgets (or save and reload the project), you will find the “events” table looks like this:

  5ed62169-3cac-423a-a4dd-977a5e16ce8a-image.png

   
  That’s because the new tab is still empty, so the column headers are missing. To fix that, we can define the columns whenever there are 0 rows in the table:

  61dffcd1-e3d2-4361-8535-533288186d35-image.png

   
  The first column “id” will be used to store the unique eventID.

   
   

  Step 7 - Add the New Event

   

  Generate a unique ID using row count + 1 (always unique) and prefix it with “E”. Then append a row:

  d3c9015d-d19e-450b-9775-22dfd1a66eca-image.png

   
   

  Step 8 - Sort the Table

   

  Since all the events have date and time, we should keep them sorted in the calendar. We will sort the rows by “time” first, then by “date”. This will ensure all events are sorted by date and then time:

  4d864702-2d7b-4f65-8a59-bab1e3025a9e-image.png

   
   

  Step 9 - Write Back to the Sheet

   

  Overwrite the user’s tab with the updated table:

  05a6e9d1-f1d6-46fe-a66a-7ca12687e8e2-image.png

   
   

  Step 10 - Report Success to the AI

   

  Let the AI know the command succeeded and pass along the new eventID:

  64a7f312-336b-4591-a3ac-840599769b41-image.png

   
  We clarify that this message is from addEvent, so the AI won’t mistake it for user input.

  Test by adding events:

  c49731ea-472d-498f-a385-ae98eb299472-image.png

   
  …and confirm they appear in the Sheet:

  48824007-a8a5-4e7a-9d2a-1ee82f9eb9d8-image.png

   
   

  Step 11 - Define the “removeEvent” Command

   

  Next, let’s work on removing an evnet. We will still start with defining the command in the system instruction:

  You are an AI assistant that helps manage the user's calendar. ## Commands You can use the following tools to manage events through an external service: * To add a new event to the calendar, you can return this command, which will return its eventID: addEvent | [date] | [time] | [event description] * To remove an existing event, after you have found its eventID, you can use this command: removeEvent | [eventID] ## Instructions * If you are using a command, then only return the one line of command and nothing else. * Use yyyy.mm.dd date format, such as "2025.02.21" * Use hh:mm time format, such as "13:30"

  removeEvent takes just one parameter — the event’s unique ID. If the AI doesn’t already know the ID, it must retrieve it using other commands first (those are defined later).

   
   

  Step 12 - Intercept the “removeEvent” Command

   

  Detect and handle it with a new block:

  fba93dd8-a0af-4516-9c8d-5316582db7f5-image.png

   
   

  Step 13 - Extract the eventID

   

  Pull the ID from the command string:

  0180e05f-19f3-44cf-b351-6589c557a27f-image.png

   
   

  Step 14 - Find and Remove the Event

   

  Read the Sheet into events again. Find the index of the row with the matching eventID. Delete that row.

  8637a6a4-4f4d-4cba-b236-d265a9214ae6-image.png

   
   

  Step 15 - Clear and Rewrite the Sheet

   

  Before we write the table back, we need to first clear the tab. The reason is that the table has one less row compared to the data in that tab, so if we simply write the table into that tab, there will be an extra duplicate row at the bottom.

  bdd2d179-e0a3-428f-b7cf-ccece0807ecd-image.png

   
   

  Step 16 - Report Removal to the AI

   

  Notify the AI so it can reply to the user:

  7d7a062a-078f-4733-9579-baf1cdda4d44-image.png

  Restart your program, add an event, then remove it. Double‑check the Sheet each time:

  f31330ff-bb58-4ed5-bd82-bb6c399c64cb-image.png

   
   

  Step 17 - Define the “searchEvents” Command

   

  Earlier, the AI already knew the eventID for the dentist appointment, so removing it was easy. But when an eventID is not in the chat context, the AI must search for the event first.

  To keep things simple, we’ll let the AI specify one keyword. Our helper will search that word in the description column and return every matching event (with its eventID).

  Updated system instruction:

  You are an AI assistant that helps manage the user's calendar. ## Commands You can use the following tools to manage events through an external service: * To add a new event to the calendar, you can return this command, which will return its eventID: addEvent | [date] | [time] | [event description] * To remove an existing event, after you have found its eventID, you can use this command: removeEvent | [eventID] * To search for existing events by a **single word**, use this command, which will return all events at that time with their eventIDs (use a generic word to get more likely matches): searchEvents | [keyword] ## Instructions * If you are using a command, then only return the one line of command and nothing else. * Use yyyy.mm.dd date format, such as "2025.02.21" * Use hh:mm time format, such as "13:30"

  We instruct the AI to choose one broad word (e.g., “lunch” or “Jon”) so we’re more likely to find the right event—even if we return a few extras, the AI can sort them out.

  Test it:

  cb3181c5-3870-40b9-90d3-8ded030b3ec9-image.png

   
  The AI should pick “lunch” or “Jon” as the word to search for.

   
   

  Step 18 - Intercept the “searchEvents” Command

   

  We will handle this new command using the block “handl search command”:

  a189acb7-0a5b-4cd4-8aa6-cfc1a1109172-image.png

   
   

  Step 19 - Extract the Search Word and Read the Data

   

  Store the keyword in a variable called search word:

  d470f68e-fb62-488c-ae99-73dbb6dfd306-image.png

   
   

  Step 20 - Iterate Through All Rows

   

  Use a for‑loop to walk through each row in events, pulling out the description:

  b1fbb7ed-c0e0-4ffc-af72-d85b4a194889-image.png

   
   

  Step 21 - Build the Search Result

   

  Create a variable called search result. Start it empty, and append every row whose description contains the keyword:

  dd7b8d9b-2dd3-46c0-bc4a-f9d10745b8e6-image.png

   
   

  Step 22 - Send the Search Result to the AI

   

  After the for-loop finishes, the “search result” will either be empty or filled with events. Either way, we will let the AI know the result, so it can generate the response to the user:

  29c74dc2-4960-40b3-883c-56ef07fd5d6d-image.png

   
  To test it, first add another lunch event to the Google Sheet with another person:

  7b4d3776-859e-4f85-9e28-083e63d91b66-image.png

  Then, restart the app, and ask about the “lunch with Jon”:

  9a7d452e-0fa5-4c22-b2e6-224b66743812-image.png

   
   

  Step 23 - Define the “findEvents” Command

   

  Our final command, findEvents, grabs every event between two time points. Perfect for “Am I free next Wednesday morning?” or “What’s on my calendar this weekend?”

