Writing on LaTeX, academic publishing, research integrity and AI-assisted writing.
What LaTeX and Markdown actually are, how each one works, side by side examples of the same document in both, what each is genuinely good for, and how a hybrid file lets you use both at once.
ACM tells you to put your AI disclosure in Acknowledgements. The CHI anonymised template hides Acknowledgements during review. Here is where the disclosure actually needs to go, and how to write one you can substantiate.
A dozen editors now compete with Overleaf. Here are the six capabilities that actually matter months into a manuscript, which tools have them, and what to watch out for in the newer arrivals.
arXiv has gone from asking authors to disclose AI use to banning them for a year over unchecked AI output, with fabricated references named explicitly. Three policy changes since 2023, and the one preventable mistake that carries the heaviest penalty.
Detectors flagged around 21% of ICLR 2026 reviews as machine written, authors started hiding instructions to AI reviewers in white text, and the NIH banned the practice outright. The reason for the ban is not that AI reviews badly.
Stanford researchers found seven AI detectors flagged 61% of TOEFL essays by non-native English speakers as machine written. OpenAI shut down its own detector for poor accuracy. Here is what the evidence actually says, and what an accused student can do instead of arguing with a percentage.
A growing number of graduate programs now require formal disclosure or committee authorization of AI use in a thesis or dissertation. Most of them put the advisor and the student in the same awkward position: agreeing on something neither can independently verify.
Every major publisher now asks authors to disclose AI use. They disagree, in specific and easy-to-miss ways, on where that disclosure goes, what it has to contain, and what counts as AI use in the first place.
The EU AI Act's transparency rules were written with deepfake images and video in mind, but the text-disclosure clause and the research exemption both reach further into how papers get written than most labs assume.
A computer scientist tried to copyright his AI's output by naming the machine as author, and lost at every court up to the Supreme Court. The same human-authorship test now decides how much of an AI-assisted book, textbook, or monograph you can actually defend as your own.
The three largest research funders in the US and EU adopted three different AI policies within about a year of each other. None of them can actually verify the thing they are asking applicants to attest to.
AI models invent citations to papers that do not exist, and they are ending up in published research. Here is how bad the problem has gotten, and how TeXposit stops fabricated references before they reach your bibliography.
Emailing manuscript_v4_final_REVISED.tex is still how a lot of research gets written. What real time editing fixes, and the four things it has to get right before anyone trusts it with a manuscript.
AI disclosure statements are unverifiable by construction. Recording authorship while the document is written replaces an attestation you have to trust with a record anyone can check.
Where TeXposit differs from Overleaf, where Overleaf is still the better choice, and the five questions worth asking before you switch anything.
Why we moved a LaTeX engine into the browser instead of making our servers faster, what it costs, and where it still falls short of a full TeX Live install.