1. AI Episode 1: Intro to Artificial Intelligence in Teaching
  2. AI Episode 2: What Does An AI Teaching Assistant Look Like?
  3. AI Episode 3: Implications for Thought Leaders and Policy Developers
  4. Introducing Simulations into Teacher Preparation Programs
  5. Assistive Technology to Support Writing
  6. Enhancing Instruction and Empowering Educators with AI Tools and Technology
  7. So, AI Ruined Your Term Paper Assignment?
  8. Step by Step Use of Chat GPT
  9. CIDDL ChatGPT: Summarizing Text
  10. CIDDL ChatGPT: Solving Multiple Choice Questions
  11. Equity, Diversity, and Access to Technology in the Age of Artificial Intelligence
  12. CIDDL ChatGPT: Writing Programs
  13. CIDDL ChatGPT: Solving Word Problems
  14. Artificial Intelligence: Positives and Negatives in the Mathematics Classroom
  15. AI to Support Literacy
  16. Using the AI Bill of Rights to Guide Education’s use of AI and the European Commission’s “Ethical Guidelines for Teaching and Learning” to Guide the Future of AI in Education Part 1 of 2
  17. Using the AI Bill of Rights to Guide Education’s use of AI and the European Commission’s “Ethical Guidelines for Teaching and Learning” to Guide the Future of AI in Education Part 2 of 2
  18. Three Free & Easy Tools to Support Tiered Reading in Your Classroom
  19. The Question of Equity in the Age of ChatGPT
  20. CIDDList: 5 AIs You Need to Check Out This Summer!
  21. Mixed Reality Simulations, Personalized Learning, AI, and the Future of Education with Dr. Chris Dede
  22. Foundations for AI and the Future of Teaching and Learning from the US Department of Educational Technology
  23. Apple Enters the AR/VR/MR/XR Scene
  24. ChatGPT, AIs, and the IEP?
  25. There’s An AI for That: A Site Dedicated to Curating AIs
  26. UDL, Design Learning, and Personalized Learning
  27. Embracing the Future: How Teachers Can Harness AI at the Beginning of the School Year
  28. Empowering Special Education Faculty: Navigating the AI Landscape in Higher Education for 2023-2024.
  29. CIDDList: Back-to-School Checklist for Technology in Teacher Preparation Courses
  30. Cracking the Code: Students with Disabilities in the Computer Sciences 
  31. UNESCO Discusses Artificial Intelligence
  32. AI-integrated Apps for Those with Visual Impairments: Camera-Based Identifiers and Readers
  33. Publishers Respond to Generative AI
  34. K-12 Generative AI Readiness Checklist
  35. CIDDL Talks How AI Will Change Special Education at TED
  36. Re-designing and Aligning an Intro to Special Education Class to the UDL Framework through Technology Integration: Minimizing Threats and Distractions
  37. Resources for Learning About AI Going Into 2024
  38. Artificial Intelligence in Education 2023: A Year in Review
  39. Revolutionizing Mathematics Education in K-12 with AI: The Role of ChatGPT
  40. Image Generating AI and Implications for Teacher Preparation
  41. Are We There Yet? AI for Statistical Analysis
  42. Answers to Your AI Questions: A Conversation with Yacine Tazi
  43. Emerging Trends in Special Education Technology: A Doctoral Scholar Symposium
  44. 2024: A Space Odyssey? How AI and Technology of the Present Compares to HAL9000 and the Predictions of 2001: A Space Odyssey
  45. Using ChatGPT for Writing Lesson Plans
  46. Updates in the World of AI
Screen Shot 2024-02-27 at 9.04.39 PM

Answers to Your AI Questions: A Conversation with Yacine Tazi

Authors: Yacine Tazi and Samantha Goldman; info@ciddl.org

CIDDL has been actively engaged in the conversation around AI and its impact on life, but especially education and personnel preparation. Beyond the conversations of our national center, AI has been the focus of local and national news. The more engrossed our population becomes with AI, the more niche and specialized terminology enter the mainstream vernacular. In this blog post, CIDDL invited Yacine Tazi, a doctoral candidate at the University of Central Florida, to share his insights on questions about AI. Yacine’s background includes an undergraduate degree in Computer Science, with a minor in Finance, and a Master’s in Business Data Analytics. Currently, he is working towards a PhD in Methodology, Measurement, and Analysis, with an area of focus in Psychometrics, specifically Measurement Invariance.

