I've been using GPT-4 quite a bit over the past year to help me develop learning materials.
Let me share a few examples.
I teach a bunch of different 1st and 2nd year comp sci courses and I have found that this tool saves me SO much time when creating learning materials for my courses. For example, in my introduction to Operating Systems course, I can feed GPT-4 a set of slides and ask it to come up with some relevant C coding for Linux exercises based on the material. It does this pretty darn well. Or I can provide a complex question and ask it to create 3 simpler questions that prepare students for the more complex question.
And for "Teaching the AI" I have given tutorials on how to use GPT-Builder tool that is available with GPT-4 to set up a custom GPT to assist learners in creating data visualization code. You can give the GPT pre-fab instructions, some conversation starters, and any extra knowledge you want it to have access to (ie. you can upload CSV files).
There are a bunch of these pre-fab GPTs out there for GPT-4. I have messed around with Khan Academy's "Khanmingo Lite" which is a code tutor and with a tool called GeoGPT+ which assists you in creating GIS maps from a dataset.
And finally, I second your comments on GPT-4's propensity for hallucinations. It is essential to research and back-check everything it produces as it is predictably unpredictable in its responses.
I love this post. I downloaded the paper and will apply what is in there for sure. I teach undergraduate and graduate business students, and I found it is more and more difficult to come up with something that will induce them to go deeper into their studies. AI is a help here, and a great one. My master one class is now hooked on the subject - leadership - because we solved problems with AI, and they enjoyed it, and I saw a lot of "AHA" moments in class. The book is great, it is not just a good read, but also I feel that I live the science-fiction I once dreamed of.
I think this is VERY important work. I'm spreading it around. I think any organization that's using LLMs needs to have its own mini-culture around them. On the one hand there's just messing around and having fun. But that over time that exploration becomes focused and results in useful tools and strategies. This kind of thing needs to filter down into the middle schools.
Thanks as always, esp on the ask to share prompts and learnings. Tomorrow we start a prompt hackathon in the german speaking Corporate Learning Community and I definitely will try to translate results so that it’s available also tour wider target group.
Ethan, I finished your book last week. It's good! Everyone, go listen or read.
I have found myself agreeing with all of this, especially level 4. I think understanding that you can get what you want through a few corrective iterations is an incredibly helpful piece of the overall strategy, and one reason it works really well for me.
Just ask the AI to do something, see what it does, and ask to modify on that basis. This is one way to very quickly learn what an AI is capable of in the first place; I often try not to be overly specific with the first prompt.
I am sure you hear this all the time, but I'll say it anyway—you are my most important 'distant teacher' about Generative AI in the classroom. Thank you for staying honest, consistent and tuned in to what educators need to learn, try, and question.
Your post on April 22 addressed the necessity for folks in education to share out about their experiments with AI, and I could not agree more. At the university where I teach, we have a faculty study group that meets monthly to talk about how we are using AI—right now, it's a small but mighty core of instructors, staff and administrators and we are learning a lot.
With another set of teacher colleagues from K-12 classrooms, we have begun publishing essays about how we are using GenAI tools with pre-service and in-service teachers to support the quality of our teaching (and to show how GenAI creates efficiencies that free up time to do the parts of teaching that mean the most to us, but that often get swallowed up by necessary-but-time-consuming tasks.)
You can see our first couple of essays in the April 19 edition of Big Fresh (https://choiceliteracy.com/big-fresh/), a newsletter published by Choice Literacy. On the landing page, you can register for free or just scroll down and click on the header for the April 19 post.
Thx for this it is great stuff and very timely. I am an IO psychologist and I am currently working on a role play based assessment. We are tracking pretty nicely with your ideas, but there is some good stuff here that will help us elevate our game!
Love the experimentation and insights you share. In support of the principle "Just like instructors often share lesson plans, or members of organizations might share tips for success, people can also share prompts. This is a useful starting point for would-be-innovators; someone else has done the work of creating and testing a prompt for you. By using other people’s prompts, you can see what an AI system can (or can’t) do." It would be helpful to have the prompts you use built in a publicly shareable location where people can iterate off of what you built as a foundation. Something like here: https://gooey.ai/compare-large-language-models/
This is a great article. I have been taking online courses on prompt engineering, but truthfully none of them come close to the clarity of this article. I have been building my own GPTs on OpenAI, and will definitely put your tips to use!
Thank you for this, I work with a lot of founders and this is helpful to my agency. Question, Do you think that the skill of prompt engineering (what this basically is) will be useful in the next 2 years? Like do you think it is only valuable now? because I am assuming in the next 2 years that we will have AI LLMs creating our prompts for us.
Amazing post. I read your book which was quite insightgul in many respects. But you are doing God's work with the dissemination of this kind of post. I am an entrepreneur and see many parallels in educating employees on different topics
Hi Ethan,
Thanks for sharing this article!
I've been using GPT-4 quite a bit over the past year to help me develop learning materials.
Let me share a few examples.
I teach a bunch of different 1st and 2nd year comp sci courses and I have found that this tool saves me SO much time when creating learning materials for my courses. For example, in my introduction to Operating Systems course, I can feed GPT-4 a set of slides and ask it to come up with some relevant C coding for Linux exercises based on the material. It does this pretty darn well. Or I can provide a complex question and ask it to create 3 simpler questions that prepare students for the more complex question.
