SocArXiv

I’m given to understand the technology can and has been harnessed to useful effect. Ok. But in aggregate it is also increasing the relative volume of garbage that editors and reviewers must now contend with. With limited time and resources, that leads invariably to inefficiencies. Not good! (2/2)

As a journal editor, the most noticeable consequence of LLMs I’ve observed is an increase in slop submissions. While this might show up on someone’s ledger as greater productivity, what it has meant for me is more toil at the desk review stage and challenges with workflow. (1/2)

Now on @socarxiv.bsky.social !

osf.io/preprints/so...

New version of our preprint on bioRxiv about bioRxiv up. Now that’s what I call a revision – 6 years after the first version!
It has new data about our progress and highlights from a massive user survey. 1/n
www.biorxiv.org/content/10.1...

Yet.

Prof. Cohen is just venting. This is not an official policy statement.

Where should SocArXiv draw the AI line? A case that helps sort out the questions for an AI policy at SocArXiv.

It's one thing to ask authors to disclose how they used #AI. But it's another to have an articulate policy on what degrees and kinds of use make a work unacceptable, esp at a #preprint repository. Thoughtful take by @philipncohen.com, director of @socarxiv.bsky.social.
socopen.org/2026/02/22/w...

Nope.

It is definitely a risk factor for poor quality work of various kinds (one pattern we have observed is some folks who do this produce papers on very different topics very quickly...)

I'm not convinced by the case for here - GenAI can't think, and writing is part of the thought process / refinement of ideas. Outsourcing all the "thinking" to GenAI leaves a gap where the core intellectual endeavour must be.

This is actually my take on the @socarxiv.bsky.social question of whether to set policy banning fully AI-generated submissions - literally writing the prose is an important part of connecting with readers. As in faith-based fellowship, the act of being in community is an important part of Science.

"no one should be looking at the corpus of SocArXiv as a repository of the best research. (...) There's a lot of bad work on it, which, unlike most journals & some preprint servers, we are not shy about admitting, because it doesn’t hurt the good work that is here, & we’re not trying to make money"

Where Should #SocArXiv Draw the #AI Line? (via @socarxiv.bsky.social) socopen.org/2026/02/22/w... #scholcomm #preprints #publishing

Raises important questions: how will #scholcomm adapt, which norms around publishing will emerge? How will research assessment work in the future?
What do we want "research" to look like?
Curious to see where @socarxiv.bsky.social will end up with their policy.

The blog post and the thread suggest that It was a well written paper that -however- marginally contributed to the scientific discourse. If not for the author's (authors'?) own declaration, it wouldn't have been removed.

The degree of AI use by the author(s) is imho a bad reason for removal.

I'm the moderator who flagged the submission Philip is talking about here. The goal is a conversation about balance of "AI" and human work in papers. I think I may write a post about my own opinions, but we social scientists need to have a broader conversation about this. In particular, 1/2

So it passes the minimal quality bar. And apparently non-hallucinatory. The basis for rejection would be the disclosed AI use. But we don't have that policy (yet). Would should our policy be?
/4

If not for the AI, we would accept this. It's boring, unoriginal, and superficial. Its literature review is deficient. It has models, apparently done competently, and graphs and tables. Cites and in-text quotations appeared to be real. As a whole, it was coherent and relevant to existing research /3

A paper submission says they used AI tools used to generate code, literature search, consult on statistics, and draft text — but claims to have formulated the question and theory, chosen variables and models, interpreted results, and verified every data, citations & quotation, and shared the code /2

Where should SocArXiv draw the AI line? A case that helps sort out the questions for an AI policy at SocArXiv.

From @philipncohen.com, a case for discussing our policy on AI-generated research.

180 164 160 SocArXiv submissions: 30 days to 16 Feb 2026 As seen by Notebook LM 140 Gender & dentiry: 54 120 100 Family & Life Course: 46 106 Artificial Intelligence & Ethics: 24 80 80 60 Data Science & Research Methods: 40 67 Total Abstracts Count Social Inequality & Class: 40 Public Health & Epidemiology: 90 40 Governance & Democratic Institutions: 36 58 Climate Policy & Conservation: 36 20 Mental Health & Webbeing: 15 Urban Planning & Infrastructure: 22 0 Migration & Integration: 14 Social Networks & Community: 10 Sociology & Demography Communication & Digital Media: 42 Technology & Media Health & Medicine International Relations & Confilct: 18 Public Policy & Administration: 13 Political Science 44 Higher Ed, Careers & Management: 28 39 Philosophy & Theory: 23 Urban & Environmental Studies Broad Disciplines 47 Macroeconomics & Development: 21 Finance & Investment: 14 Behavioral & Microeconomics: 12 Economics & Finance Pedagogy & Instruction: 16 Education Archaeology & History: 14 Humanities 7 General/Misc: 7 Uncategorized A NotebookLM

Subject coding on @socarxiv.bsky.social is very messy - people choose any subject(s) and we don't police it. I had Notebook LM categorize the last ~500 titles+abstracts into 15-30 topics nested within disciplines. It made this. (Didn't check it beyond a few word frequencies, which look reasonable.)

Introducing COS’s 2026–2028 Strategic Plan: Advancing Lifecycle Open Science The Center for Open Science (COS) has released its 2026–2028 Strategic Plan, outlining a focused, three-year direction for advancing openness, integrity, and trustworthiness in research.

COS has released its 2026–2028 Strategic Plan, outlining a focused direction for advancing openness, integrity, and trustworthiness in research. This plan aligns our work around Lifecycle Open Science (LOS)—research with publicly accessible plans, contents, and outcomes.

💡:

SocArXiv New Submissions at Record Pace socopen.org/2026/02/17/s... #preprints #repositories #scholcomm @socarxiv.bsky.social

In the old days, our hosts (Center for Open Science) used to provide regular data on uploads, downloads, and views. Now that they don't, I've been meaning to come up with a regular report mechanism. I don't know about APIs and JSON, but I am semi-fluent in Stata, so I tried using ChatGPT...
/1

line graph showing cumulative socarxiv papers, starting in 2016, with an upward kink in 2025

SocArXiv accepted 3,162 new papers in 2025, an increase of 20% over 2024. As we are now accepting only social science papers, and turning away new papers on technical AI topics, this seems to reflect an increase in (human) social science. Thanks for sharing, social scientists!

Congratulations! The best paper-acceptance skeets are SocArXiv preprint paper acceptance skeets.

🚨My first preprint is out on @socarxiv.bsky.social!
How do AI-generated content labels shape what people see as authentic on social media — and do labels have unintended side effects? osf.io/preprints/so...

A thread 🧵

Very excited that @lauralindberg.bsky.social and I just posted a preprint of a paper that we've been working on for a while, looking at the (lack of) comparability of the most recent NSFG with prior waves of data. Feedback extremely welcome!
osf.io/preprints/so...

Surveys are important for miscarriage research, yet questionnaire design and sample characteristics can influence reported miscarriage prevalence

We compared miscarriage reporting across British surveys with @mccompans.bsky.social & @heinivaisanen.bsky.social - take a look at our pre-print here ⬇️

If they do we won't tell. We only share threads that reject the null, so to speak