  Here is the updated system instruction with the “findEvents” command:

  You are an AI assistant that helps manage the user's calendar. ## Commands You can use the following tools to manage events through an external service: * To add a new event to the calendar, you can return this command, which will return its eventID: addEvent | [date] | [time] | [event description] * To remove an existing event, after you have found its eventID, you can use this command: removeEvent | [eventID] * To search for existing events by a **single word**, use this command, which will return all events at that time with their eventIDs (use a generic word to get more likely matches): searchEvents | [keyword] * To find existing events inbetween 2 time points (they can be the same date or time), you can use this command, which will return all events within the given time interval: findEvents | [start date] | [start time] | [end date] | [end time] ## Instructions * If you are using a command, then only return the one line of command and nothing else. * Use yyyy.mm.dd date format, such as "2025.02.21" * Use hh:mm time format, such as "13:30"

   
   

  Step 24 - Intercept the “findEvents” Command

   

  We will handle this command using a new block called “handle find events”:

  779b348f-b6c7-4853-9f3c-83d3cf946c74-image.png

   
  The “handle find events” block should be very similar to the “handle search events” block. Can you try to implement it before looking at the steps below?

   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   

  Step 25 - Extract Parameters

   

  We will first extract the 4 parameters:

  e2561586-799a-4b27-954b-592d8e8b08d8-image.png

   
   

  Step 26 - Reading Events in a For-Loop

   

  We will still use a for-loop, but this time, we will extract the eventDate and eventTime from each row:

  974f08b8-7ec7-40ea-b506-955b752045b8-image.png

   
   

  Step 27 - Skip Events Before the Start Date/Time

   

  As we check out each row, if we find its date is earlier than the start date, or its time is earlier than the start time, then we will continue to the next row:

  fbb2c495-08cf-43ce-b613-c6b890d109f1-image.png

   
   

  Step 28 - Stop After the End Date/Time

   

  Because events are already sorted, once we reach a row after the end date/time, we can break the loop. Until then, add each valid row to search result:

  413abb89-a679-4f75-9fa0-dbde4e7d29d4-image.png

   
   

  Step 29 - Send these events to the AI

   

  After the for-loop completes, we will give the result list to the AI:

  db5cbeff-0384-4655-991e-88892902fd8c-image.png

   
  Now let’s test it. Suppose this is the calender:

  b6fb265d-d965-4b4e-be6a-b6b90d7ac4d9-image.png

   
  The AI will be able to find all events for next week:

  32bd8a87-72ba-4769-9ba0-48126f9b0a10-image.png

   
   

  Step 30 - Refine the output

   

  The reply above feels a little robotic. Let’s tweak the system prompt so the AI talks more like a human:

  ## Instructions * If you are responding to the user, then talk like a human * If you are sending a command, then only return the **single line** of command and nothing else. * Use yyyy.mm.dd date format in commands, such as "2025.02.21" * Use hh:mm time format in commands, such as "13:30"

   
  Unfortunately, this sometimes makes the AI mix user responses with commands:

  1d59cee0-244b-44b9-9f4c-9283717ac8a7-image.png

   
  Fix: Apply the T.I.R.E. prompting method by adding an explicit Example to the end of the system instruction:

  ## Example [User]: I need to go to Jon's birthday party this Saturday [You]: What time is the party? [User]: I think it is 2pm [You]: addEvent | 2025.05.31 | 14:00 | Jon's birthday party at 2 PM [Helper]: addEvent result: successfully added event with ID: E123 [You]: I have added Jon's birthday party to the calendar for May 30 at 2pm. [User]: So what's my schedule next week? [You]: findEvents | 2025.05.26 | 00:00 | 2025.06.01 | 23:59 [Helper]: findEvents result: E99|2025.05.28|18:00|gym class with friend at 6 PM E123|2025.05.31|14:00|Jon's birthday party at 2 PM [You]: Looks like you have a light schedule next week. You will be meeting with your friend for a gym class next Wednesday at 6, and you will also attend Jon's birthday party on Sunday at 2 o'clock.

   

  Another problem that might appear in testing is that the AI sometimes think of “next week” as the “next 7 days”, rather than the next calendar week. This can be fixed by adding a new line in the “## Instructions” section:

  * When the user refers to the next week, assume they mean the next calendar week, not the next 7 days.

   

  Finally, we are getting a much more human-like response:
   
  be331e0c-680c-49ad-b023-822c792fa45e-image.png

   
   
  Here is a final demo:
   
  calendar.gif

   
   

  Important Note:

   
  When you release your app for other users, you should publish the project instead of sharing it. That is because if you share the project, any user can look inside it to acquire the URL of the Google Sheet, then they will be able to look at other users’ calendars.

   
   
   
   
   

  Additional Challenges

   

  For practice, here are some improvements you can make to this app:

  Multiple Commands: Users may combine requests (“My lunch with Jon is cancelled and schedule a dentist at 4 p.m. today”). Rescheduling should trigger a removeEvent followed by an addEvent.

   

  Multiple Search Keywords: Allow two or three keywords for searchEvents. “Lunch with Jon” could search for “Jon” and “lunch.”

   

  Remove Past Events: When the app starts, delete events more than one month old to keep the Sheet tidy.

   

  Recurring Events (advanced):

  Extend addEvent with a fifth parameter, interval (e.g., “weekly,” “daily”). Add an interval column to the Sheet. When interval is present, auto‑add the event for the next six months. On startup, scan for recurring events and extend them six months into the future.
• 2D Blocks

  How to use each 2D block

  133
  Topics
  138
  Posts

  

  Introduction

  You can now make everything on the screen (the stage and all sprites) shake like this:

  shake.gif

   
  Note that this only works for 2D projects.

