Image Generated by DALL·E 2
Given my last post, it would be easy to assume I despise ChatGPT and think it is worthless. Au contraire, I am just a hater of over hype. I do suggest you go and read that post if you haven’t yet. It was my attempt to clear up the hype and doom-and-gloom-talk about ChatGPT.
Now, it turns out I am really excited about LLMs and ChatGPT, but for different reasons than most people are talking about.
I see ChatGPT as a tool that can bring a lot of value to writing and teaching.
I AM NOT SAYING CHATGPT IS A GREAT REPLACEMENT FOR TEACHERS AND THEY SHOULD FEAR FOR THEIR JOBS!
You might be saying “Wait Steve, I thought you said ChatGPT is often wrong, how can it be good for teaching and writing?”
Glad you asked, let me lay it out for you.
Writing has been a very difficult task for me. I know writing is hard for most people, but this is different. I get no joy in the finished product, and the process can actually feel painful to me.
Some of this is my ADHD, where writing is too slow and too exact,so I get frustrated. But a large part of it for me is the mechanics of making sentences and paragraphs. Even if I had good ideas, I would constantly get bad grades on my writing. I loved taking French and would have probably stuck with it much longer had I not been required to write in French as part of the class.
With ChatGPT, I feel like I have a calculator for writing. And just like we use calculators for math, I feel like we should do the same with ChatGPT for writing.
While everyone should learn how to do math without a calculator, once you understand the concepts, you’re allowed to use a calculator, especially if math is difficult for you. For some reason, not being able to write everything from scratch is viewed differently from finding math difficult.
With ChatGPT that no longer needs to be the case. The model has done a good job of figuring out grammar, constructing sentences, arranging those sentences into paragraphs, and using those paragraphs to form a coherent argument. The content may have errors,but the structure is still good and that is the hardest part for me (and others).
I used this process to begin a blog post just a few weeks ago. I went from weeks of dragging my feet and avoiding making a blog post, to having a good product that I felt happy putting my name on in just 2 days. I still had to rewrite major sections of the text, as well as add other sections, but I didn’t have to start from scratch and that made all the difference.
The way I see it helping is that a user gives ChatGPT a prompt for the topic they want turned into the written product. Once ChatGPT produces its response, the user still has to go back and edit the text along with fact checking for mistakes. The user becomes an editor, which, while still demanding, is not the same difficulty as writing de novo text.
Of course, people (mostly children) still need to learn the fundamentals of writing, and how to construct a written argument. But they won’t be required to do it for the rest of their lives without the help of LLMs.
There are so many subjects I studied in school that would have been so much more satisfying and interesting had I not been forced to write all the text from scratch. If I could have focused more on my ideas, History, Literature, French, Psychology… would have been more intriguing areas of study. A good chunk of my course selection revolved around avoiding classes that required “too much writing.”
I almost didn’t get my PhD,and I definitely left academia because I had such a hard time with writing. I thought there was something wrong with me,and I just shouldn’t do it. I had won NSF grants and a Fulbright, but still, my difficulties contributed to giving up following my passion, research and teaching, for a career which involved far less writing.
Perhaps you are thinking that surely ChatGPT can’t help with writing scientific papers. I had the same doubts as well. Then I thought, what the hell, let’s give it a try. As you can see, ChatGPT starts out with a good student essay “about the importance of animal movement in understanding landscape ecology and conservation biology”. But then I prompt it to make the text more technical. The result is actually a good start for the introduction of a scientific paper. It needs fleshing out and citations, but this would have been really helpful for getting me going with writing the journal article. It also provided me with a reasonable and current list of sources for further reading.
There is now a “calculator” for those of us that struggle with writing. And in the same way the calculator has removed some of the stigma and difficulties for people who struggle with math, I am hoping we can see the same change for those of us who struggle with writing.
There is now a tool available that can help people like me get past our fear/struggle with the mechanics of getting words on the page and instead, let us focus on communicating our ideas.
Like other technological advances, ChatGPT as a writing tool will likely need to go through the same process of social acceptance as calculators did, from being banned in classrooms to being encouraged as a teaching aid.
There are already good examples of faculty who are digging in and creating innovative learning.
I also had a transformative experience working with ChatGPT to teach me how to use Spring Data REST (see the blog post comment above). I believe that for those who like interactive learning, using ChatGPT is the best online experience out there.
