10 Ways I Use LLMs like ChatGPT as a Professor

ChatGPT-4o, Gemini 1.5 Pro, Claude 3.5 Sonnet, custom GPTs - you name it, I use it. Here's how...

[image created with Dall-E 3 via ChatGPT Plus]

Welcome to AutomatED: the newsletter on how to teach better with tech.

Each week, I share what I have learned — and am learning — about AI and tech in the university classroom. What works, what doesn't, and why.

In this week’s piece, I share how I use large language models (LLMs) like ChatGPT as a professor, including which tools I prefer for each task, as well as why I think it’s a good idea.

Remember that pre-registration for my October 4th webinar on “How to Use LLMs like ChatGPT as a Professor" closes this week. Now’s the chance to register at a 50% discount for a 1.5 hour webinar on prompting and LLM-workflow integration.

💡 10 Ways I Use AI as a Prof

Given that this newsletter is, well, a manifestation of my attempt to ethically lean into AI integration in higher education, I am sure you are not surprised to hear that I use AI tools regularly.

But you may be surprised how much I use AI — 5-20 hours per week, depending on the complexity of the prompting required — or the varied ways that I use it.

With our rapidly approaching October 4th webinar on how to prompt and integrate large language models (LLMs) like ChatGPT as a professor, I figure now is as good of a time as ever to share how I improve my teaching and save time with AI.

By the way, pre-registration for the webinar is open now, with a 50%-off discount available for those who sign up this week (and a money-back-for-any-reason guarantee):

So, without further ado, below are 10 ways I use AI, with a focus on LLMs that I use as a teaching and researching professor. For each, I briefly explain what I use the LLM for, how I use it (including which tool), and why…

1. To plan lessons (especially activities)

How: I use Claude (namely, Claude 3.5 Sonnet) to plan lessons, providing it with course context such as previous and upcoming lessons, readings, expected student knowledge, and assessment goals. I typically use a single chat session for each course for continuity, occasionally utilizing Claude Projects for more complex planning.

Why: Claude's creativity enhances my lesson plans, making classes more engaging and fun for students. It's particularly useful for adapting activities when the course schedule changes.

2. To create course content (especially quizzes)

How: I use Gemini 1.5 Pro, available in Google AI Studio, to create course content, primarily quizzes. I can input lecture recordings or transcripts, often from a classroom microphone that records only my voice. I can even repeat student questions to ensure the context of my answers is captured. I also use AI to create summary handouts from these question-and-answer transcripts, as often this is where the action happens in class.

Why: Gemini 1.5 Pro's long context window allows it to process extensive inputs, making it ideal for extracting content from lots of information without losing track of what it means (more on the context window here). This method saves me a ton of time, and I’ve found Gemini 1.5 Pro to be just as accurate as I am, with the right prompting. The process can be automated for Canvas uploads, as detailed in my recent Premium tutorial, streamlining the content creation workflow.

3. To tutor my students

How: I use a custom GPT as a virtual teaching assistant in each class (more on custom GPTs here). This involves careful setup of the GPT and regular maintenance, including implementing content guardrails and designing it to take a tutoring stance. I also focus on educating students about AI literacy and the GPT's potential fallibility — misunderstandings about the latter can be a serious issue if left unaddressed (some students assume it is infallible).

Why: This approach offers 24/7 support to my students, enhancing their learning experience (I’ve never had human TAs, unfortunately — or fortunately). While it requires balancing the benefits of AI support with the need for critical thinking and AI literacy, it has been more than worth it and my students love it.

4. To grade faster and give better feedback

How: I employ a custom GPT, similar to the Educational Feedback Accelerator GPT available to ✨Premium subscribers, to enhance my grading process (plain ChatGPT version explained here). This tool transforms my brief jotted notes on each student’s work into comprehensive, professional feedback. I focus solely on content evaluation, while the GPT handles the conversion into polite, formatted (legible), and psychologically effective feedback. (No student data is shared with the AI.)

Why: This method significantly reduces the cognitive load associated with providing feedback. It allows me to concentrate on assessing content while ensuring students receive high-quality, constructive comments. The time savings are substantial, ranging from 5 to 15 minutes per student per assignment, which accumulates to a considerable amount across a larger classes.

5. To draft grant applications

How: I use Gemini 1.5 Pro for drafting grant applications, leveraging its extensive context window to synthesize large amounts of information, including referee feedback. I work section by section, providing detailed instructions, guidelines, and formatting requirements for each part. The prompting process is crucial and highly specific to ensure the LLM stays on track with the application's objectives and the complex field-specific content.

Why: This method streamlines the often complex and time-consuming process of grant application writing, while maintaining the necessary attention to detail and adherence to specific guidelines. I’ve found I can write grants in roughly 1/3 of the time, with improved quality and less soul death.

