From the Cool Cat Teacher Blog by Vicki Davis
Follow @coolcatteacher on Twitter
People everywhere are asking, “What is Chat GPT?” and “Why does it matter?” I'll admit when I first went into Chat GPT earlier this year, I went in and came away not super impressed. Then again, I didn't really know how to prompt it, so I wasn't getting great results. I also didn't know I could give it feedback to improve the prompt. I also didn't really understand how it worked or what it was used to do. So, as I've always done on this blog, my students and I began studying this tool to understand the big fuss.
This post is written entirely by me with Grammarly Pro to help me with typos/comma issues. I'm not an AI expert but a classroom teacher studying this technology to break down the essential high-level elements to teach to my students. I've quoted sources who know far more than I do about this topic to give you places to dig deeper. This is Day 4 of the 80 Days of AI where I'm learning and sharing what I'm learning. There's so much to learn and so much I personally don't know, so I'm always open to connecting and conversing. Twitter @coolcatteacher is where I chat most.
The purpose of this post is to:
Chat GPT is created by Open AI. It is an Artificial Intelligence (AI) Chatbot. This means that you can have a conversation with it.
While the Chat GPT on the Open AI site is text-based, they have created an API. An API allows other tools to use Chat GPT technology. So, for example, the Promptheaus extension or TalkBerry for Google Chrome lets you use your voice to ask Chat GPT questions.
Learning to create good prompts has a new term called “Prompt Engineering,” which, as my friend Steve Dembo said in the post on Thursday, may be important initially but not so important in the long run as AI improves.
Chat GPT
Stands for "Chat-based generative pre-trained transformer." According to the Engage AI Institute
"In this case, the model: (1) can generate responses to questions (Generative); (2) was trained in advance on a large amount of the written material available on the web (Pre-trained); (3) and can process sentences differently than other types of models (Transformer).
In simple terms, as you teach about this tool, understand that:
Point 1. Chat GPT needs to be prompted. (Generative)
Point 2. It was trained by written material ON THE WEB (more on this later) (Pre-Trained)
Point 3. It processes sentences differently than other models out there. This means that no two responses are ever alike. As it uses the algorithms to generate the next word, it gives a different word, so responses are unique. (Transformer)
CLASSROOM OBSERVATION: This is why when my students and I used Chat GPT to write TicTacToe programs in the Python language, all ten were different, and only eight of them worked.
RESOURCE: The Engage AI Institute's Glossary of AI Terms for Educators has some fantastic definitions for more terms than I will share here as you embark on your own learning journey.
Chat GPT is NOT a search engine. It can give you wrong information or “hallucinations” as it is being called. We should verify the “facts” that it tells us, particularly because it does not always accurately cite sources (even when you ask it to.)
If you look at the Chat GPT information from Open AI, you will learn that this tool has been trained on “vast amounts of data from the internet written by humans, including conversations.” This data was captured before 2021, so knowledge of the world and events after 2021 may be inaccurate.
When they initially trained Chat GPT, not everything could be vetted by a human and so no one told whether the information was fact or fiction. So, if it studied a body of false or misleading information, it may think it is true and give you something that isn't. Therefore, you should verify the information found in a search engine.
However, as inaccurate information is caught in the feedback process, the information becomes more accurate, particularly if the original user tells Chat GPT it has an error.
Training Data
Training data is "an extremely large dataset that is used to teach a machine learning (ML) model" (Techopedia) ChatGPT is a type of Machine Learning called a Large Language Model (LLM). According to Molly Ruby of Toward Data Science, "LLMs digest huge quantities of text data and infer relationships between words within the text." OpenAI called this process "generative pre-training (GP).
TEACHING TIP: ChatGPT has studied humans. Humans have biases, make mistakes, and draw wrong conclusions. Therefore, you can expect ChatGPT to have potential bias, make mistakes, and draw wrong conclusions as well.
As you study AI, understanding the data used to train the model is vital to understanding how to use it. Training data has three forms:
Training Data Form 1. Unsupervised ML Models. In unsupervised models, the training data is not labeled.
TEACHING TIP: In their June 2018 abstract, Improving Language Understanding by Generative Pre-Training, Open AI scientists created a method allowing ChatGPT to learn from mass quantities of unlabeled data followed by "discriminate fine-tuning" on each task because it would provide "a significant performance boost.) Therefore, we have to know that much of the data studied by ChatGPT has not been reviewed by a human. However, as they state in their paper, without this, ChatGPT might not exist.