   
   

  The Shake Screen Block

  The shake screen block is in the motion category:

  5b666ce1-eecc-4b7a-aba7-12b1786bed56-image.png

   
  It takes 3 inputs:

  Duration: how many seconds to keep the shaking effect Frame Rate: how many frames per second. In each frame, the whole stage and all sprites are randomly shifted to a new position. Magnitude: This is the maximum distance the screen may be shifted in both the x and y direction. For example, if magnitude is 20, then for any frame, the screen can be shifted left or right, up or down, but never more than 20 units from the center.
• 3D Blocks

  How to use each 3D block

  90
  Topics
  90
  Posts

  

  Introduciton

   

  You can now add joystick-style user controls for your 3D games, which is a natural input method for users on touchpads and smartphones.

  Note that this new feature only works in 3D projects, and a 3D scene must be initialized first.

  joy.gif

   
   

  Add a Joystick

   

  A virtual joystick is made of 2 parts:

  A “base” that is composed of 2 circles. Whenever the pointer is pressed down, the base will not move as the pointer is dragged around it. A “puck” that is a smaller circle with a dot at its center. It shows the current position of the pointer relative to the base.

   
  a8027378-beb1-4d73-9a39-d4d128f4f0c5-image.png

  You can add a new joystick using the following block:

  4bb6149b-de59-4249-9f70-59f503e8e92e-image.png

   
  The inputs are:

  Left or Right: you can add a left or right joystick using this block. The left one only covers the left half of the screen, and the right one only covers the right half. Use this block 2 times to add both joysticks. Color1 and Color2: These two color inputs control the colors of the base and the puck. Scale: This is a scaling factor that controls the size of the virtual joystick. By default, it is 100, which means 100% of the default size. If you would like to double the size of the joystick, you can set this value to 200.

   
   

  Read joystick properties

   

  You can read the properties of a joystick using this reporter block:

  232b6132-f657-4750-8c3c-ff51abe69ddc-image.png

   
  It has 2 inputs:

  Side: left or right Property: x or y: the position of the puck relative to the center of the base. Note that both x and y are between -1 and 1. For example, if the puck is at the right edge of the base, then x will be 1 and y will be 0. dir: the direction of the puck relative to the base in degrees, which is between -180 and 180. Distance: the distance of the puck from the base between 0 and 1. pressed: whether the joystick is pressed down. 0 means not pressed down, and 1 means it is pressed down. This should be the first value to check, since if the joystick is not pressed down, then all the other properties are not valid.

   
   

  Remove all joysticks

   

  You can use this block to remove all the joysticks you have added so far:

  a2e02fc4-f3bc-474b-9f7f-35f17a72c3e8-image.png

   
   

  Demo

   

  In this demo program, we first add a 3D glass object, then add 2 virtual joysticks of different colors and scales. Then we enter a forever loop: whenever the left or right joystick is pressed down, we read its properties into a few variables, and also rotate the 3D glass based on the “dir” property:

  blocks (42).png

   
  Here is the project link:

  play.creaticode.com/projects/673e98585e434ba9fd49f96a

• Live Books

  Free K-12 books that you can use in combination with the CreatiCode platform to learn coding and technologies. Their content will be continuously updated based on your feedback.

  ChatGPT for K-12 Students - Understand, Build, Innovate
  8
  Topics
  55
  Posts

  

   
   

  What’s a Prompt?

   

  DALL·E 2023-10-20 20.10.21 - Cartoon style image showcasing a cute chatbot labeled 'ChatGPT'. The chatbot has large friendly eyes and a digital screen on its chest displaying a bl.png

   

  You’re already familiar with ChatGPT, but let’s go deeper into the mechanics of your interactions: the prompts. Every message you send to ChatGPT is actually a prompt. Instead of thinking of a prompt as a standalone question or command, imagine it as your contribution to an ongoing dialogue with ChatGPT. This isn’t just a Q&A session; it’s more like a flowing conversation where each prompt serves a specific role.

  Here are the core types of prompts you’ll use:

  Questions: These are straightforward inquiries that seek a direct answer. For instance, “ChatGPT, how does photosynthesis work?” or “What is the square root of 64?”

  Requests: This type is for when you need ChatGPT to perform a specific task. For example, “ChatGPT, summarize this chapter for me,” or “Create a list of renewable energy sources.”

  Information: These prompts add extra context or clarification to the conversation. Maybe ChatGPT misunderstood a previous prompt, or perhaps you want to add more detail to get a better answer. For example, “I meant to summarize the chapter about World War I, not World War II,” or “By renewable energy, I’m referring to sources that are not fossil fuels.”

  In a real-world chat with ChatGPT, you’ll often use a mix of these types of prompts. You could start with a question, then make a request based on ChatGPT’s answer, and provide additional information to clarify or steer the conversation.

  Understanding these different types of prompts allows you to have a more dynamic and meaningful conversation with ChatGPT. With this knowledge, you’re better equipped to make ChatGPT a valuable tool for learning, creating, and exploring.

   
   

  Building a Great Prompt: The CORE Framework

   

  If you are just having a casual chat with ChatGPT, you can pretty much speak your mind freely in your prompts. However, when your goal is to ask ChatGPT to carry out a task and deliver satisfactory results, the prompt often needs to be more complex and structured.

  Although there are many ways to write a prompt, some are more effective than others in getting a high-quality response. You may often hear others giving very different opinions on what ChatGPT can do: some people may tell you ChatGPT is awesome and can do great things, while other people may say ChatGPT is very confused and can’t get anything useful out of it. Most of the time, this difference is caused by the difference in the prompts they have used. ChatGPT is like a gold mine, and a well-written prompt is like a well-calibrated metal detector that allows you to find gold in it.

  To help you get started, let me introduce a simple but powerful framework called CORE. It stands for Context, Objective, Requirement, and Example. Using these four elements, you will be able to create prompts that bring out the best in ChatGPT, giving you answers that are as awesome as you hoped they would be.