Training Data

  1. What does it mean when people say the AI was trained on data? What data was it trained on? How was it trained?

Yacine: When people say training data, it means the AI system learned from large amounts of information, like texts and prompts. These data teach the AI to understand and generate responses. The training process involves showing the AI examples and adjusting its internal rules to improve accuracy. This way, it learns to recognize patterns, interpret language, and make decisions.

2) What is training data?

Yacine: Training data can be things such as prompts, code, pdfs, docs, etc. that you feed to the AI and gives its knowledge about whatever you gave it.

3) Does the AI train off the information I give it?

Yacine: Yes with documents and prompts you give it as well as the conversation history itself. That's why when you use ChatGPT, it's good practice to save your chat history and stick to the topic of that chat window. For example, if you asked ChatGPT a bunch of questions about World War 2, then it would assume (to some degree) that any future questions would be related to World War 2.

Tokens and APIs

4) What’s a token?

Yacine: In AI, tokens are chunks of text/data that are fed into an AI model (such as ChatGPT) and then the model processes them.

5) Is a token the same as a query or prompt that I give the AI?

Yacine: No. When you talk to the AI using a question or instruction, it breaks down what you said into smaller pieces called “tokens”. This helps the AI understand and think about your question before it answers. The longer the prompt, the more tokens are used.

6) Explain APIs.

Yacine: An API, or Application Programming Interface, is like a set of instructions or a menu. software developers use an API to tell a computer program what they want it to do or what information they want it to provide. This makes it easier for different computer programs to work together, share information, and get tasks done automatically.

Confusingly enough, “token” is also a term often used for API “keys”, which are strings of characters that authenticate a user or application to whatever service they are trying to connect to (such as Open API Assistants).

7) Should I be concerned about my token usage?

Yacine: It depends on how you're using it, which model you're using, and for what purpose. If you're using the free version of ChatGPT, then you can use it as much as you want. For the paid version of ChatGPT, they don't limit you per token, they limit you per message. Currently, the cap is 50 messages every 3 hours. So it's recommended to try to get the most out of each message, if you can. For example, if you want to rewrite an email, try to be as specific as you can be, so that you don't go through a lot of iterations and use up your current message cap. The reason why ChatGPT limits you, at the moment, is that it's very expensive to run AI on servers. Personally, even with daily usage, I have only hit the limit twice. But, there are reports that some people talk to the AI for hours everyday, which costs the company a lot more than the 20 dollars a month they charge. Right now, many of the AI companies are taking losses in revenue because they want to promote their product and want people to use it. I foresee in a year or two, that using the “best” AI models may cost a lot more than the current prices.

For APIs, many APIs charge you to use them and, in the case of OpenAI, you pay for tokens to use the tool. You can prepay a certain amount or auto renew with or without limits. When you run out of tokens, you can no longer use the API and your software will stop working. 

8) I’ve used ChatGPT but have never paid for it. Am I stealing tokens?

Yacine: No, because the company is aware that you haven't paid for it. That's why they are giving you access to an AI that is nowhere near as powerful and sophisticated as the paid version. Think of it as a sample you get at a supermarket. Some people take the sample and are content with it and move on, while others really like and decide to purchase it. You are not stealing by sampling. 

Using ChatGPT and Other Large Language Models

9) What are the big differences between the free and paid ChatGPT?

