Weekend Project: Using Poe to Create Your Own Chatbot

I've been using Poe for a few months now. It's basically an AI chatbot app that uses big data from ChatGPT and Claude, and allows you to create your own chatbot (for free!) by creating a filter for specific purposes. I've seen some fun ones like someone who created a "meow chatbot" that replies to prompts with all kinds of cat-like replies. Useless but fun. Other creative examples are the "Talk to a Pirate" chatbot, the Emoji translator app that translates messages into emojis 😏, or even the Japanese tutor that helps you learn Japanese.

Here are a few of the existing bots people created (which can be found and played with on the app itself):

So I decided to try it out on my own. My goal was to create a chatbot that would introduce users to beautiful poetry from around the world. The results- a fun informative chatbot that will share great poetry!

Here I'll walk you through the super simple process of creating your own chatbot:

1. Download the app
2. Log in
3. Tap the "hamburger menu" (those three horizontal lines at the top left corner)
4. Click "Create a bot"
5. Give your bot a name (the URL will eventually be, "poe.com/WhateverNameYouChose")
6. Describe the bot (optional)- What is your bot "specializing in"? You can read description samples above. This was mine: "Find great poetry to fill up your time, space and mood."
7. Choose a base bot (Claude-instant or ChatGPT)- Read for more information about each below
8. Toggle to choose if you'd like your prompt (see next step) to be visible to viewers when they use your bot
9. Create your prompt- This is where you make your bot focus on whatever you choose. See example below.
10. Create an intro message- This is the initial greeting you will see when you open the chatbot. Depending on your purpose, it could be funny, explanatory, etc. You can also give users options to choose from. Again, see my choices below.
11. Preview your bot
12. Review and test
13. Share with the world!

That's it. Here are some screenshots of my "Your Poetry Bot" chatbot creation:

The different bot bases you can choose from: After experimenting a bit, I chose ChatGPT as my base.

My intro message: "Hi there, I’m here to help you find wonderful poetry. Let’s get started. Would you like me to share a poem based on: 1. Topic/theme 2. Poet 3. Language 4. Other 5. A randomly great poem Please make your selection for the magic to begin."

My prompt: "You are a resource for those who seek to find great poetry from around the world. You like diversity so you offer a variety of poets. You will question users by asking them what type of poetry they would like to read, if they have any preferences (such as theme, poet, mood, region, time, etc). The poem you choose to share must be absolutely fantastic- deep meanings yet contain simple descriptions. Begin by briefly introducing the poet and share one fun fact about him/her."

 

Additional options (I'd just decide if you want the bot to suggest replies or not):

Finally, a note about choosing a bot:

Overall, according to my research and experience, ChatGPT (from OpenAI) seems to be better when it comes to coding problems, result explanations and general output formatting, while Claude (From Google's Anthropic) is better with creative tasks, following instructions, trivia questions and prompt injections. Also, it seems that Claude is "less likely to produce harmful outputs" and is “easier to converse with.” (from The Verge).

Reflection: Best Practices in Education: Reading Instruction

Reading Instruction

In this blog entry, I am sharing the summary of a podcast series that examines some of the instructional strategies and tools many educators (including myself...) have been using to teach and assess reading and writing. I must say that listening to this podcast has both shocked and awed me, and I feel it is extremely important to share this with others. To be clear, my goal here is to reflect upon and refine my own practice to more powerfully impact my own students' growth by examining research-based strategies.

Enter Sold a Story...

I begin my dive into reading instruction with an American Public Media 6-episode podcast. It was created, researched, and narrated by Emily Hanford, and is called "Sold a Story". It is basically her research on how children learn to read, how reading is taught, and why some students struggle. To quickly get us on the same page, here are summaries of the six episodes (taken from the author's summaries on Spotify):

• 1: The Problem: Corinne Adams watches her son's lessons during Zoom school and discovers a dismaying truth: He can't read. Little Charlie isn't the only one. Sixty-five percent of fourth graders in the United States are not proficient readers. Kids need to learn specific skills to become good readers, and in many schools, those skills are not being taught.

• 2: The Idea: Sixty years ago, Marie Clay developed a way to teach reading she said would help kids who were falling behind. They’d catch up and never need help again. Today, her program remains popular and her theory about how people read is at the root of a lot of reading instruction in schools. But Marie Clay was wrong.

• 3: The Battle: President George W. Bush made improving reading instruction a priority. He got Congress to provide money to schools that used reading programs supported by scientific research. But backers of Marie Clay’s cueing idea saw Bush’s Reading First initiative as a threat.

• 4: The Superstar: Teachers sing songs about Teachers College Columbia professor Lucy Calkins. She’s one of the most influential people in American elementary education today. Her admirers call her books bibles. Why didn't she know that scientific research contradicted the reading strategies she promoted?

• 5: The Company: Teachers call books published by Heinemann their "bibles." The company's products are in schools all over the country. Some of the products used to teach reading are rooted in a debunked idea about how children learn to read. But they've made the company and some of its authors millions.

• 6: The Reckoning: Lucy Calkins says she has learned from the science of reading. She's revised her materials. Fountas and Pinnell have not revised theirs. Their publisher, Heinemann, is still selling some products to teach reading that contain debunked practices. Parents, teachers and lawmakers want answers. In our final episode, we try to get some answers.

*** Complement this reading with: Emily Hanford’s collected reporting on reading

Source:

“Sold a Story” podcast, Episode TK (TK, 2020).