   
   

  Context: Setting the Stage

   
  Context is like the setting in a story. It tells ChatGPT the background details: who you are, what you are trying to do, and why you’re asking ChatGPT for help. Good context is like giving ChatGPT a backstage pass to your situation. Without it, ChatGPT might give you an answer that’s sort of like a one-size-fits-all t-shirt — kind of okay but not a great fit. For example, if you ask ChatGPT “How can I earn some coins?”, its answer will be very different depending on whether you are talking about “Super Mario” or real life.

  Usually, an easy way to describe the context is to start with “I”, such as “I’m a 6th-grade student, and I need to …”, “I am trying to …”, or “I need your help with …”. Alternatively, some people prefer to start with “You” (i.e. ChatGPT), such as “You are helping me with …”, or “You are talking to a 6-th grade student”.

  Sometimes, you may need to provide additional information or reference to ChatGPT in the context. For example, if you need ChatGPT to help summarize an article, then you need to include that article in your prompt. Your prompt may look like this:

  I am reading an article about the French Revolution: [insert the entire article] It is a bit too long. can you summarize it for me?

   

  Examples of Using Context

   

  Math Homework Help: Less Effective Prompt: “Help me with fractions.” Effective Prompt: “I’m a middle school student struggling with my math homework on fractions. Can you help me?” Explanation: The effective prompt specifies that it’s a middle school student needing help with homework. This helps ChatGPT tailor the explanation to a middle school level, making it more accessible for the student.
    Science Fair Project on Renewable Energy: Less Effective Prompt: “How do solar panels work?” Effective Prompt: “I’m working on a science fair project about renewable energy, specifically focusing on solar panels. How do they work?” Explanation: This prompt sets the stage for a more educational and project-focused response, informing ChatGPT that the user isn’t just curious but has a specific academic goal.
    Writing a Poem for English Class: Less Effective Prompt: “How do I write a poem?” Effective Prompt: “I have an English class assignment to write a poem about nature. How do I get started?” Explanation: The effective prompt makes it clear that the task is for an English class assignment about nature, allowing ChatGPT to offer more targeted tips and suggestions.
    Learning to Play a New Instrument: Less Effective Prompt: “Tips for learning the guitar.” Effective Prompt: “I just joined my school’s music club and chose to learn the guitar. Any tips?” Explanation: The context specifies that the user is a beginner who has joined a music club. This helps ChatGPT provide tips that are suitable for someone who is new to playing the guitar and perhaps working in a group setting.

  Each of these examples shows how a little bit of context goes a long way in getting a response that really fits what you need.

   

  Objective: What’s the Mission?

   
  Once you’ve set the stage with your context, it’s time to get down to business with the Objective. This is where you tell ChatGPT exactly what you want it to do for you. You’re essentially giving it a mission to complete. Without a clear objective, ChatGPT would have to guess what it needs to do, and sometimes may take the wrong action.

  For example, imagine you tell ChatGPT “I failed my math test today”. If you don’t say anything else, most likely ChatGPT will offer some advice on how to improve your math skills. However, if you say “I failed my math test today. How should I tell my parents about it?” you will get very different answers.

  Usually, the objective should contain a verb that describes the action that ChatGPT needs to take, such as “explain”, “describe”, “list”, “design”, “summarize”, etc.

   

  Examples of Great Objective Descriptions

   

  Math Homework Help: Less Effective Prompt: “I’m a middle school student struggling with my math homework on fractions. Help me.” Effective Prompt: “I’m a middle school student struggling with my math homework on fractions. Can you walk me through how to solve ( \frac{1}{2} + \frac{3}{4} )?” Explanation: The highlighted objective clearly asks for a walkthrough on solving a specific fraction problem. This directs ChatGPT to provide a step-by-step solution tailored for a middle school student.
    Science Fair Project on Renewable Energy: Less Effective Prompt: “I’m working on a science fair project about renewable energy, specifically focusing on solar panels. How do they work?” Effective Prompt: “I’m working on a science fair project about renewable energy, specifically focusing on solar panels. Please explain how they convert sunlight into electricity.” Explanation: The highlighted objective narrows down the query to ask how solar panels convert sunlight into electricity. This specificity guides ChatGPT to deliver a more focused explanation, ideal for a science fair project.
    Writing a Poem for English Class: Less Effective Prompt: “I have an English class assignment to write a poem about nature. How do I get started?” Effective Prompt: “I have an English class assignment to write a poem about nature. Please give me a template or outline to start my poem?” Explanation: By asking for a template or outline, the highlighted objective specifies the type of assistance needed. This helps ChatGPT provide a useful starting point for writing the poem.
    Learning to Play a New Instrument: Less Effective Prompt: “I just joined my school’s music club and chose to learn the guitar. Any tips?” Effective Prompt: “I just joined my school’s music club and chose to learn the guitar. Please list the basic chords I should start with.” Explanation: The highlighted objective asks for specific information on basic chords, directing ChatGPT to provide a beginner’s guide that’s useful for someone in a school music club.
   
  By stating your objective clearly, you’re setting up ChatGPT for success, making it way more likely that you’ll get the answer you’re really after. Up next, we’ll add even more finesse to our prompts by introducing requirements and examples.

   
   

  Requirement: The Nitty-Gritty Details

   

  After you have laid out the Context and Objective in your prompt, ChatGPT would know what it needs to do. However, there are many different ways to carry out the same task, so we need to provide some requirements that specify how to do it.

  There are many things you can include in your requirements. For example, you might specify the tone, such as “formal” or “casual,” or the length, like “detailed” or “brief.” You can even specify a limit on the number of words, though ChatGPT may not be able to strictly control its output word count. You could also ask for the answer to be formatted in a particular way, like a list of bullets or a paragraph. You may also tell ChatGPT what you don’t want in your requirements.

  In many ways, the requirement is very similar to the rubrics your teacher gives you for some assignments. So imagine you are a teacher assigning a task to ChatGPT, then design your rubrics for ChatGPT to follow.