Yacine: There are two major differences. First, is the access to updated information. The free version was last updated September 2021 and is not connected to live internet. With the paid version, the last updated April 2023 but, technically is connected to live internet through the use of GPTs/Plugins. The second difference is the ability to create and use plugins and GPTs. Plugins are add-ons for ChatGPT, such as ScholarAI that would look at research databases for articles. Plugins are pretty much outdated GPTs. GPTs are custom “chatbots” that are trained specifically for a task. For example, there is a Kayak GPT that is connected to Kayak.com’s services and can be used to plan trips looking at upcoming hotels/flights. 

People can also make custom GPTs. For example, I have one called Psychometrics (my dissertation field) that is prompted to be a dissertation assistant. It helps me with the mathematics section of my dissertation topic, as well as, helps APA 7 formatting and grammar rules. To create my custom GPT, I trained it by uploading my dissertation proposal as well the main articles related to my dissertation. So whenever I talk to it, it has preloaded knowledge of my dissertation.

10) ChatGPT, Gemini, Perplexity… which should I use?

Yacine: There is a lot of debate about this. I have heard some are better for certain tasks than others. For example, I have heard Gemini is better when given “roles” for certain perspectives. Personally, I have only played with ChatGPT and StableDiffusion. It's currently an arms race, so some may be better than others depending on the current update or which company has lower costs. For example, ChatGPT wasn't profitable till last quarter and Microsoft is probably taking a huge hit in server costs with promoting Copilot.

11) I heard that it costs money to ask ChatGPT questions. I don’t pay for it. Who does? 

Yacine: That depends on what model and the context. When using the free version, OpenAI is taking a financial hit. It costs money to run AI models. The main reason it’s free is to promote the product. Also, this incentivizes the user to pay for ChatGPT plus. 

AI and Education

12) How do I protect my privacy and my students’ privacy when using AI?

Yacine: You can opt in to say “don't use my data for training purposes or improve the model” but, who knows to what extent the companies (such as OpenAI) actually keep private or not. Best practice is to not upload confidential or private student data (or any confidential private data, for that matter), including names, birthdates, or other personal information. 

13) Should I teach my students to use AI?

Yacine: I personally think yes. People who don't use AI aren’t going to be as prepared as people who do use it. For example, think about Excel. There are people who don't use Excel and rather map things out with pen and paper or use the functions of Word. While these processes may work, their efficiency and functionality are way behind someone who is proficient at Excel. This is very clear in the computer science field, especially with entry level positions. Why would I hire someone who takes a day to write some code where someone else who uses AI can do it in 5 minutes?

Implications for Teacher and Personnel Preparation

There are several takeaways and implications from the above for teacher and personnel preparation. First, as Yacine mentioned, teaching our pre-service personnel how to utilize this technology and how to implement it in their future classrooms with their future students is imperative. AI, and technology as a whole, always is, and continues to, rapidly change and evolve. Exploring these emerging and innovative technologies with our pre-service personnel helps positively position them to use what might come in the future. Next, it’s important to realize that as teachers and teacher educators, we will never know it all. Having resources to learn about these emerging and innovative technologies is essential. While having humans in your network, such as Yacine, is one method to answer your questions, it is not the only option. Believe it or not, you can simply ask AI, such as ChatGPT, to answer your questions. And, best of all, if the answer still doesn’t make sense, you can ask the AI to rephrase it or, even, change the comprehension level. In fact, it’s a great idea to teach your pre-service personnel how to find answers to their questions themselves. Rather than the teacher educator being the expert, teach them to ask the AI their questions. Come up with a list of questions as a class and assign students to ask questions to different AI and compare the responses. Let the pre-service personnel, with the assistance of the technology, lead their own learning. 

Join the Conversation in our Community!

We answered A LOT of questions in this blog post. Personally, as Yacine answered one of my questions, I had at least five follow-ups. Do you have follow-up questions or thoughts on the topics discussed here? Join the conversation in our community!