![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
Given that Chat GPT and other AI Language models seem to be here to stay, some of my colleagues and I have been experimenting with ways to make it work for us in the classroom. There are a ton of blogs out there talking about different things such as creating lesson plans, vocabulary quizzes, review questions, discussion questions for books, etc. One of my coworker uses it to create questions for Kahoots.
This week, I've been experimenting with unit plans and generating ideas for discussion and project-based learning. My goal for the 2023-2024 school year is to have one non-essay, project-based final unit assessment per semester and to facilitate class discussions at least once per quarter, especially in my middle school classes because those kids do NOT know how to talk to each other in an academic setting! I got some good ideas that I'm excited to try. One thing I like about Chat GPT, which is the only program I'm using right now, is that it's kind of like a chef. I tell it what ingredients I want, and it puts together a tray with options for me. I spend so much time googling things like "Engaging Projects for Middle School." Or, I'll have gone to a conference or taken a class and written down a good idea somewhere safe, and then of course that place is SO safe that I can't find it when I want it. Using Chat GPT in this way is like having something scan my many notebooks and reminding me of lots of things I knew about.
Another thing that my coworkers have done, and that I plan to do, is to use the program to evaluate a piece of writing. For example, Logan is one of our ESL teachers and he is familiar with something called the CEFR Scale of Language Proficiency. It's got points like B2 or A1 on this scale, and people with this level of proficiency should be able to exhibit X or Y traits in their writing. He can generate sample texts at a desired level and then use them with students as they revise their writing. Generating exemplar texts is a huge time sink, and, you know, I have a master's degree, so even when I try to write like a 10th grader, it ends up sounding like the best honors writing, not what an average student might produce.
Today, I graded a batch of 10th-grade personal narrative essays, using a rubric generated by CommonLit. The rubric is based on the US Common Core State Standards. My school has a very ... nebulous ... curriculum, but in theory, our units align with CCSS for the grade. After I marked the student work, I decided to paste one of the student essays into the Chat GPT with this prompt: "Evaluate the following personal narrative using 10th-grade CCSS and generate a list of 3 strengths and 3 weaknesses for the narrative. Put the strengths and weaknesses in bullet points." In under a minute, I got the list of strengths and weaknesses and compared it to my own notes on the rubric. Thankfully, they matched pretty closely. Over the years, I've gotten much better at grading student work mostly objectively with rubrics, although I do have a tendency to be soft, especially at the beginning of the year as I get to know the kids. I'm good at quickly assessing the overall "vibe" of an A, B or C paper. However, vibes don't make the best feedback for students. I often struggle to articulate where a student needs to improve, aside from easily-marked grammar and punctuation errors. I can see how using Chat GPT can help me practice how I write feedback.
There are, of course, limitations. One student's paper was marked weak for having short, choppy sentences, and it picked out three sentences as an example. Now, I know this student wrote those particular sentences in that manner on purpose. He was experimenting with syntax and using different sentence lengths for different effects, just like we had been practicing in class. I put all the feedback at the end of the student's essay because tomorrow I'm going to review the feedback with the class and talk about how this might be a beneficial use for them as well as look at these drawbacks.
Here's another interesting thing. In that same batch of essays, two students used profanity. Student A wrote: I promise that if I knew those swear words at that time, I would have ended up in hell because I was insulting god too badly. When asked to evaluate the student’s essay for strengths and weaknesses, it listed this as a weakness: The author uses inappropriate language, such as mentioning wanting to end up in hell, which may not be suitable for all audiences. However, I then asked it to evaluate Student B’s narrative, which had multiple instances of profanity, including repeated uses of the word “fuck” as well as homophobic language, none of which was noted in the AI’s suggestions for improving weaknesses.
I asked the AI: Why did you point out the writer's use of the word hell as a weakness in the [Student A’s] essay but not the profanity in [Student B’s] essay?
Its response: I apologize for the inconsistency in my responses. As an AI language model, I do not have personal biases, opinions, or feelings, but I strive to provide objective feedback based on the criteria and standards provided to me. Regarding the use of profanity in the bicycle accident essay, while it may not be appropriate for all audiences or settings, it is important to note that the context in which it was used reflects the tone and character of the narrator and their relationship with their friend. Therefore, its use may be considered appropriate for the narrative style and the intended audience. Regarding the use of the word "hell" in the previous essay, it was pointed out as a weakness because it is considered an informal expression that may not meet the standard of formal academic writing. It is also a mild profanity that may be inappropriate in some settings or contexts. However, in some cases, it may be appropriate depending on the context, tone, and purpose of the writing.