  One commonly used technique in writing the requirement is to assign a “role” to ChatGPT. For example, you can say “Write the answer as an expert in …”, and ChatGPT will use the tone and level of details that are typically used by experts. On the other hand, if you say “Pretend you are 5 years old”, then ChatGPT’s response will look like written by a very young kid with very simple words and sentences.

   

  Examples of Giving Requirements

   

  Math Homework Help: Less Effective Prompt: “I’m a middle school student struggling with my math homework on fractions. Can you walk me through how to solve ( \frac{1}{2} + \frac{3}{4} )?” Effective Prompt: “I’m a middle school student struggling with my math homework on fractions. Can you walk me through how to solve ( \frac{1}{2} + \frac{3}{4} )? Please use simple language and show all steps.” Explanation: The highlighted requirement specifies the language level and detail needed, ensuring that the explanation is both understandable and comprehensive for a middle school student.

   

  Science Fair Project on Renewable Energy: Less Effective Prompt: “I’m working on a science fair project about renewable energy, specifically focusing on solar panels. Please explain how they convert sunlight into electricity.” Effective Prompt: “I’m working on a science fair project about renewable energy, specifically focusing on solar panels. Please explain how they convert sunlight into electricity. Keep it concise but include the basic science behind it.” Explanation: The requirement narrows down the focus to a concise explanation while still including the basic science. This helps prevent ChatGPT from writing lengthy explanations.

   

  Writing a Poem for English Class: Less Effective Prompt: “I have an English class assignment to write a poem about nature. Please give me a template or outline to start my poem?” Effective Prompt: “I have an English class assignment to write a poem about nature. Please give me a template or outline to start my poem? Make sure to include at least three different poetic devices.” Explanation: The requirement specifies the inclusion of different poetic devices, guiding ChatGPT to offer a more educational and enriching starting point for the poem.

   

  Learning to Play a New Instrument: Less Effective Prompt: “I just joined my school’s music club and chose to learn the guitar. Please list the basic chords I should start with.” Effective Prompt: “I just joined my school’s music club and chose to learn the guitar. Please list the basic chords I should start with. Pretend you are an experienced music teacher.” Explanation: By asking ChatGPT to take on the role of an experienced music teacher, the highlighted requirement ensures that the guide will be intuitive and well-designed.

   

  By being meticulous with your requirements, you’re fine-tuning the ChatGPT’s responses to fit like a glove. You get to be the director of how you want your information served.

   
   

  Examples: Show and Tell

   

  So, you’ve given ChatGPT the context, objective, and even the fine details of your requirements. What’s left? Examples! While this step is optional, it can be incredibly effective in getting the response you’re after. If you find ChatGPT’s answers are not what you expect, but you don’t know how to change the objective or requirements, then it is probably a good time to provide some examples directly.

  Examples can illustrate the style, tone, or format you prefer. If you’ve ever heard the saying, “A picture is worth a thousand words,” well, in this case, an example can be worth a thousand guidelines. It’s like giving ChatGPT a mini “preview” of your ideal answer.

   

  Examples of Providing Examples

   

  Math Homework Help: Less Effective Prompt: “I’m a middle school student struggling with my math homework on fractions. Can you walk me through how to solve ( \frac{1}{2} + \frac{3}{4} )? Please use simple language and show all steps.” Effective Prompt: “I’m a middle school student struggling with my math homework on fractions. Can you walk me through how to solve ( \frac{1}{2} + \frac{3}{4} )? Please use simple language and show all steps. For example: ( \frac{1}{3} + \frac{1}{4} = \frac{1 \times 4 + 3 \times 1}{3 \times 4} = \frac{4 + 3}{12} = \frac{7}{12} ).” Explanation: The example provides a sample walkthrough for adding fractions, setting the expectation for the style and format of the answer.

   

  Science Fair Project on Renewable Energy: Prompt: “I’m working on a science fair project about renewable energy, specifically focusing on solar panels. Please explain how they convert sunlight into electricity. Keep it concise but include the basic science behind it.” Explanation: For this prompt, it is not easy to provide an example since you would have to include an entire paragraph that explains another concept.

   

  Writing a Poem for English Class: Less Effective Prompt: “I have an English class assignment to write a poem about nature. Please give me a template or outline to start my poem? Make sure to include at least three different poetic devices.” Effective Prompt: “I have an English class assignment to write a poem about nature. Please give me a template or outline to start my poem? Make sure to include at least three different poetic devices. For example, start with a simile, include some alliteration, and end with a metaphor.” Explanation: The example outlines the structure and poetic devices desired, providing ChatGPT with a blueprint for crafting the template or outline.

   

  Learning to Play a New Instrument: Prompt: “I just joined my school’s music club and chose to learn the guitar. Please list the basic chords I should start with. Pretend you are an experienced music teacher.” Explanation: For this prompt, it is also not easy to provide an example, and it might not be necessary.

   

  In short, providing an example is like giving ChatGPT a cheat sheet for acing your query. It’s your way to make sure you get the answer that fits your needs to a T.

  With this, you’ve completed the C.O.R.E framework, and you’re all set to craft prompts that will make your interactions with ChatGPT as fruitful as possible.

  Absolutely, a hands-on activity is a fantastic way to cement the understanding of the CORE framework. Here’s how you could set up this educational exercise:

   
   

  Hands-On Activity: Fix that Prompt!

   

  You’ve learned about the C.O.R.E framework — Context, Objective, Requirement, and Example. Now it’s time to put your knowledge to the test! Below are some less effective prompts that lack one or more elements of the CORE framework. Your mission is to figure out what’s missing or poorly written, and then rewrite the prompt to make it better.

  For each case, you can try both the original and the improved versions on ChatGPT to see how they compare. Does your prompt yield a better answer? You can keep improving your prompt until you see a clear improvement.

  You can use this chat project to run the tests:

  https://play.creaticode.com/projects/6531b7e60fdce080a4481c1d
  
  

   

  Prompts for Analysis and Improvement Prompt: “Tell me about the solar system.” Hints: Is the context clear? What’s the objective here? Any requirements or examples?

   

  Prompt: “I need to write a story for my class.” Hints: Who are you? What exactly do you need for your story? What should it look like?

