AI for Higher ED - News, Analysis, Tutorials | Join 4k+ Profs and Learning Specialists who read my newsletter to improve their pedagogy and save time with AI.
As #AI becomes increasingly integrated into academic work, the question of how to properly cite it is crucial. In today's newsletter, I explore the purposes of citation and how they apply to AI. I also engage with MLA and APA recommendations, and I make some of my own.
“How to Use LLMs like ChatGPT as a Professor” = the title of Friday's AutomatED Zoom webinar, led by yours truly. More information here, including webinar schedule, registration link, discount options, etc: https://t.co/21jhXC0Vc1
Will #AI ultimately kill the prestige race in #HigherEducation, at least from the perspective of undergraduates? Likely so, or at least I argue as much in today's very-big-picture newsletter: https://t.co/OPk9kRzMKB
@NourHourani12 Thanks! I found it to be unreliable in doing so, but I didn’t explore at length, given the reliability of the converter. How lengthy are the custom GPT’s instructions? Is it almost always reliable?
Lecture recording (e.g. from prof mic only) or course readings or your notes or ... -> Gemini 1.5 Pro -> High-quality quizzes -> ChatGPT -> QTI converter -> Canvas LMS. Curious? It's all pre-paywall here: https://t.co/KXgxWNTpCy
After meetings, I use voice-to-text with LLMs. I dictate key points and outcomes, which the AI uses to draft follow-up emails or create reminders. It helps me capture insights while they're fresh in my mind.
I analyze my teaching using ChatGPT's Advanced Data Analysis on an anonymized gradebook. I look for correlations between assignments and use insights to brainstorm improvements. It helps me continuously refine my teaching strategies. More here: https://t.co/QOHs39QIm6
I use various LLMs, particularly Claude, to prepare for meetings. I input relevant docs for analysis and generate questions and talking points. It helps me think more comprehensively about meeting goals and outcomes.
As a professor, I use #LLMs like Claude 3.5 Sonnet and Gemini 1.5 Pro 5-20 hrs/wk. Here are 10 ways I use LLMs, including to lesson plan, draft grant apps, create quizzes, write better feedback quickly, prepare for (and follow up after) meetings, and tutor my students. A🧵...
For routine emails, I use Copilot for Microsoft 365. I provide a brief, essential response to inquiries, and Copilot expands it into a more formal email. It reduces time spent on repetitive correspondence. Old post on this here: https://t.co/vEePWNGry9
I draft grant applications with Gemini 1.5 Pro. I work section by section, providing detailed instructions. It helps me synthesize large amounts of info, including referee feedback. I can write grants in about 1/3 of the time now. More here: https://t.co/30VbKUSY8i
To grade faster, I use a custom GPT. I input brief notes, and it creates comprehensive, formatted feedback. I focus on content evaluation while the AI handles phrasing. It saves me 5-15 minutes per student per assignment. Plain ChatGPT version here: https://t.co/xT6eArAWYX
I use an @OpenAI#customGPT as a virtual TA in each class. It requires careful setup and maintenance. I focus on educating students about #AI literacy and potential fallibility. It offers 24/7 support, which my students appreciate. More on custom GPTs: https://t.co/SysqVPnwja
For course content creation, I use @Google's Gemini 1.5 Pro. I input lecture recordings or transcripts to create quizzes and summaries that I then send to Canvas. I've found it saves time and can be as accurate as I am with the right prompting. More here: https://t.co/KXgxWNTXs6
I use @AnthropicAI's Claude 3.5 Sonnet for lesson planning. I input course context and get creative activity ideas. It helps me keep classes engaging and adapt to schedule changes (e.g. after an interruption). I typically use a single chat session per course for continuity.
@emollick It couldn’t represent the actual chain of thought, except insofar as it may be a high level description of it, right? (Ie that is best case scenario.)
@emollick How do you square this confidence about clear improvement with the worry--that I believe you've expressed--that evaluating models > o1 will become harder and harder as they push into territories where only top experts can evaluate their effectiveness?
3 upshots of GPT-o1 for #Education:
1. Content-based AI immunity for reasoning-heavy assignments has diminished stepwise.
2. The need for student AI training with o1 is uncertain, despite @OpenAI claiming prompts should be kept simple.
3. Everyone needs a multi model toolkit.