My name is Ashley Orehek Rossi, and I am a STEM Librarian (at Western Kentucky University (WKU) (Bowling Green, KY) and an assistant professor in the University Libraries.
My tagline is meteorologist-turned-librarian because the traditional meteorology career paths didn’t resonate with me and I fell in love working in an academic library during college.
Naturally, I found myself on a science librarian track for graduate school at the University of Tennessee - Knoxville.
What does a STEM Librarian do?
My job encompasses a little of everything and tasks change daily - It isn’t all just about books! I teach one-time classes about library resources or various aspects of the research process; maintain and update the book collections; answer reference desk questions and offer research consultations; perform research either on my own or with colleagues; attend or present research at conferences; and help out with other things as needed.
How did you get into generative AI?
Not by choice 😅 I taught a physics junior seminar in February 2023 about literature reviews. I anticipated the question arising but chose to not indulge unless prompted – faculty had (and still have) individual opinions about AI and classwork use (overall, it’s mixed).
By the end of class, the question arose. I told the students I didn’t know much about it and needed to research before I could provide a clearer answer. Fortunately, I was scheduled to revisit the class in a few weeks so we tabled the discussion until then.
I played with ChatGPT, consulted Brandon and various sources, and conversed with colleagues also interested in this new hot topic of generative AI in academia. I incorporated what we knew then into a presentation about plagiarism and ChatGPT.
How did you teach about it then?
Before I left the first instruction, I asked for a physics topic I could inquire ChatGPT about. They chose spaghettification — an astrophysical phenomenon. Imagine holding a piece of chewing gum with your teeth and you pull it away from your mouth, stretching it thin. That’s what happens to a star when it gets too close to a blackhole – It gets stretched so thin until it is completely absorbed by the black hole.
Anyway, I prompted ChatGPT to explain it to an undefined population:
» Write two paragraphs about spaghettification in physics.
Then I modified the prompt slightly:
» Write two paragraphs about spaghettification in physics with citations.
I got the same exact response with seven citations embedded and a list of those citations.
I heard ChatGPT could possibly hallucinate information so I carefully vetted each citation. Of the seven citations, three were hallucinations of varying degrees!
I compared the hallucinations with what I thought were the citations it meant (if I could find one – One citation was straight-up fabricated!).
Essentially, ChatGPT mixed and matched real citation information from actual astrophysics journals and authors to make the citations.
For fun, I reentered the original prompt to describe spaghettification to a specific population:
Explain spaghettification to a 5th grade student.
Explain spaghettification to the average adult.
I displayed my findings to the students to be wary of using sources directly from ChatGPT, mainly to discourage using citations as is.
I repeated the same lesson to a Senior physics seminar in September 2023, and inputted the same prompts in ChatGPT for a then-and-now comparison:
» (then) Write two paragraphs about spaghettification in physics.
» (now) Write two paragraphs about spaghettification in physics with citations.
Interestingly, I got two different explanations and, instead of a citation list automatically provided, I had to prompt for one:
» Write two paragraphs about spaghettification in physics with citations and a list of those citations after the paragraphs in APA 7th edition style.
It gave me two citations this time. I vetted both - they were in fact real citations! Although, I cannot confirm if the information was properly used so I will need to consider that next time.
This time around, the responses seemed more humanlike, however a bit more lengthy and often repeating information it already stated.
Other AI uses?
Bard, Bing, and ChatGPT were useful at creating keyword lists so students could create search queries. I learned about this 24/7 rule WKU associate professor Thad Crews developed:
In between 50-100 words, explain [topic]. Call it the "24 second explanation." Then provide exactly 7 keywords [topic], no more no less.
Do you use any other AI tools than ChatGPT?
I played with DALL-E 3 via Microsoft Bing for image generation. My main reason was financial – I chose to not pay for ChatGPT+ and Midjourney (Brandon’s favorite) is a “freemium” platform.
DALL-E 3 via Bing is free to an extent – It appears you get 15 tokens per day (or week) before the generation speed drops. OpenAI just recently released DALL-E 3 with the ChatGPT+ subscription ($20/month).
Have you taught any more about generative AI?
After this newsletter drops, I will collaborate with library colleagues on a sandbox session. We will demonstrate a few common generative AI platforms to educate campus faculty and staff who may have reservations about using AI or are interested but didn’t know where to start.
Afterwards, we will provide a safe space to try the platforms themselves. The platforms we chose were ChatGPT, DALL-E 3 (via Bing), and Google Bard. We may host another similar session in the spring semester(?) - TBD.
I hesitate to teach about it widely because it really is instructor-dependent. Because WKU is still developing a campus-wide policy, they left it up to the individual instructor on how to incorporate AI, if the instructor chooses to. If I’m asked to, awesome! If not (and the syllabus states no generative AI), then I avoid it.
Will you incorporate AI into your everyday life?
It is situational-dependent. As I get more comfortable with it myself, I’ll probably start incorporating it. I’m mixed because I’m one of those people who wants to adopt it but appreciates the value of human intuition with technology more.
I do love using Siri and Google Home for simple tasks and daily recommendations and reminders.
What are some of the major takeaways from your journey so far?
There are different ways to prompt and the style of the prompt does affect the outcome.
Verify the outputs because we cannot assume it is factually correct.
Models update quickly and end up producing different outputs.
Check the citation content, not that it just exists.
Large language models (e.g., ChatGPT) are easily accessible and provide a great place to start learning how to use AI.
Keeping up with the industry is very difficult and quick.