And for "Teaching the AI" I have given tutorials on how to use GPT-Builder tool that is available with GPT-4 to set up a custom GPT to assist learners in creating data visualization code. You can give the GPT pre-fab instructions, some conversation starters, and any extra knowledge you want it to have access to (ie. you can upload CSV files).
There are a bunch of these pre-fab GPTs out there for GPT-4. I have messed around with Khan Academy's "Khanmingo Lite" which is a code tutor and with a tool called GeoGPT+ which assists you in creating GIS maps from a dataset.
And finally, I second your comments on GPT-4's propensity for hallucinations. It is essential to research and back-check everything it produces as it is predictably unpredictable in its responses.
I love this post. I downloaded the paper and will apply what is in there for sure. I teach undergraduate and graduate business students, and I found it is more and more difficult to come up with something that will induce them to go deeper into their studies. AI is a help here, and a great one. My master one class is now hooked on the subject - leadership - because we solved problems with AI, and they enjoyed it, and I saw a lot of "AHA" moments in class. The book is great, it is not just a good read, but also I feel that I live the science-fiction I once dreamed of.
Love such classification systems for an emerging practice 🙏 That's true trailblazing ✨
Another great article. I like the guide and clarity to navigate this new field.
To extend this into business, I show people four categories of how Gen AI can help them:
- supergoogle - context-senstivie info (I guess I would include simulation and tutoring in that)
- speed up - take over tasks and subtasks (email, outlines, etc)
- improve - challenge, critique and take on various personas to
- extend - using your knowledge and skills in a new area that you haven't worked in before
I find improve & extend is completely underrated and has real potential to give people superpowers.
I think this is VERY important work. I'm spreading it around. I think any organization that's using LLMs needs to have its own mini-culture around them. On the one hand there's just messing around and having fun. But that over time that exploration becomes focused and results in useful tools and strategies. This kind of thing needs to filter down into the middle schools.
Thanks as always, esp on the ask to share prompts and learnings. Tomorrow we start a prompt hackathon in the german speaking Corporate Learning Community and I definitely will try to translate results so that it’s available also tour wider target group.
Yep. Prompting for AI feels so much like the analog of learning HTML in the early days of the web.
Ethan, I finished your book last week. It's good! Everyone, go listen or read.
I have found myself agreeing with all of this, especially level 4. I think understanding that you can get what you want through a few corrective iterations is an incredibly helpful piece of the overall strategy, and one reason it works really well for me.
Just ask the AI to do something, see what it does, and ask to modify on that basis. This is one way to very quickly learn what an AI is capable of in the first place; I often try not to be overly specific with the first prompt.
Great stuff; please keep it up!
Hi Dr. Mollick-
I am sure you hear this all the time, but I'll say it anyway—you are my most important 'distant teacher' about Generative AI in the classroom. Thank you for staying honest, consistent and tuned in to what educators need to learn, try, and question.
Your post on April 22 addressed the necessity for folks in education to share out about their experiments with AI, and I could not agree more. At the university where I teach, we have a faculty study group that meets monthly to talk about how we are using AI—right now, it's a small but mighty core of instructors, staff and administrators and we are learning a lot.
With another set of teacher colleagues from K-12 classrooms, we have begun publishing essays about how we are using GenAI tools with pre-service and in-service teachers to support the quality of our teaching (and to show how GenAI creates efficiencies that free up time to do the parts of teaching that mean the most to us, but that often get swallowed up by necessary-but-time-consuming tasks.)
You can see our first couple of essays in the April 19 edition of Big Fresh (https://choiceliteracy.com/big-fresh/), a newsletter published by Choice Literacy. On the landing page, you can register for free or just scroll down and click on the header for the April 19 post.
Rock on!
Thx for this it is great stuff and very timely. I am an IO psychologist and I am currently working on a role play based assessment. We are tracking pretty nicely with your ideas, but there is some good stuff here that will help us elevate our game!
Hi Ethan,
Love the experimentation and insights you share. In support of the principle "Just like instructors often share lesson plans, or members of organizations might share tips for success, people can also share prompts. This is a useful starting point for would-be-innovators; someone else has done the work of creating and testing a prompt for you. By using other people’s prompts, you can see what an AI system can (or can’t) do." It would be helpful to have the prompts you use built in a publicly shareable location where people can iterate off of what you built as a foundation. Something like here: https://gooey.ai/compare-large-language-models/
In each of the posts on this SubsIn each of the posts on this Substack, the comments section are learning avenues as well.
This is a great article. I have been taking online courses on prompt engineering, but truthfully none of them come close to the clarity of this article. I have been building my own GPTs on OpenAI, and will definitely put your tips to use!
Thank you for this, I work with a lot of founders and this is helpful to my agency. Question, Do you think that the skill of prompt engineering (what this basically is) will be useful in the next 2 years? Like do you think it is only valuable now? because I am assuming in the next 2 years that we will have AI LLMs creating our prompts for us.
Great write-up, thanks for sharing!
Amazing post. I read your book which was quite insightgul in many respects. But you are doing God's work with the dissemination of this kind of post. I am an entrepreneur and see many parallels in educating employees on different topics