Commentary on AI use in peer review, out now at LSA:
"We argue for professionalization and transparency rather than prohibition, and propose guidelines built on three aspects: model selection and data protection, human guidance and accountability, and publication integrity."
Human reviewers pull 8.4% of their criticisms from a paper's appendix.
Frontier LLM reviewers: 62%.
A new study had GPT-6 and Claude 5.5 models review ICLR 2027 submissions, then mapped each weakness to the section it came from. The AI critiques keep going past page 30 of the appendix.
They skew toward detail audits: more seeds, significance tests, confidence intervals. Rarely the core idea.
Same pattern via API, web chat or coding agent, which points at the models, not the interface.
arxiv.org/abs/2611.17811#PeerReview#LLMs#AcademicTwitter
Had a really wonderful time at @UrologySBUR in Memphis this past weekend! Great science, great people, great bbq and blues, and a great @BassProShops pyramid! And was really honored with the Young Investigator Award I was lucky enough to share with my girl, @AyeshaShafi_PhD!
I love getting immersed in a tight and supportive community! SBUR combines excellent research with strong cohesion and mutual support.
It was an honor to discuss manuscript prep and the publication landscape w the trainees and faculty!! @UrologySBUR@LSAjournal
> be me, reading a new paper about directed evolution in human B cells
> B cells naturally mutate antibody genes; called somatic hypermutation
> seems like a really good idea to "borrow" that machinery to evolve other types of proteins, too
> researchers put a gene into the B cells, add a short stretch of DNA upstream (to recruit the antibody-making proteins), and let the cells generate a bunch of mutations
> periodically sort the cells via flow cytometry to keep the best ones
> use this to make flu-antibody variants with 3x better binding
> no need to build and keep cloning a mutant library between every round
> start thinking about selecting for several useful properties at once, directly in human cells
> yeah, this seems really useful
> but wait
> feel like I’ve heard this idea before
> find a 2004 paper by Roger Tsien
> they did the exact same thing, using B cells to evolve new fluorescent proteins, 22 years ago
> (this is where mPlum and mRaspberry came from)
> why have I never heard of this?
> suddenly realize most things in biology actually came from somewhere else and most ideas are much older than people think
> Francis Crick was calling for control over neurons in Scientific American in 1979
> by 1999 he was explicitly saying we should use light as the signal
> optogenetics is 'invented' a few years later
> people say future laboratories are going to be fully automated
> Zymark said the same thing in the 1980s, and it never happened
> ideas in biology repeat themselves
I’m recruiting 1–2 PhD students for Fall 2027 in the Deep Lab at UC Berkeley!
In the Deep Lab, we study how ocean microbes build living structures and interact with fluid flows, spanning cellular self-organization to ocean carbon transport.
This commentary urges professionalization and transparency for AI tool use in peer review, proposing disclosure standard and guidelines that preserve human judgment and efficiency gains.
@theNCIhubs.la/Q04F4tD70
Great point. Journals who already publish reviews could add an AI disclosure there.
Transparency about AI use in peer review not only for the authors but also for the reader.
Agreed. Editorial offices: LSA already publishes review files (this one's is up), so reviewer disclosure could sit there as fields: tool, task, which human checked it. The draft COPE, STM and International Science Council standard asks authors the same. datamercs.net/posts/2026-10-…
Chi-Ping Day and colleagues observe that banning AI use by peer reviewers is not enforceable and leads to undisclosed use. They propose a mix of transparency and strict guidelines grounded in human accountability.
Read the piece here: life-science-alliance.org/co…
"These findings indicate that AI ... can predict high-resolution banding from genome sequences alone, thereby bridging classical cytogenetics and modern AI-based genomics."
This is so cool! Out now at @LSAjournal
Today I'm posting a new video to my YouTube channel! PubMed has stopped indexing Scientific Reports (for the time being at least)... Is anyone surprised? youtu.be/HXcrOkRUkLY
One of the most common dynamics I discuss with authors:
A reviewer might behave poorly or have questionable motives, but still be objectively correct about a critique.
It puts the editor in a difficult position: uphold scientific values without validating poor behavior.
Moreover, it is entirely possible for two things to be true at the same time:
1) The reviewer is farming citations in bad faith.
2) The reference in question is important to situate the manuscript relative to established literature.