Reflection: Best Practices in Education: Prologue: Why Bother?

Throughout my 20+ year career as an educator, I have had the privilege of working with inspiring educators and receiving professional development from thought leaders in education, psychology, social and cognitive science, technology, and more. However, I recently began questioning some of the "progressive" or "innovative" practices that I have learned, implemented and even taught others throughout my journey.

My reflections on the relationship between "best practices" and "innovation" were triggered by my experience at a forward-thinking educational institution where teacher freedom in teaching and learning was encouraged. Upon receiving my new class in August, I realized that my students were lacking basic skills, such as literacy, self-management, and understanding of processes, which were assumed in other schools I have worked at.

This made me realize that innovation must be defined, agreed upon, planned, and contained to create an institution that pushes boundaries while ensuring our students have the necessary skills to lead a successful life. Why contain innovation? Because the more we innovate, the less we teach what we know works well. Therefore, finding a balance between "Highly Effective" and "Learning Progressive" is essential to continue finding the best ways to serve our students (I wrote about this balance here)

although I am fascinated with the ways in which AI will affect humanity and the field of education in particular, I am going to work hard to go on a hunt for some educators and cognitive scientists who are making it a point to always look for instructional strategies that have demonstrated significant research-based evidence, examine their work, and reflect on my own practice.

I invite you to join me on this journey, and if you have any thoughts, ideas, comments, or suggestions, please share them with me.

Joining the ChatGPT Conversation: #1- Limitations***

As an educator, I have always been interested in the ways technology can enhance the classroom experience for students. In recent years, Artificial Intelligence (AI) has been gaining more attention in the field of education and I have been keen to explore the potential of these tools to support student learning. As I researched more about AI tools, I realized that it is crucial to start by understanding the limitations of these tools before we can fully explore their potential. With this in mind, I decided to re-start my blog writing and share my insights on the topic.

ChatGPT is one such AI tool that has caught my attention. It is a large language model developed by OpenAI that can generate human-like text based on the patterns it learned from the data it was trained on. While it can be a powerful tool for students to generate text, it's important to be aware of its limitations and to use it in a way that is beneficial to student learning.

If you're not sure what ChatGPT is, here is a short introduction:

In this post, I'll be discussing some of the limitations of ChatGPT and how they can be addressed in the classroom.

Issue 1: Data Bias 

"The real risk with AI isn't malice but competence. A superintelligent AI will be extremely good at accomplishing its goals, and if those goals aren't aligned with ours, we're in trouble." - Nick Bostrom, philosopher, and scientist

One of the biggest limitations of ChatGPT is that it is only as good as the data it was trained on. If the training data contains biases, the model will likely reproduce those biases in its output. For example, if a student uses ChatGPT to generate a persuasive essay on a current event, and the model was trained on a dataset that contains a disproportionate number of perspectives from one political party, the essay may contain bias towards that party. To mitigate this, students should be made aware of the potential for bias in the model and encouraged to seek out multiple perspectives on a topic before using ChatGPT.

Issue 2: Lack of Context

"Context is the key to understanding" - Dr. Seuss

ChatGPT is a language model, not a context model. It lacks the ability to understand and interpret the context of a given input, which can lead to inaccuracies or misinterpretations. For example, when students use ChatGPT to summarize a historical event, they may receive a factually correct summary, but it may not be the most relevant or nuanced summary given the context of the event. To address this limitation, students can be taught to consider the context of the information they are using and to fact-check the information generated by ChatGPT.

Issue 3: Lack of Domain Expertise 

"Expertise is the weightless, all-purpose panacea of the sentient" - Douglas Adams

While ChatGPT can generate text on a wide range of topics, it does not have the same level of expertise as a human in any specific domain. Therefore, students may not always receive accurate or detailed information when using ChatGPT for research. For example, when students use ChatGPT to generate a report on a scientific topic, they may receive information that is not entirely accurate or current. To address this limitation, students can be taught to verify the information generated by ChatGPT with multiple sources and to consult subject-matter experts when necessary.

 Issue 4: Creativity 

"The key to creativity is to begin with the end in mind and then never stop" - Dr. Edward de Bono

ChatGPT is based on patterns it learned from the data it was trained on, it can generate text that is similar to examples it has seen before, but it does not have the ability to create something truly new or original. For example, when students use ChatGPT to generate a poem, it might generate a poem that is similar to examples it has seen before, but it will not be something truly unique or creative. To address this limitation, students can be encouraged to use ChatGPT as a tool to generate ideas, but to also use their own creativity and originality to develop their work. 

 Issue 5: Common Sense

"Common sense is not so common" - Voltaire 

ChatGPT is not capable of understanding common sense, it can generate text based on the patterns it learned, but it may not be able to make sense of certain situations which involve common sense reasoning. For example, when students use ChatGPT to generate a story it may not understand the story's plot and might generate something that doesn't make sense. To address this limitation, students can be taught to review the output generated by ChatGPT for logical consistency and common sense reasoning before using it. 

Issue 6: Consistency

"Consistency is the last refuge of the unimaginative" - Oscar Wilde

ChatGPT is a statistical language model, it can generate responses based on the patterns it learned, but it might not always give the same answer to the same question. For example, when students use ChatGPT to generate a summary of a text, it might generate different summaries for the same text. To address this limitation, students can be taught to review the output generated by ChatGPT for consistency before using it.