What if there are no "best practices" when using generative AI for research?
What if all use of genAI is inherently harmful?
What if I don't want to participate in research using generative AI?
These questions seem to be completely off the table now.
This feels like 2006 all over again, when everyone everywhere seemingly thought they had to use corporate social media for their professional activities.
I think it is possible to completely avoid the BigTech chatbots, with the stance you describe, yet at the same time not fall behind those who use then (incentive structures of the neo-liberal university...), and use LLM based tools for eg. Lit searches, first pass lit review, document summary, copy editing for those who need it, and helping broader research diffusion.
Using the same two letters to describe a whole bunch of different systems does not help with the 'should we? how?'
Me too!
I'm referring to the quick pass scan to see if an article is worth reading. Think exclusion criteria in a PRISMA.
Your mention of biais betrays a misunderstanding. When the sole human input in the context window is a one line prompt, the output will draw in the training set, with ensuing biais.
When you fill the input with human-written sources, then ask a question about what is in it isn't in the text is much less of a problem. Although of course, it's all stochastic, so still contingent on a dice roll.馃槓
I agreed both the searches and the reading are necessary, and those are not the steps I'm talking about.
Say I use keywords on reliable databases, and pull 100 papers, out of which 10 are actually relevant to me.
I can scan through abstracts, potentially paper themselves, or ask say NotebookLM to find which 10 are relevant.
Or if I have a doctoral thesis I know says something, having read it, but need to fine the specific paragraph.
A space for Bonfire maintainers and contributors to communicate