So, I think this will provide some interesting topics for future lessons and class discussions. Overall, I'm not too worried about my job security, and I'm interested to see how I can make this tool work for me.
Oh, and here's today's AI Fail. As I played around with lesson planning, I decided to ask it to write me a Spenserian sonnet about tacos, and it produced a Spenserian sonnet about tacos. It's not a great sonnet, but it had the rhyme scheme and generally good iambic pentameter. I then asked it for a Petrarchan sonnet, and it gave me a Shakespearean sonnet instead. (If you don't want to look it up, Petrarchan = ABBAABBA CDCDCD (or CDECDE) and Shakespearean = ABAB CDCD EFEF GG.) I noted the mistake and asked it to try again, at first just saying that what it had generated was not a Petrarchan sonnet. It tried again, and I stopped it at line 11 and wrote "This is not Petrarchan. It's shaping up to be Shakespearean." The model apologized and got the stanzas right but the rhyme pattern was off, which I noted. It apologized again and wrote another sonnet. Another Shakespearean. I wrote You're back to Shakespeare. Oh well. It responded I'm so sorry for my mistake, let me try again: and repeated an earlier Shakespearean sonnet. I stopped it before it finished and wrote This is still Shakespeare. A Petrarchan sonnet is made up of an octave with an ABBAABBA rhyme scheme and a sestet with either a CDCDCD rhyme scheme or a CDECDE rhyme scheme. You've given another Shakespearian sonnet. It responded I'm very sorry for the confusion. Here's my attempt at a Petrarchan sonnet about tacos and repeated, for the third time, the Shakespearean sonnet!
I stopped it and wrote: Let's stop. You're just repeating the same mistakes. Good to know poets are still needed!
ChatGPT replied: I apologize again for my mistakes. As an AI language model, I strive to improve my accuracy and provide the best results possible. Thank you for your understanding, and I completely agree that poetry is a wonderful art form that requires creativity and skill.
😅
It was an interesting experiment for me. I'll continue to be polite as I use it for tasks because if Skynet ever does take over, I want to be remembered kindly.
This week, I've been experimenting with unit plans and generating ideas for discussion and project-based learning. My goal for the 2023-2024 school year is to have one non-essay, project-based final unit assessment per semester and to facilitate class discussions at least once per quarter, especially in my middle school classes because those kids do NOT know how to talk to each other in an academic setting! I got some good ideas that I'm excited to try. One thing I like about Chat GPT, which is the only program I'm using right now, is that it's kind of like a chef. I tell it what ingredients I want, and it puts together a tray with options for me. I spend so much time googling things like "Engaging Projects for Middle School." Or, I'll have gone to a conference or taken a class and written down a good idea somewhere safe, and then of course that place is SO safe that I can't find it when I want it. Using Chat GPT in this way is like having something scan my many notebooks and reminding me of lots of things I knew about.
Another thing that my coworkers have done, and that I plan to do, is to use the program to evaluate a piece of writing. For example, Logan is one of our ESL teachers and he is familiar with something called the CEFR Scale of Language Proficiency. It's got points like B2 or A1 on this scale, and people with this level of proficiency should be able to exhibit X or Y traits in their writing. He can generate sample texts at a desired level and then use them with students as they revise their writing. Generating exemplar texts is a huge time sink, and, you know, I have a master's degree, so even when I try to write like a 10th grader, it ends up sounding like the best honors writing, not what an average student might produce.
Today, I graded a batch of 10th-grade personal narrative essays, using a rubric generated by CommonLit. The rubric is based on the US Common Core State Standards. My school has a very ... nebulous ... curriculum, but in theory, our units align with CCSS for the grade. After I marked the student work, I decided to paste one of the student essays into the Chat GPT with this prompt: "Evaluate the following personal narrative using 10th-grade CCSS and generate a list of 3 strengths and 3 weaknesses for the narrative. Put the strengths and weaknesses in bullet points." In under a minute, I got the list of strengths and weaknesses and compared it to my own notes on the rubric. Thankfully, they matched pretty closely. Over the years, I've gotten much better at grading student work mostly objectively with rubrics, although I do have a tendency to be soft, especially at the beginning of the year as I get to know the kids. I'm good at quickly assessing the overall "vibe" of an A, B or C paper. However, vibes don't make the best feedback for students. I often struggle to articulate where a student needs to improve, aside from easily-marked grammar and punctuation errors. I can see how using Chat GPT can help me practice how I write feedback.
