"We measure AI-generated writing in the dissertations of a near-census of US STEM PhD graduates from 2019 to 2026 and link graduates to their subsequent employment. AI writing is absent before 2023, present in 29% of dissertations filed in 2026, and rapidly growing." (link below)

A paper recently came out claiming to show that nuclear power plants cause cancer. Using the exact same methodology, researchers at @TheBTI showed the following also "cause cancer". (link below)

.@arXiv is making some updates in light of a deluge of paper submissions: blog.arxiv.org/2026/10/01/up…

Interesting. "Robustness is better assessed with a few thoughtful models than with billions of regressions" pnas.org/doi/abs/10.1073/pna…

"... almost all participants had experience with irreproducibility... 58% had failed to replicate a result from the published literature.... Participants reported feelings of self-doubt, frustration, and depression while experiencing irreproducibility.. " journals.plos.org/plosone/ar…

Here's the duplication issue we found in "LDL receptor-related protein-1 is a sialic-acid-independent receptor for myelin-associated glycoprotein that functions in neurite outgrowth inhibition by MAG and CNS myelin" - It's awaiting moderation here: pubpeer.com/publications/E2C…

We found another duplication issue in the @J_Cell_Sci paper Sholto found issues in (10.1242/jcs.113191). Sholto suggests looking at PhD theses. In one co-author's PhD thesis we found same image but w different coloration and diff labels. Video shows transform & blink comparison.

We found a minor duplication issue in the supplementary info of 2017 Oncogene paper. What's interesting is this is from the lab of Dr. Kenneth Anderson at DFCI. @addictedtoigno1 previously flagged 16 of his papers previously as part of case that ended up causing quite a stir..

We're flagging a case of cross-paper image reuse, submitted to pubpeer. The 2014 paper caption says the image shows "EGFP expression in pAcISUBC transfected BuMECs" pAcISUBC is a human insulin expression vector constructed in the 2014 paper that does not appear in 2013 paper.

IMAGE DUPLICATION DETECTED Nature Medicine 18, 892–901 (2012). The cells in Fig. 1 are described as from a human patient, while the matching cells in Fig 6 are described as from a tumor grown subcutaneously in a mouse from human cells (a "mouse xenograph").

The paper names three mouse strains in the Methods, assigns Figure 1 to one and Figure 4 to the others, then asks the reader to "compare Fig. 4 with Fig. 1" precisely because the strains differ.

it's very curious! The two panels are different windows onto one micrograph, offset by 60 px and rotated 1.8 degrees

There is a duplication across figures that correspond to 2 different mouse strains: mice w & w/o crv4. Submitted to PubPeer. "Permissive role for mGlu1 metabotropic glutamate receptors in excitotoxic retinal degeneration" Neuroscience (2017) 10.1016/j.neuroscience.2017.09.005

There is a duplicated region across two different staining conditions. (Submitted to PubPeer) "Aberrant expression and constitutive activation of STAT3 in cervical carcinogenesis: implications in high-risk human papillomavirus infection" Molecular Cancer (2010)

Now Metascience Observatory Founder Dr. Dan Elton (@moreisdifferent) is carefully going through all the findings along w/ biology/neuroscience expert @GregSFitzgerald, PhD and we are posting to @PubPeer when we find things that we think deserve more scrutiny! (2/3)

We ran our image duplication detection system on 1,000 papers. The papers 1. have one author who has a retraction according to @RetractionWatch, 2. are not retracted, and 3. have no image-related @PubPeer comments. We use AI to screen findings & suppress false positives... (1/3)