Deyan Ginev

@dginev.bsky.social

411 Followers 436 Following

Software Engineer at arXiv. LaTeXML dev. The Web was created to share science. Let's keep it that way. creator of http://ar5iv.org

LaTeXML: Perl source vs Rust port, weekly statistics for lines of code in the repository, over 20 years, 2006 to 2026. LaTeXML perl shows gradual growth over the period reaching 74,432 lines. latexml-oxide, the Rust port, grows in 2 phases from 2015-2019, then 2022-2024, reaching 48,000 lines of code. It then exponentiates up to 240,506 lines in the months since March 2026.
arXiv's monthly submission growth followed by a giant beanstalk leading up into a cloud. AI generated image in the style of Studio Ghibli.
State of HTML 2025 survey, Content section, multiple choice question on inline MathML
Signs on fence: No man's land. Stop. "Attention. Prohibited area. Don't approach. Don't take photographs."
A giant cataclysm class robot labeled "Deep Learning" rampages through a city, with a structure on fire labeled "wall". The robot is holding a cookie labeled "arXiv" with a helicopter buzzing overhead.
The number 3 in square brackets is auto-recognized as a citation, but it is actually a mathematical notation for the permutations of 3, as the text explains.
The #1 DevOps excuse for legitimately slacking off: - Hey! Get back to Work! - Terraform! - Oh. Carry on.
A smiling banner at the top of arXiv.org exclaiming: "Happy Giving Tuesday - support arXiv today! Thank you to everyone who makes arXiv possible. Give to arXiv today and help keep science open. Donate!"
An old Indiana Jones looking skeptical.
An AI-generated image of a wall with a graffiti sign for "LaTeXML" with a regal Siamese cat sitting in front.
ChatGPT 4o successfully explaining why a math-based joke is funny.
ar5iv logo date: 04.2024 license: C-UDA
Recommendation 5: Encourage innovative approaches to integrating AI assistance into scientific workflows. The scientific enterprise is an excellent "sandbox" in which to practice, study, and assess new paradigms of collaboration between humans and AI assistants. Examples of such paradigms could include scientific advisory AI agents247 that would serve as sophisticated natural language interfaces for operating complex software or laboratory equipment; coupling generative AI algorithms with expert human feedback and formal verification methods (or with other types of AI systems) to improve accuracy of generative AI; or the use of AI tools to enable interdisciplinary, decentralized, or crowdsourced research projects at a scale that would not otherwise be feasible. The objective should not be to maximize the amount of automation, but rather to take advantage of complementarity and allow human researchers to do high quality science that utilizes AI assistance responsibly.
screenshot from an article abstract page of arXiv.org, showing a new "HTML (experimental)" link underneath the usual "Download PDF" link.
A typst sticker stating "LaTeX, it's not you - it's me" next to a poster with a red crab holding a flag of The Rust Programming Language and stating "Rewrite it in Rust"
A block of code from TeX's B book, assigning the pointer "last" the value of the pointer "first", then adding a comment of "see Matthew 19:30", referring to the Bible passage.