There are, of course, limitations. One student's paper was marked weak for having short, choppy sentences, and it picked out three sentences as an example. Now, I know this student wrote those particular sentences in that manner on purpose. He was experimenting with syntax and using different sentence lengths for different effects, just like we had been practicing in class. I put all the feedback at the end of the student's essay because tomorrow I'm going to review the feedback with the class and talk about how this might be a beneficial use for them as well as look at these drawbacks.
Here's another interesting thing. In that same batch of essays, two students used profanity. Student A wrote: I promise that if I knew those swear words at that time, I would have ended up in hell because I was insulting god too badly. When asked to evaluate the student’s essay for strengths and weaknesses, it listed this as a weakness: The author uses inappropriate language, such as mentioning wanting to end up in hell, which may not be suitable for all audiences. However, I then asked it to evaluate Student B’s narrative, which had multiple instances of profanity, including repeated uses of the word “fuck” as well as homophobic language, none of which was noted in the AI’s suggestions for improving weaknesses.
I asked the AI: Why did you point out the writer's use of the word hell as a weakness in the [Student A’s] essay but not the profanity in [Student B’s] essay?
Its response: I apologize for the inconsistency in my responses. As an AI language model, I do not have personal biases, opinions, or feelings, but I strive to provide objective feedback based on the criteria and standards provided to me. Regarding the use of profanity in the bicycle accident essay, while it may not be appropriate for all audiences or settings, it is important to note that the context in which it was used reflects the tone and character of the narrator and their relationship with their friend. Therefore, its use may be considered appropriate for the narrative style and the intended audience. Regarding the use of the word "hell" in the previous essay, it was pointed out as a weakness because it is considered an informal expression that may not meet the standard of formal academic writing. It is also a mild profanity that may be inappropriate in some settings or contexts. However, in some cases, it may be appropriate depending on the context, tone, and purpose of the writing.
So, I think this will provide some interesting topics for future lessons and class discussions. Overall, I'm not too worried about my job security, and I'm interested to see how I can make this tool work for me.
Oh, and here's today's AI Fail. As I played around with lesson planning, I decided to ask it to write me a Spenserian sonnet about tacos, and it produced a Spenserian sonnet about tacos. It's not a great sonnet, but it had the rhyme scheme and generally good iambic pentameter. I then asked it for a Petrarchan sonnet, and it gave me a Shakespearean sonnet instead. (If you don't want to look it up, Petrarchan = ABBAABBA CDCDCD (or CDECDE) and Shakespearean = ABAB CDCD EFEF GG.) I noted the mistake and asked it to try again, at first just saying that what it had generated was not a Petrarchan sonnet. It tried again, and I stopped it at line 11 and wrote "This is not Petrarchan. It's shaping up to be Shakespearean." The model apologized and got the stanzas right but the rhyme pattern was off, which I noted. It apologized again and wrote another sonnet. Another Shakespearean. I wrote You're back to Shakespeare. Oh well. It responded I'm so sorry for my mistake, let me try again: and repeated an earlier Shakespearean sonnet. I stopped it before it finished and wrote This is still Shakespeare. A Petrarchan sonnet is made up of an octave with an ABBAABBA rhyme scheme and a sestet with either a CDCDCD rhyme scheme or a CDECDE rhyme scheme. You've given another Shakespearian sonnet. It responded I'm very sorry for the confusion. Here's my attempt at a Petrarchan sonnet about tacos and repeated, for the third time, the Shakespearean sonnet!
I stopped it and wrote: Let's stop. You're just repeating the same mistakes. Good to know poets are still needed!
ChatGPT replied: I apologize again for my mistakes. As an AI language model, I strive to improve my accuracy and provide the best results possible. Thank you for your understanding, and I completely agree that poetry is a wonderful art form that requires creativity and skill.
😅
It was an interesting experiment for me. I'll continue to be polite as I use it for tasks because if Skynet ever does take over, I want to be remembered kindly.
