Upgrade StorageNew Item Type for artificial intelligence
Hi All,
First post here, so please go easy on a newbie!
Are there any suggestions for creating an item type for Generative AI? The genie is out of the bottle, and students are going to need this functionality to cite AI as its use increases.
Any practical ideas would be sincerely appreciated! :-)
Have a great day
Andy
First post here, so please go easy on a newbie!
Are there any suggestions for creating an item type for Generative AI? The genie is out of the bottle, and students are going to need this functionality to cite AI as its use increases.
Any practical ideas would be sincerely appreciated! :-)
Have a great day
Andy
I think the norms on this (and what even counts as 'AI') are very much in flux and I don't think throwing an item type at it is necessarily the best way to go, certainly at this point (for now, I'd advise citing it as software as noted there, which it certainly is).
1. One wants to cite the tool itself: the Software type feels appropriate.
2. One wants to cite an output (full output, with prompts so that the reader can understand what happened): I guess Dataset or Web Page should work, as long as one obtains a stable snapshot of the experiment.
3. One wants to cite some specific, factual information obtained by a Generative AI: I don't think it should be done. Either Generative AIs become smart enough to cite the original, verifiable source (I assume that's how things could evolve, and then citing the primary source will be the way to go), or they will continue to produce unverified statements that should not be trusted for academic work.
Open for debate, of course :-)
It all seems up for grabs at the moment:
- APA recommends 'Personal Communication' - e.g.(OpenAI, personal communication, January 16, 2023).
- More at Cite-Them-Right: https://www.citethemrightonline.com/sourcetype?docid=b-9781350927964&tocid=b-9781350927964-217
Thanks for all your help, really appreciated!
Have a good one
Andy :-)
Provided we have a Forum post type, with the addition that when saving a question from Stackexchange we are even offered the option of saving the question or the answers individually, a very convenient feature, I think chatbot question/answers should be given the same treatment. I don't see much difference (in regards to item types) between a Stackoverflow answer by smartyduck_79 or a Bard one by GPT_4; but I wouldn't object to generative AI having its own item type.
I've been lurking in my cave for nearly a year, watching the gradual adoption of Generative AI in the Higher Education sector, and notice that several of the main referencing/citation styles are offering solutions for hand-entering GenAI:
> https://apastyle.apa.org/blog/how-to-cite-chatgpt
> https://libguides.ucd.ie/harvardstyle/harvardgenAI
Given that most students tend to use a reference management tool manage in-text citations and reference lists, would it now be time to think of a Zotero resource-type for GenAI.
Students don't want to be using reference management tools for the the rest of their work, then having to hand-enter Gen AI!
The Software resourse type could have been a solution, but it's end product doesn't comply with the emerging formats above.
Many thanks,
Clueless Andy ;-)
Luis
Assistant. (2024, November 28). Response to "How does Relation Extraction (RE) differ from Named Entity Recognition (NER)" [Personal communication]. Perplexity AI.
If you enter info as software as in the screenshot, Zotero/CSL will produce the APA reference as
OpenAI. (2023). ChatGPT (Version May 14) [Large Language Model]. https://chat.openai.com/chat
i.e with the exception of the order of day and version exactly as APA suggests in the blogpost linked above.
https://s3.amazonaws.com/zotero.org/images/forums/u2433/jg2dwrho3gygco66b6bj.png
As for citing individual queries, if you really must, you can cite it as personal communication (I think Interview in Zotero will work), but I think it's a misunderstanding of the role of citations. LLMs aren't people, so can't receive credit, and replies are stochastic, so reproducing queries doesn't make outputs verifiable (both of these things distinguishes them from Stack Overflow answers, which have authors and stable, verifiable links). The idea that you should include full queries also suggests/encourages poor LLM usage: As is by now well documented, good prompt engineering frequently requires longer prompts and refinement over multiple interactions (which are captured in the model's context window, so can't be reduced to the final query). If that is central to your work, you should document query strategies in the text or an appendix. It certainly doesn't make sense to have individual bibliography entries that span half a page.