   

  Prompt: “I am preparing for my Spanish exam. How do I improve?” Hints: Which area are you looking for improvement?

   

  Prompt: “I’m hosting a party. Give me a recipe for salads.” Hints: Any dietary restrictions? How detailed should it be?

   
   

   
   

  Follow-Up Prompts: It’s a Conversation, Not a Monologue

   
   
  DALL·E 2023-10-28 00.10.54 - Minimalistic illustration with the phrase 'Follow Up Prompt' in bold and clear letters positioned at the top center. Below, there is a user on the lef.png

   

  You’ve got your CORE framework down, but let’s be real — nobody’s perfect. It’s rare to get exactly what you need from ChatGPT with just one prompt. And guess what? That’s totally okay! Working with ChatGPT is more like a conversation than a monologue. Sometimes, you need to clarify, adjust, or even backtrack to get to the answer you’re looking for.

  The key to effective follow-up prompts is to identify what went wrong (or what could be better) with the initial response. Is ChatGPT missing some background information? Were your original requirements unclear? Did your example mislead the model? Once you spot the issue, your follow-up prompt can address it, refining ChatGPT’s understanding and output.

  Don’t worry if you have to go through several iterations. The goal is to gradually narrow down to the most accurate and helpful response, and that usually takes a bit of back-and-forth.

   

  Common Issues and How to Address Them Through Follow-Up Prompts

   

  Missing Background Information

  Explanation: Sometimes ChatGPT doesn’t have enough context to give a tailored response. This usually happens when the initial prompt is too vague or general. In such cases, you’ll likely get a broad or generic answer that might not meet your specific needs. The key is to pinpoint what additional information would make your question clearer and include that in your follow-up prompt.

  Example 1: Initial Prompt: “Tell me about photosynthesis.” Initial Response: “Photosynthesis is the process by which green plants and some other organisms use sunlight to synthesize foods with the help of chlorophyll pigments.” Follow-Up Prompt: “I’m interested in understanding how chlorophyll plays a role in photosynthesis. Could you elaborate?” Improved Response: “Absolutely! Chlorophyll is essential for photosynthesis in plants. It’s the molecule in plant cells that absorbs sunlight and uses its energy to convert CO2 and water into glucose and oxygen.” Example 2: Initial Prompt: “Explain the concept of renewable energy.” Initial Response: “Renewable energy is energy that is collected from resources that are naturally replenished on a human timescale, such as sunlight, wind, and rain.” Follow-Up Prompt: “Can you focus on how solar energy is harnessed and converted into electricity?” Improved Response: “Certainly! Solar energy is harnessed using photovoltaic cells commonly found in solar panels. These cells convert sunlight directly into electricity by the photovoltaic effect.”

   

  Misleading Descriptions

  Explanation: At times, the words you use in your prompt might inadvertently guide ChatGPT down the wrong path. For example, asking for a “quick recipe” could still yield a complex dish if ChatGPT’s definition of “quick” differs from yours. It’s important to clarify any terms or phrases that could be interpreted in multiple ways.

  Example 1: Initial Prompt: “Give me a quick recipe.” Initial Response: “How about Beef Bourguignon? It’s a delightful French dish, although it takes about 2 hours to make.” Follow-Up Prompt: “I’m looking for something quick and easy to prepare. Do you have a recipe that takes less than 30 minutes?” Improved Response: “Of course! A quick and easy option would be a classic Spaghetti Aglio e Olio. You’ll need spaghetti, garlic, red pepper flakes, and olive oil, and it takes just about 20 minutes to prepare.” Example 2: Initial Prompt: “How do I become popular?” Initial Response: “Gaining popularity can involve a myriad of factors such as your social media presence, networking skills, and even your charisma.” Follow-Up Prompt: “I was referring to becoming well-liked among my local community members. Do you have any tips for that?” Improved Response: “Ah, I understand. In a local community setting, becoming well-liked often involves being actively involved in community events, being kind and approachable, and showing genuine interest in the well-being of others.”

   

  Unclear Requirements

  Explanation: If you have specific requirements for how you’d like the information to be presented, it’s crucial to state them clearly. Whether you want a detailed explanation, a summarized list, or any other specific format, not stating this clearly can lead to getting a response that doesn’t meet your expectations. Always specify your requirements in your prompt or follow-up to get the answer you’re looking for.

  Example 1: Initial Prompt: “Write me a poem about the ocean.” Initial Response: “The ocean’s waves dance to and fro, a ballet of nature, don’t you know.” Follow-Up Prompt: “Could you write the poem in a sonnet form?” Improved Response: “The ocean’s waves in moonlight gleam so bright, a dance of liquid sapphire through the night. Each crest and trough an artist’s stroke, it seems, a living, breathing, ever-changing dream.” Example 2: Initial Prompt: “Tell me how to take care of a dog.” Initial Response: “Taking care of a dog involves feeding, exercise, grooming, and regular veterinary check-ups.” Follow-Up Prompt: “Could you provide the information in a detailed list format?” Improved Response: "Certainly! Here’s a detailed list: Feeding: Choose nutritionally balanced food and provide fresh water. Exercise: Daily walks and playtime are essential. Grooming: Brush regularly and keep nails trimmed. Vet Visits: Annual check-ups and vaccinations are a must."

   

  Misleading Examples

  Explanation: Sometimes, you might use an example to indicate the style or format you’re interested in. But if your example isn’t spot-on, ChatGPT might take it too literally or misunderstand your intention. Be mindful of the examples you use and make sure they align closely with what you’re actually looking for.

  Example 1: Initial Prompt: “Write me a story like Harry Potter.” Initial Response: “In a dystopian future, young Sarah discovers she’s a tech wizard capable of controlling machines with her mind.” Follow-Up Prompt: “I meant a fantasy story set in a magical world.” Improved Response: “Ah, got it. In the magical kingdom of Eldoria, a young orphan named Tim discovers an ancient wand that bestows upon him incredible magical powers.” Example 2: Initial Prompt: “Suggest a hearty dinner recipe.” Initial Response: “How about a seafood paella?” Follow-Up Prompt: “I was thinking of something more along the lines of a meat and potatoes kind of meal.” Improved Response: “Ah, I see. How about a classic meatloaf with a side of mashed potatoes? It’s a hearty, comforting meal that’s sure to satisfy.”