Training Data Form 2. Supervised ML Models. In this case, the training data is labeled in some way. It could be labeled "true or false" or with some other sort of data. Obviously, this could slow down the training process, but if data is labeled accurately, it can provide many benefits to the output. For example, when ChatGPT moved from its 3.0 model to the 3.5 model we are using today, Open AI hired 40 contractors to create a supervised data set. Prompts were collected from user input, and the labelers wrote appropriate responses. So, now the new GPT 3.5 model is called the SFT Model for (Supervised Fine Tuning). That said, it is a huge dataset and much hasn't been reviewed by humans.
TEACHING TIP: We need to know that our prompts can and likely will be read by those hired as part of Open AI's SFT work. So, we should not put anything in a prompt we do not want to read by a stranger. This includes asking ChatGPT to revise a strategic plan, revision of confidential information of any kind, or things that include names, addresses, places, or anything that should be protected for privacy or legal reasons. This applies to any AI tool using SFT. However, it seems you need SFT to create a good model. This is the Human Intelligence (HI) that makes AI so powerful.
Training Data Form 3. Reinforcement. In reinforcement learning, the model's performance is evaluated and it learns based on feedback.
TEACHING TIP: As you interact with ChatGPT you can give it feedback on its response. As you do, you may even see it admit mistakes, converse with you, or seemingly change "opinion." However, the reviewers labeling the data behind the scenes may also review your prompt and the response. For this reason, you should scrutinize each site's privacy policy using AI Chatbots for how their reinforcement learning is happening.
RESOURCE: Read more at 365 Data Science, Toward Data Science, and Open AI's Research Brief.
OpenAI has warned users about not using it for medical information. OpenAI says
“OpenAI's models are not fine-tuned to provide medical information. You should never use our models to provide diagnostic or treatment services for serious medical conditions.”
However, as some users have reported, if you tell it not to give you negative responses, it will only tell you positive things. It does not warn users every time when they ask a question that could be potentially harmful or where it could have dire consequences.
For example, a company named Nabla attempted to see if they could use ChatGPT to see how it would work in healthcare use cases. In one case, the GPT-3 version actually recommended suicide to a pretend patient.
For example, Rob Morris, founder of Koko, a free therapy program admitted on Twitter that they were using GPT-3 to provide mental health support to their users. They had a “copilot” approach where humans supervised the AI “as needed” on about 30,000 messages. He did this without any further consent than the standard agreement users had signed when they signed up for the service. Many condemned this as highly unethical, and, personally, I agree. This is a case study to help us understand that chatbots may not be disclosing their use of the ChatGPT technology and trusting chats without such disclosures should be concerning to us.
TEACHING TIP: Every teacher studying this tool should read Open AI's “Disallowed usage of our models.” on their Usage Policies. While you might not share this list with students (for obvious reasons), if it is on the list it is because it has been observed.
Open AI Disclosure Requirements
“We have further requirements for certain uses of our models:
- Consumer-facing uses of our models in medical, financial, and legal industries; in news generation or news summarization; and where else warranted, must provide a disclaimer to users informing them that AI is being used and of its potential limitations.
- Automated systems (including conversational AI and chatbots) must disclose to users that they are interacting with an AI system. With the exception of chatbots that depict historical public figures, products that simulate another person must either have that person's explicit consent or be clearly labeled as “simulated” or “parody.”
- Use of model outputs in livestreams, demonstrations, and research are subject to our Sharing & Publication Policy.
You can use our free moderation endpoint and safety best practices to help you keep your app safe.”
From OpenAI: Usage Policies.
The Risk of ChatGPT Use by Minors
My Dad always told me, “don't gamble with what you cannot afford to lose.”
The risks with Chatbots that can authoritatively mimic human language without the morality or ethical compass to understand the implications of response can be very dangerous. To eagerly jump into AI without understanding the harmful implications is foolhardy.
Our children are asking ChatGPT questions about their mental health, and as I've said in my podcast Generative AI, ChatGPT, and Learning, age considerations and mandatory disclosures should happen. While the requirement for ChatGPT use is 18, I know many students who joined before the age question was asked. And without verification, students will do what they have done throughout the ages – lie about their age to get access to the “free” technology they want.
They don't understand privacy concerns or that the content could be inaccurate. This is why we're discussing, and I'm teaching about this tool. The risks are too great for my students to be uneducated.
Asilomar AI Principles
Recently, a group of AI scientists and leaders, wrote an open letter demanding the halt of any technology past GPT 4 being developed for the next six months.
(I pondered this on my own blog post Should AI Be Stopped as many of the business people signing the letter, I wonder if they signed it for competitive reasons vs. altruistic reasons. However, as a former pastor once said "low motivation is better than no motivation" so the concerns remain valid.)