The problem for me with reading blog posts, watching YouTube videos, or even online pre-recorded classes is that, if you don’t relate to the way the material is being taught, you are out of luck. If the example doesn’t make sense or you don’t quite get what they are talking about, you have no ability to ask questions. There is no ability to dig deeper if you don’t understand one particular area.
I did my google searches looking for an introduction to Spring Data REST with PostgreSQL. You are interested in PostgreSQL, right? Because if you are not using PostgreSQL you need to re-evaluate your life choices. I found plenty of examples and tutorials online in both blog posts and videos.
All the posts I found assumed:
- You were quite familiar with the concept of a Repository in Spring. I could write a whole aside about why I am not familiar with the concept, but that is not for this post.
- Just showing how it worked with 1 table in the database was enough to get you to understand the whole thing.
Since neither of these were true for me, I was struggling to get up to speed. I turned to ChatGPT with this question:
Can you please show me a complete example of using Spring Data Rest with Spring Boot and PostgreSQL
ChatGPT’s response began with instructions on how to create a PostgreSQL database, then how to create a Spring Boot Application, and finally how to wire them together.. While this may seem ChatGPT went too basic, it actually worked for me to quickly skim over the flow I already understood and anchor myself in the example.
Like the other resources, it only used 1 table. And here is where things got interesting. I then asked:
Can you show that with more than 1 table
It proceeded to add another table THAT MADE SENSE GIVEN THE FIRST TABLE! It didn’t just make up some random table – it actually kept them conceptually linked.
After I looked at that discussion and example, I thought I was getting stuck because I didn’t understand one of the Spring Annotations being used. Again, the great part for me is I could, right in this teaching session in this browser tab, dig deeper on that annotation:
Please explain the @RepositoryRestResource annotation arguments
After reading its explanation, I was then able to figure out my real problem. I was having a problem visualizing what the collectionResourceRel parameter of the annotation would look like for multiple tables.
What would the collectionResourceRel look like if I had 4 tables in my database
And with that answer, a basic understanding finally clicked into place for me. Now I could ask a more advanced question:
How do I produce a custom query in the example above
Which gave a really good example and I was ready to start digging in and experimenting with the code.
ChatGPT allowed me to create a custom curriculum on the fly given my knowledge and problem areas. I didn’t have to keep bringing up Firefox tabs trying to stitch information together from a bunch of different sources. I could get examples relevant to the requirements of the application I was going to build. It prevented hours and hours of frustration and got me to a solid foundation where I could make forward progress.
From now on, when I am struggling understanding the basic (or intermediate) concepts for a new technology, I will probably turn to ChatGPT to give me a foundation. The inability to ask questions and get real time consolidated answers, limits the usefulness and pleasure I get from learning from blog posts, articles, and pre-recorded classes.
I could easily see students using this technology to help them understand concepts being taught in class. Given our mass production style of teaching, if you don’t really get a concept given the way your teacher presents it, you can fall behind or struggle to learn the concept. With LLMs, a student can ask for clarification on topics their teacher might not have time to cover o, or ask a question they didn’t feel comfortable asking in class.
As I mentioned in my first post, ChatGPT gets information wrong -not just the text but also the code samples. I could not run the examples as provided by ChatGPT; they gave compile errors.
It also didn’t necessarily teach me the “proper” way to use the technology. With many programming tasks there are many ways to solve a problem, some of which are better or worse given your use case. ChatGPT had its way of showing,but I don’t think it was the best in my application. It let me get my head around the concepts, but I still had to do the work to translate those concepts into better implementation.
While my first post was mostly pouring cold water on the ChatGPT excitement, I am still excited for the possibilities created by this technology. It can provide a tool making it easier for students to both learn new ideas and then share their thoughts on those areas. It can help people who might be great in a given profession pursue that area despite having difficulties producing topical, written communication.
It can remove some persistent stigmas that we have about how people learn and who they are as a person. I am hoping these can help people who are afraid of “asking stupid questions” and who struggle to recognize they are good at a subject simply because they have trouble expressing themselves through writing.
My next post on ChatGPT will be some of my thoughts on the implications of ChatGPT and Dall-E-2 (AI image generator) on the future of our current “open web”.
Would love to hear your thoughts on what exciting outcomes we can see from LLMs. Drop a comment here or reach out to me on Mastodon.