   

  Needs Additional Information

  Explanation: Even when you think you’ve provided enough details, sometimes ChatGPT might still need more information to give you the best possible answer. This could be additional context, specific preferences, or other factors that weren’t included in your initial prompt. If the initial response is lacking, think about what additional details could make it more precise.
   

  Example 1: Initial Prompt: “What laptop should I buy?” Initial Response: “The ideal laptop depends on various factors, including what you’ll be using it for, your budget, and personal preferences.” Follow-Up Prompt: “I’ll mainly be using it for video editing.” Improved Response: “For video editing, you’ll want a laptop with a strong processor, at least 16GB of RAM, and a high-resolution display. MacBook Pro and Dell XPS 15 are good options.” Example 2: Initial Prompt: “What pet should I get?” Initial Response: “Choosing a pet depends on your lifestyle, living situation, and personal preferences. Dogs and cats are popular choices, but there are also birds, fish, and small mammals like hamsters.” Follow-Up Prompt: “I should add that I live in an apartment.” Improved Response: “If you live in an apartment, you might consider smaller pets that require less space. Small dog breeds, fish, or even a rabbit could be a good fit.”

   

  Understanding these common pitfalls and knowing how to navigate them can greatly improve your interactions with ChatGPT. A good prompt often requires some iteration and fine-tuning, so don’t hesitate to refine your questions in follow-up prompts.

   
   

  Hands-On Activity: Mastering the Art of Follow-Up Prompts

   

  Objective

  In this activity, you will practice writing follow-up prompts to fine-tune ChatGPT’s responses. You’ll start with some initial prompts that are intentionally less effective. After getting the initial responses, you’ll identify issues based on the categories we’ve discussed and write follow-up prompts to address them. Finally, you’ll evaluate the improved responses to see if they meet your expectations.

  Guidelines for a Great Response Relevance: Does the response answer your question directly and completely? Clarity: Is the response easy to understand, with no ambiguous or confusing parts? Depth: Does the response provide a sufficient level of detail? Tone: Does the tone of the response match what you were looking for (e.g., formal, casual, educational, etc.)? Formatting: Is the response formatted in a way that makes it easy to read and understand?
    Initial Prompts “Tell me how solar energy works.” “What’s a good meal I can make quickly?” “How can I make my bedroom more comfortable?” “How do I get better at playing basketball?” “Explain how a car engine works.”
    Steps Step 1: Copy one of the initial prompts and put it into the same chat project used in the previous activity. Step 2: Examine the response you get. Identify any issues based on the categories we’ve discussed (Missing Background Information, Misleading Descriptions, Unclear Requirements, Misleading Examples, Needs Additional Information). Step 3: Write a follow-up prompt to address the issue(s) you’ve identified. Put your follow-up prompt into ChatGPT. Step 4: Again, examine the new response. Is it improved? Does it meet the guidelines for a great response? If not, go back to Step 3 and write another follow-up prompt. Step 5: Once you’re satisfied with the response, move on to the next initial prompt and repeat the process.
    Example of the Process Initial Prompt: “Tell me how solar energy works.” Initial Response: “Solar energy works by capturing the sun’s energy and turning it into electricity for your home or business.” Issues Identified: Missing Background Information, Unclear Requirements Follow-Up Prompt: “I’m doing a project on renewable energy sources. Can you explain how solar panels capture the sun’s energy and convert it into electricity, in a detailed manner?” Improved Response: “Certainly! Solar panels are made of photovoltaic cells, which contain semiconductors. When sunlight hits these cells, it energizes the electrons, creating an electric current. This electricity is then converted from DC to AC and can be used to power your home. Solar inverters and battery storage systems can also be used for a more constant energy supply.” Evaluation: The improved response is relevant, clear, detailed, educational in tone, and well-formatted, meeting all the guidelines for a great response.

  Feel free to go through all the initial prompts to get the most out of this hands-on activity. Remember, the key to a meaningful interaction with ChatGPT often lies in the art of the follow-up prompt!

   

   

  Chapter Summary

   
  In this chapter, we explored the CORE framework to craft effective prompts for ChatGPT. The CORE framework consists of Context, Objective, Requirement, and Example, each adding clarity and precision to your questions. We illustrated this with detailed examples and explained how to fix common issues through follow-up prompts. Hands-on activities were included to offer practical experience in writing both initial and follow-up queries. By mastering these principles, you’re set to get the most relevant and accurate responses from ChatGPT.

   
   

  Project - An Improved Chat App

   

  It’s time to do some coding now. Please follow this tutorial to build an improved chatbot.

• Programming

  How to design and develop a solution program

  1
  Topics
  1
  Posts

  

   

  Prerequisite The For-Loop Block The Repeat Block The Repeat Until Block The Forever Block

   

  What are Loops?

  A “loop” is a block for playing other blocks repeatedly. Below are some loop blocks you can use on the CreatiCode playground;

  
   

  What are Nested Loops?

  When you put one loop inside another loop, it becomes a “nested loop”. You can even have a third loop inside the inner loop.

  
   

  How Nested Loops Work?

  When you run a program with a nested loop, the inner (child) loop will work like a normal loop. However, because of the outer (parent) loop, the inner loop itself will be repeated.

  Here is a step-by-step example:

  Step 1: The main loop is played.

  
   

  Step 2: The Nested loop is started.

  
   

  Step 3: The blocks inside the nested loop are played until the condition of the nested loop is completed.

  
   

  Step 4: The blocks after the nested loop are played (if there are any).

  
   

  Step 5: The main loop is restarted.

  Note: All of the above 5 steps keep repeating until the condition of the main loop is completed.

  For example, as shown below, “repeat until ” is the nested loop, which keeps the monkey moving until it touches the edge of the stage. And “repeat (5)” is the main loop which repeats the nested loop 5 times.