6. To email rote emails

How: I use Copilot for Microsoft 365 to handle routine emails efficiently (old post on this here). For basic inquiries about syllabus details or office hours, I provide Copilot with a brief, essential response. The AI then expands this into a more formal and polite email.

Why: This approach significantly reduces the time spent on repetitive email tasks. By allowing Copilot to handle the formalities and politeness of routine correspondence, I can focus on addressing the core content of inquiries quickly.

7. To prepare for meetings

How: I employ various LLMs, particularly Claude, to prepare for meetings. I input relevant documents for analysis and use the AI to generate questions and talking points. This process involves considering the meeting's context, goals, and potential benefits for all participants.

Why: Using LLMs for meeting preparation broadens my perspective on the meeting's purpose and potential outcomes. It helps me think more comprehensively about what I want to achieve and what other participants might gain. This is especially useful if I am short on time or energy.

8. To follow up after meetings

How: After meetings, I use voice-to-text in Apple Notes or directly speak into LLM apps like ChatGPT or Claude. I provide a stream-of-consciousness report of the meeting's key points, outcomes, and my next steps. This input is then used to draft follow-up emails, create personal reminders, or plan.

Why: This method allows for quick, efficient capture of meeting insights while they're fresh in my mind. Unlike automated transcription services like Zoom AI Companion, this approach focuses on my personal takeaways and perspectives.

9. To write letters of recommendation

How: I use a preset prompt with Claude to draft letters of recommendation. The prompt includes a framework for the letter and a section where I input a stream-of-consciousness account of the student's performance, attributes, and my personal assessment (I always record this stream-of-consciousness account with my Apple Notes app; I don’t use their real name). Claude then synthesizes this information into a structured, high-quality recommendation letter. I replace the dummy name with their real name once I paste it into Word.

Why: This method allows me to quickly convert my thoughts and observations about a student into a well-crafted, professional letter. It maintains the personalized nature of the recommendation while saving significant time on formatting and phrasing. The stream-of-consciousness input ensures that the letter captures my genuine impressions and specific details about the student, resulting in more authentic and impactful recommendations.

10. To analyze my teaching

How: I utilize ChatGPT's Advanced Data Analysis capabilities to evaluate an anonymized version of my gradebook, following a method similar to the Tutorial available to Premium subscribers. I input data from various assignments and assessments, focusing particularly on identifying correlations (or lack thereof) between formative and summative assignments. When ChatGPT identifies a lack of correlation, I use it to brainstorm adjustments to assignments or student preparation strategies.

Why: This data-driven approach allows me to critically evaluate the effectiveness of my teaching methods and assignments. By analyzing correlations between different types of assessments, I can identify which formative assignments are successfully preparing students for summative tasks and which ones might need improvement. This method helps me continuously refine my teaching strategies, ensuring that each component of the course contributes meaningfully to student learning outcomes. Additionally, using ChatGPT to plan adjustments based on these insights streamlines the process of improving my course design and delivery.

If you want to see some of my prompts, learn a range of prompting techniques, and better integrate your LLMs of choice in your workflow(s), be sure to sign up for the upcoming October 4th webinar on these topics. It’s going to be a good one.

How do you use LLMs in your teaching or research?

Login or Subscribe to participate in polls.

1. Researchers have developed PaperQA2 (code here; discussion here), an AI agent capable of conducting entire scientific literature reviews autonomously. In tests, it outperformed PhD and postdoc-level biology researchers on multiple literature research tasks. The creators claim this is the first AI to exceed human performance in this aspect of scientific research.

2. Harvard physics professors Gregory Kestin and Kelly Miller conducted a study showing that a custom-designed AI tutor doubled student engagement and learning outcomes compared to traditional active learning methods. The success of this experiment has inspired other large Harvard classes to test similar AI tutoring approaches this fall.

3. Fort Lewis philosopher Justin McBrayer argues that AI systems like ChatGPT and Gemini likely have built-in political biases, predominantly leaning liberal. He cites anecdotes and research showing these AIs often refuse to produce conservative content while readily generating liberal perspectives. McBrayer warns that as we increasingly rely on AI for work and personal tasks, we should be aware of these potential biases rather than assuming AI is politically neutral.

4. Researchers conducted a large-scale study comparing research ideas generated by an AI agent against those from over 100 natural-language-processing (NLP) experts. The study found that AI-generated ideas were judged as more novel than human expert ideas, though slightly less feasible, marking a significant milestone in AI's potential for scientific ideation.

5. You can upload files to GPT-o1 on Poe. (This is not native functionality on OpenAI’s site.)

6. Phil Hill analyzed Risepoint's Voice of the Online Learner survey, noting that while it shows a slight increase in preference for fully asynchronous online courses, a deeper look reveals 89% of prospective students are willing to participate in synchronous sessions, with most preferring weekly sessions.