In the Asilomar principles state guidelines for research, ethics, and values, and long-term issues with the goal of AI being "to create not undirected intelligence, but beneficial intelligence." It also states that "teams developing AI systems should actively cooperate to avoid corner-cutting on safety standards." Furthermore, it talks about avoiding an arms race in lethal autonomous weapons, something which should also concern every person on this planet.
Teaching Tip: Discuss the Asilomar AI Principles with students. Read them yourself. Consider the implications if we use AI Tools from companies not committed to these principles or if the government does not consider regulations of industries using AI tools like ChatGPT.
Resources: Asilomar AI Principles, Pause Giant AI Experiments: An Open Letter, OpenAI: Usage Policies
First, sign up for an account at https://openai.com/blog/chatgpt. Remember that while it uses the GPT 3.5 or GPT 4.0 engine, it isn't the same because ChatGPT contains additional instructions from OPENAI.
Is ChatGPT free? Remember that you start with a “Research” version when you sign up. The “free” version will not work when they are over capacity. However, a paid version for $20/month lets you use the GPT4 engine and always gives you access to ChatGPT.
The ChatGPT Interface
It looks so simple, but let's go through the four screens. Amazingly, in just four screenshots, you'll know all you need to know about interacting with ChatGPT's interface.
Note that in this example, I'm using a sample prompt recommended for college students studying tough subjects on the prompt site FlowGPT. (I do not endorse the word “dumb” but nevertheless wanted the experience of using this prompt as is.)
The Chat List Box
This box on the left saves your chats. You can rename the titles and delete chats. However, deleting chats from the left does not delete it from the training data in OpenAI. It remains in the data attached to your account inside their database.
Selection of the Engine This Chat Will Use
In the pro version of ChatGPT, you can select the current 3.5 engine or the experimental 4.0 version. You cannot change this after you have begun. GPT-4 has limitations in the number of prompts you can give it a day.
Prompt
In this prompt, I created a bot named LAN GPT. This was a prompt idea I located on a prompt website. These prompt idea websites are springing up. While I can copy this prompt, once I've created the "bot" I can return to it in my chat list on the left. It gets "smarter" as I provide it more feedback and use it more.
Response
In this case, I've trained ChatGPT to respond in a way to set up our interaction. However, it will respond with the typical icon as you interact with it.
Prompt Engineering
When I started, I didn't get great results because I didn't know what to ask it. While this may be the word of the moment, there is some truth to knowing how to ask it questions. I've included some examples to get you started and an infographic I've created to help “prompt” some ideas for those of you experimenting.
I will return to this on Monday's post as I don't want this post to be too long already. I will give you a few prompts to get you started and then share the infographic.
OK Prompt
“Write a story.”
Better Prompt
Write a story about two boys named Bob and John who live in Alaska and have a dog named zipper. They fight a polar bear and save their village.
Best Prompt
“Write a 500-word story about two boys named Bob and John who live in the cold wilderness of Alaska. They have a dog named Zipper who likes to run off at the worst times playfully. They fight the dangers of the wilds of Alaska, including a polar bear, and save their village. Use descriptive language depicting the beauty of Alaska as you tell the story.”
So, in case you decide to copy and use my prompt, I'm going to share the output I received from GPT-4 with this last prompt. See what you think and compare what it tells you in your story. See how they are different?
Teaching Tip: If you teach older students who may have an account, use the same prompts as I did in my AI Investigation Lesson Plan here and have discussions about how they are different. Or modify them.
Prompt Engineering Tips
I will blog on this more next week as part of the 80 Days of AI, but here's the infographic I've created to use for that post so you can get started on your journey.
Intentional learning is a decision. I hope this gets you started on your journey. I look forward to learning from you. Please share your resources and thoughts in the comments here or message me on Twitter. And come right back here or join my newsletter in the box below to join me for #80DaysofAIandHI. I'm so excited to learn.
And as I've said before. If you want to learn, write about it. And to write about it, you have to use your Human Intelligence and write it yourself. Good writing will be part of learning, so let's figure out how to make it happen. And if you're concerned about academic honesty, I shared some about what to do when you suspect a paper is written by AI. This is an emerging area as well.
Have a happy day, and always keep learning!
The post What is Chat GPT? appeared first on Cool Cat Teacher Blog by Vicki Davis @coolcatteacher helping educators be excellent every day. Meow!
from Cool Cat Teacher Blog
https://www.coolcatteacher.com/what-is-chat-gpt/
No comments:
Post a Comment