  
   

  Use of Nested Loops

  We use Nested loops when we want to repeat multiple blocks at the same time, until multiple conditions are completed.

  For example, as shown below, nested loops are used to make the monkey move and eat bananas at the same time.

  
   

• Tools

  Design tools, math tools, collaboration tools, programming tools

  33
  Topics
  35
  Posts

  

  Introduction

   

  Traditional assessment tools, such as multiple-choice quizzes, often fail to accurately measure a student’s understanding and are highly susceptible to cheating.

  As Large Language Models become more intelligent, we can use AI tools to check student understanding in new ways. In this article, we will introduce an example tool, which is a chatbot that can discuss a given program with a student and assess the student’s understanding. All chat messages will be stored in a Google Sheet under each student’s user name, so that the teacher can review them at any time.

  This tool project is shared at

  play.creaticode.com/projects/67d957d70b962dba34213579

  It looks like this when a student runs it:

  chatexample.gif

   

  The chat messages will be stored in the Google Sheet like this:

  1145210a-7e47-4bc4-9672-e2ce35addbb2-image.png

   
   

  To use this tool with your students, you need to do the following:

  Remix the project above Change the Google Sheet URL Change the system instruction. Publish the project and send its URL to your students.

  Next, I will explain how this tool works and how you can make these changes.

   
   

  Overview

   

  As shown below, this program only has 3 key modules:

  Initialize Sheet: adds a new tab in the Google Sheet you provide so that the chat messages with this user can be stored there. Initialize Chat: set the instruction for AI and then get the first message from AI Handle User Message: send new user message to AI and get AI’s response

  3599db8d-e1bb-42fb-8722-3580f72f8bd0-image.png

   
   

  Initialize Sheet

   

  In this module, a few steps are taken to set up the Google Sheet. You will only need to change the Google Sheet link in the first part:

   

  1. Set Sheet URL (TODO)

  f552af37-16d2-4aa9-882e-6a27c29fd089-image.png

  All chat messages with all users will be stored in one Google Sheet, and you will need to provide the link to it. The link should be set up so that “anyone with link” can edit it. To make sure no one else can see the content of this sheet, you should not share this project with students. Instead, you should publish this project and give its URL to your students, so they can run this project without looking inside its code.

   

  2. Set Tab Name

  5a7f7138-47ca-4428-9f2c-885601dc8c10-image.png

  Each student’s chat messages will be stored in one new tab. To easily identify the student, we will use their “full name” from their profile. To handle duplicate full names, the first 4 letters from their random user ID are appended. Note that students must log in to CreatiCode before running this project so their full names in the profile can be used.

   

  3. Add a New Tab

  a384f7fb-d2ec-4e60-a268-76dc81e53e7d-image.png

  We load all tab names into a list, so we can check if the tab for this student already exists. If it does not exist, we will add a new sheet with that name We will remove most rows and columns first. This makes sure there are no empty cells. In Google Sheet, there is a limit that the total number of cells in a sheet should be less than 10 million. We will add the table headers (Time/Who/Message) to the first row of the new tab

   

  4. Add a row of dashed lines

  a7ffe25e-100e-4b47-a7fc-d7bd8244b2c4-image.png

  Sometimes, a student may run this program multiple times. To mark the beginning of a new run, we will add one row with dashed lines.

  For example, the new tab will look like this after initialization:

  98abcb8f-b1fb-40b4-8a8a-69443aa555c0-image.png

   
   

  Initialize Chat

   

  In this module, you will need to modify the system instruction based on your requirements.

   

  1. Set System Instruction (TODO)

   

  30eb7695-0211-4672-8cb2-a87922ef359a-image.png

   
  The system instruction is given to the AI model before the chat begins. It contains all the instructions and requirements. You need to modify it to specify the code that you want the student to discuss and any key points you want to check. Here is an example system instruction:

  62355385-49e0-450d-b601-e0afba5f2104-image.png

   

  The pseudocode is enclosed in the “```scratch” tags. You can get this pseudocode for any scratch program by right-clicking on any stack of blocks, and select “copy pseudocode of this script”:

  c6ff0f43-3226-4afc-9524-a223dddf2e36-image.png

  You can also specify other coding languages after the 3 ticks, such as python, javascript, java, etc.

  Feel free to update the “Key points to check”, so that the AI will try to assess how well the student understand these points about the given program.

  Feel free to update the other instructions below. For example, the example instruction asks AI to conclude the chat with a rating. You can choose to do it differently or not give any rating.

   

  2. Display the chat window

   

  545f047f-0614-4b79-8f03-51f3e15ff70f-image.png

   

  3. Get AI’s initial message

   

  1c9a62a3-3d5c-4b08-b925-aea6ce45ec2e-image.png

   
  To make the AI present the given program to the student and ask its initial question, we simply need to send a message like “hi” to the AI. The response from AI will be added to the chat window.

   

  4. Store AI’s message

   

  ba844c79-f00e-4338-8828-7d0e258e24c0-image.png

  Laslty, this message from AI will be stored in the Google Sheet, using the “store message” block. This block will get the current time, then append one row at the bottom of the tab for this student:

  c1fcd769-9f5d-45b5-a5ff-ccfc1db1f63b-image.png

   
   

  Respond to student message

   

  Whenever the student inputs a new message, we will get AI’s response to it. You don’t need to change anything here.

   

  1. Display and store student message

  7f0c10b7-14a1-4e3e-b043-5f9d0c113745-image.png

  When the student clicks the “send” button, we will append that message (“value of widget [chat1]”) to the chat history. We will use the “store message” block to store this message in the Google Sheet as well.

   

  2. Get AI’s response

  bf221635-d72c-4c2b-9852-ce7989ed4ea1-image.png

  We will send the user message to the AI model and wait for its response. Then we add the response to the chat window Lastly, we also store the AI’s response in the Google Sheet.

   
   

  Summary

   

  In summary, you can reuse the example project by replacing the Google Sheet URL and the system instruction. You can also modify the system instruction to assess the student’s understanding of some concept instead of actual code. You can always review student’s conversation with the AI in the Google Sheet.