Hi everyone,
I wanted to ask a quick question about using AI in research. Lately, I’ve noticed more people around me are starting to use AI tools to support their academic work.
So I decided to try out a few myself like tlooto and Consensus. They seem helpful, but to be honest, I’m still not fully confident about trusting AI for research yet.
What do you think? Has anyone here used these tools? I’d love to hear your thoughts or experiences.
I haven’t used those tools, but in a world that’s ever more concerned with productivity and where I have to have four ongoing projects to be taken seriously, sadly using an LLM to organise data and proofread writing has become necessary.
Right now, my biggest worry is stunting my own skills.
About two years ago, NSF started to fund and frame AI-assisted research as “co-science” and I agree that this is one of the most promising areas of development for AI tooling. My advice for finding a co-science tool for your work is to look at a lib guide, like this one from Oklahoma State. The thing about lib guides is that they are prepared by information scientists, so you will get excellent guidance and extremely well curated resources.
A bit unrelated, but what do you think about this new kind of scientific journal?
Excerpt from their introduction:
“We’re more than just a journal—we’re a multidisciplinary Open Science platform designed to meet modern academic needs by combining human expertise with advanced AI.”
Quite a few people at the Ronin Institute publish preprints there (e.g. @martin.bohle ). They have a clear interface for users and readers and they also generate DOIs for each preprint, which is very helpful. I recommend it, even though I have never published my work on there.
Qeios is a bit different as far as I understand: it also provides a system that allows a sort of public process of peer review. And in case of success, the paper is automatically upgraded from pre-print to scientific article!
Yes, Qeios does also have a sophisticated peer review system, which makes it even better!
Hi Keith,
thank you, your comment helped me to make a decision (in agreement with my two co-authors)! Our last paper is now on Qeios as a pre-print.
BTW, you at IGDORE are welcome to read it, although you are probably not allowed to write a review, since you have the same affiliation as me, which is considered a conflict of interest. However, since IGDORE is widely distributed, you may be allowed to do so, or maybe you can do it by using a different affiliation… I don’t know exactly.
Yes, I am also a fan of Qeios although I have not published there myself.
Regarding AI - I find that it is very useful for development activites (i.e. designing based on existing knowledge) and can support a lot of research activities in various ways. I’ve also been quite surprised with how well it develops theories and connects ideas in the adjacent possible, but you have to assess its responses very critically in these cases.
That said, I have found GPTs tend to follow orthodoxy and don’t like contrarian ideas - when I ask o3 about my research on RF antivirals I consistently defy its predictions that what I am doing won’t work, even when I give it the results from previous stages of my work to use as a reference. So my impression is that even reasoning models tend to prefer producing ‘answers’ that already exist in their training dataset rather than working out answers out from first principles (even when you tell them to do just that).
For using prompt based AI (e.g. ChatGPT), I’ve found using Ethan Mollick’s ideas are very useful in improving the quality of my response. I recommend both his substack and book
This is a very useful guide for using the current generation of reasoning models: Using AI Right Now: A Quick Guide - by Ethan Mollick