7. Microsoft researchers have developed “WINDOWSAGENTARENA,” a comprehensive benchmark for evaluating AI agents' ability to perform diverse tasks in a Windows environment, alongside a new multi-modal agent called ‘Navi’ that demonstrates promising performance on these tasks.

8. Researchers created “RAID,” a large benchmark dataset for evaluating AI-generated text detectors, containing over 6 million text samples across 11 AI models, 8 domains, 11 adversarial attacks, and 4 decoding strategies. Using RAID to test 12 detectors (8 open-source, 4 commercial), they found current detectors struggle significantly with adversarial attacks, different sampling methods, repetition penalties, and unfamiliar generative models. The researchers have released the dataset and a leaderboard to encourage further work on improving detectors' robustness.

9. Philippa Hardman has analyzed how instructional designers are using AI tools across all phases of course development. Her survey found AI adoption has expanded beyond just content creation to include data analysis, design ideation, and evaluation. She notes a trend toward using specialized AI tools for specific tasks — that she lists and links — rather than relying solely on general-purpose AI models.

10. Researchers conducted a randomized, controlled Turing test with GPT-4, GPT-3.5, ELIZA, and human participants. In 5-minute conversations, GPT-4 was judged to be human 54% of the time, compared to 22% for ELIZA and 67% for actual humans. Interestingly, the most common reasons for judging GPT-4, GPT-3.5, and even humans as AI were forcing a persona, being overly informal, or lacking personality. Conversely, human-like or informal tone, plausible responses, and spelling/grammar errors were top reasons for human verdicts.

11. ICYMI: Google has gathered together the AI tools and features they provide to assist students, educators, and parents. These include research tools like NotebookLM (which is having a bit of a moment due to the latest updates, which let you listen to a conversation/podcast about your documents), a course on generative AI, and customizable AI experts called ‘Gems’.

12. Google has also introduced DataGemma, new open AI models designed to reduce hallucinations in large language models by grounding them in real-world statistical data from Google's Data Commons.

📬 From Our Partners:
A Newsletter That’s Just News

Daily News for Curious Minds

Be the smartest person in the room by reading 1440! Dive into 1440, where 4 million Americans find their daily, fact-based news fix. We navigate through 100+ sources to deliver a comprehensive roundup from every corner of the internet – politics, global events, business, and culture, all in a quick, 5-minute newsletter. It's completely free and devoid of bias or political influence, ensuring you get the facts straight. Subscribe to 1440 today.

⏳ Our Next Webinar:
“How to Use LLMs like ChatGPT as a Professor”

If you want to get better at using AI for…

  • lesson planning

  • creating course content (e.g. quizzes from lecture recordings),

  • grading and assessment

  • research tasks (e.g. drafting grant applications or reviewing literature)

  • administrative tasks

  • field-specific tasks

Then the next AutomatED webinar is perfect for you.

I will cover how to use large language models (LLMs) to complete a range of professorial tasks like those listed above, including effective ways to prompt, how to prompt long context, and how to best leverage retrieval augmented generation (RAG) with custom GPTs.

I will cover ChatGPT (including 4o, the new o1, and custom GPTs), Claude, Gemini 1.5 Pro, Copilot for Microsoft 365, and Gemini for Workspace.

And I will show you how to better integrate these LLMs in your workflow, so they fit neatly into how you operate on a day-to-day basis.

This is a rare window into a service I provide to professors in my one-on-one consultations.

The last AutomatED webinar, which occurred on September 6th and covered how to train your students to use AI, had 24 registrants from SUNY Fredonia to UNC-Chapel Hill and Butte-Glenn Community College.

100% of responding participants gave it an A (“Excellent!”) afterwards.

Here’s how one participant described it:

“Clearly presented, with acknowledgment that we are all learning as we go. Ample time for questions and, most significantly, thoughtful answers. Thanks!”

An attendee at my Sept. 6th webinar

Another simply said “more please.”

Wish granted!

The next AutomatED webinar — “How to Use LLMs as a Professor” — will occur on Friday, October 4th from 12pm to 1:30pm Eastern Daylight Time.

More information can be found on our website here, where I will add a detailed webinar schedule later today.

Pre-registration is open now, with a 50%-off discount available for those who sign up now and a money-back-for-any-reason guarantee:

Premium subscribers can pre-register with an additional 10% discount (-$15) here:

[Link visible to Premium subscribers only]

✨Upcoming and Recent Premium Posts

September - Tutorial on All Major Functionalities of Microsoft 365 Copilot

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.

Graham

Let's transform learning together.

If you would like to consult with me or have me present to your team, discussing options is the first step:

Feel free to connect on LinkedIN, too!