Not sure having AI produce more research faster means better research. We already have enough issues with salami slicing, p hacking, paper mills. AI stands to exacerbate these issues more. It will likely benefit research in some capacity, but higher number of X metric is not beneficial
Exactly! Little to no discussion on p hacking, false discovery rates, salami slicing etc. More on producing more papers, more grant applications. How is that making academic research better? We already have thousands of poor quality articles, is millions better?
Confused why people argue that agentic AI is good for academia. It seems the arguments revolve around time savings and more papers produced quicker but ignores the impact this will have on peer review and an already overburdened system. Not to mention just more slop
#academicsky #metascience
A funding organisation aimed at more meta science ideas. They have had a few calls so far and a couple focused on AI, but importantly looking at AI's impact on science rather than build a new AI tool. I'm a fellow with them so ask any questions!
A great read by @aidybarnett.bsky.social and @jabyrnesci.bsky.social!
I also love funding opportunities aimed to develop applicants into independent researchers. Wasn't that the point of the PhD!?!? 😂
I think people in 30s/40s can be ECRs if they changed careers etc. But I do agree that its funny that 8 years experience (with PhD) counts as a junior in academia when there are seniors with less experience in other industries.
Many follow a similar pattern: trends and disparities in the title, similar methodology, limitations written in a similar manner. See more info here: www.medrxiv.org/content/10.6...
#science #openscience #metascience #academicsky
Our team recently highlighted the abuse of the CDC WONDER dataset, with a significant increase in the number of publications since 2023 (GenAI onset). In the first month of 2026, 276 new papers have been published, ~20% of the output in 2025 and ~79% of 2024 already.
Thoughts on averaging over 10 thousand citations a year in the last 10 years? Is this possible in healthcare research or a sign of unethical behaviour by inflating citations? 🤔
Curious to hear some perspectives!
#academicsky #metascience
Agreed worsening an already common problem, we discussed potential GenAI generated manuscripts below but this would just continue to fuel it
methods to p-hack a publishable finding or randomly splitting cohorts provides little, if any, benefit to the scientific literature
#metascience #academicsky #openscience
concerns that articles still need to meet peer review standards. It's obvious that peer review is struggling to be a gate keeper at this stage and more needs to be done to limit these articles.
There is still obviously a place for analyses of large datasets, but the mass repetition of
Very good read from @rossandersen.bsky.social and @theatlantic.com
It summarises nicely the downside to AI, and why the abuse of Open Access data is becoming a larger problem. It's also disheartening to see publishers state they will continue to publish some of these works, hand waving away
#academicsky #metascience #opencience @medrxivpreprint.bsky.social
CDC WONDER appears to be the latest open access dataset to become exploited by unethical actors, though is unique in authorship patterns and collaborations, continuing to highlight the tension between Open Science and research integrity.
We also see a large portion of these outputs are being submitted to conferences and published in journals such as Chest or Gastroenterology showcasing the link between the CDC WONDER exploitation and medical researchers as conferences have previously been a source of low-quality/unethical outputs.
Unethical actors can utilise the pressure of medical residents/trainees to publish to advertise courses with an almost guarantee to publish to at the end by using such templates. No single course is being mentioned or linked as it's hard to gauge which of these may be unethical or ethical.
medical residency and training programs. Recent articles (doi: 10.1001/jama.2025.23320; doi.org/10.1093/asj/...) have highlighted concerns about the research coming out from these programs as they appear focused on quantity of output rather than quality and integrity.
We also see strong connections between authors from Pakistan, India, United States, and the United Kingdom, including authors associated with Imperial College London and Duke University. The reason behind the connections is unknown, but we suspect it could possibly be due to
We can see a large, unprecedented increase in the number of articles being published, with most of them following similar patterns including the use of "trends" and "discrepancies" in the title, as well as eerily similar methodology and language in the write-up.
New preprint out now! This should be interesting particularly for those interested in paper-mills and their abuse of open-access datasets. This one stands to be particularly interesting as its one of the first to implicate actors associated with Western institutions.
Questions not just A is related to B without considering any confounders and no hypothesis really other than we are looking for significant relationships
Agree with Rich here, asking Chat to identify associations in a dataset would likely produce exactly the kind of low-quality publication we are seeing. Yes we need to continue to analyze existing data (always collecting data is too expensive), but we need to ensure people are asking legit and robust
How would you like to see it broadened?
I don't think having novelty in this context is bad, again unless they are incredibly strict about it. This checklist will probably just be ticked off anyway with some sentence of novelty regardless if there's anything there. But steps do need to be taken to manage this
Still p-hack and trick their way to significance regardless of novelty. I don't think focusing on a broad definition of novelty (don't repeat the same study just to sell a publication) automatically encourages p-hacking. That feels like a broader issue of academia
It is interesting that the default thought that novelty contributes to p-hacking. Because to me something can be novel but show no significant finding. If X has never been studied and then there's no relationship that's still novel. And as you can see from the crap research out there people will
They become very strict and reject everything unless it's ground breaking then that's an issue. But these open datasets are being sliced in so many ways mainly to serve the interests of paper mills or other unethical actors
Best add nothing. There is no reason in 2025 to dove a cohort into 2010-2015 and then have the same paper but the cohort is 2015-2020 unless there was some significant change but then that would be noted and likely considered novel (e.g., COVID). That's hopefully how they handle it but yes if
