I am delighted to share that I am a member of the Scientific Committee for the research project ∀ISION_E. The project's call for abstracts on "extended intelligences" in the field of drawing and architecture is open until November 14. www.visioneuid.com/call-for-vis...

Transformer LMs get pretty far by acting like ngram models, so why do they learn syntax? A new paper by sunnytqin.bsky.social, me, and @dmelis.bsky.social illuminates grammar learning in a whirlwind tour of generalization, grokking, training dynamics, memorization, and random variation. #mlsky #nlp

Join us in Charleston on November 4 for our preconference, "How Can Libraries and Publishers Collaborate to Make Backlist Monographs Open Access?, which is free to attend through the support of the California Digital Library, the De Gruyter eBound Foundation, and University of Michigan Library.

In principle, open access means that anyone, anywhere, can read and reuse scholarly work. In practice, many works labeled as “open” are constrained by restrictions that limit how they can be used. These constraints dilute the value of openness and conflict with its foundational definitions.

Computationally, whitespace gets little attention—it’s usually standardized or stripped. But in poetry, whitespace matters! Yet actually *preserving* that poetic whitespace is v tough. Its slipperiness points to bigger issues w/ text processing & LLMs. New paper ⬜️ aclanthology.org/2025.emnlp-m...

As DH grows, it’s increasingly important to publish conference papers, but there hasn’t been a clear venue for that. So I’m thrilled to share this new home for DH proceedings, which will include CHR papers & more. Thanks to @taylor-arnold.bsky.social for leading this effort! bit.ly/ach-anthology

New issue of my newsletter: "The Index and the Vector" — Converting ambiguity into precision can help a broader audience discover and learn from collections newsletter.dancohen.org/archive/the-...

New issue of my newsletter: “The Library’s New Entryway” — An interface that combines the advantages of the traditional index with the power of LLMs is the path forward newsletter.dancohen.org/archive/the-...

highly recommend!

Bartz v. Anthropic has had a couple of major developments. Though the lawsuit was initially brought to address the legality of using copyrighted materials for training AI, the suit now focuses on Anthropic’s storage—without training use—of copies of books downloaded from LibGen and PiLiMi.

Anthropic’s copyright settlement is historic, but it’s also not what many authors and publishers think. Check out our latest on what’s inside the proposed settlement:

I have updated my in-depth analysis of Bartz v Anthropic to reflect this important and overlooked aspect of the proposed settlement: “In what may be a rude surprise for authors, partial or full payments for many books may go to publishers rather than authors.” newsletter.dancohen.org/archive/land...

With @yh-huang.bsky.social, I'm excited to share our Digital Collections Explorer, an open-source, multimodal viewer for digital collections! Users can search with both natural language inputs and reverse image search. Paper: arxiv.org/abs/2507.00961 Public demo: digital-collections-explorer.com

New issue of my newsletter: “AI and Libraries, Archives, and Museums, Loosely Coupled"—A new framework provides a way for cultural heritage institutions to take advantage of the tech with fewer misgivings, and to serve students, scholars, and the public better newsletter.dancohen.org/archive/ai-a...

A new translation of @espejolento.bsky.social‬'s lesson! doi.org/10.46430/phe... We’re grateful to Javier Cisneros Brito + Alberto Santiago Martínez for their translation. Thank you to @betovargas.github.io‬ + Marisol Andrade Muñoz for their reviews, and to @giuliataurino.bsky.social for editing.

What does a "function" mean? What does it look like on a Wikimedia project? It might be something that checks leap years, tests for prime numbers, or decodes a cipher. These are small, clear examples that you can experiment with easily on Wikifunctions. 🧵⬇️ (1/3)

Everyone’s dropping VLM-based OCR models lately… But are they actually better than traditional OCR engines, which output XML for historical docs? I built OCR Time Machine to test it! 📄 Upload image + ALTO/PAGE XML ⚖️ Compare outputs side by side 🔗 huggingface.co/spaces/davan...

New this morning, a Comment I contributed to Nature Computational Science on the interaction between large language models and the humanities. 🧪 🤖 #MLSky rdcu.be/etk07 The link above will be open-access for a month — plus, I'll reply to this post with a link to a permanently open preprint. +

A new translation of @espejolento.bsky.social‬’s lesson: doi.org/10.46430/phe... We’re grateful to Javier Cisneros Brito + Alberto Santiago Martínez for their translation. Thank you to @betovargas.github.io + Marisol Andrade Muñoz for their reviews, and to @giuliataurino.bsky.social for editing.

Finally documented the Beyond Words dataset from the @librarycongress.bsky.social labs / @bcgl.bsky.social for the BigLAM @hf.co org! - 3.5K annotated historical newspaper pages - Bounding boxes + category labels - Photos, ads, headlines, cartoons & more

Can you train a performant language model using only openly licensed text? We are thrilled to announce the Common Pile v0.1, an 8TB dataset of openly licensed and public domain text. We train 7B models for 1T and 2T tokens and match the performance similar models like LLaMA 1 & 2

Many libraries now use automated tools to measure diversity in their collections. We examined how these tools work and whether library workers find them useful. A complex case study of libraries navigating automation, DEI, & shrinking public funding. Our new FAccT paper: arxiv.org/abs/2505.14890

I am excited to announce our latest work 🎉 "Cultural Evaluations of Vision-Language Models Have a Lot to Learn from Cultural Theory". We review recent works on culture in VLMs and argue for deeper grounding in cultural theory to enable more inclusive evaluations. Paper 🔗: arxiv.org/pdf/2505.22793

"Ardam was confronting a different, somewhat sensitive question about navigating a person’s digital history. When Sontag donated her laptop to the archive, did she realize how much she was giving away?" on archiving digital ephemera and culture -- www.theatlantic.com/culture/arch...

New preprint from @lauraknelson.bsky.social, @mattwilkens.bsky.social, and myself tests different ways of simulating the past with LLMs. We don't fully answer the title question here—just show that simple strategies based on prompting and fine-tuning are insufficient. +

“When Information is Networked” — My tribute to Clifford Lynch, who sadly passed away last week. Cliff saw before anyone else how digital technology would enable new forms of research and learning, and completely transform the production and dissemination of knowledge

Remapping the canon: a global history of photobooks by women photographers from the Getty Library. www.getty.edu/exhibitions/...

I recently co-authored a paper on Copyright, Privacy, and Public Access in News Archives (*). Thanks to @lisejaillant.bsky.social for including our work in AI & Society's Special Issue "When data turns into archives: making digital records more accessible with AI." link.springer.com/article/10.1...

"Photos are disappearing, one archive at a time." www.washingtonpost.com/opinions/int...

Library as home www.nytimes.com/2025/03/05/r...

What's in an anthology? @jdporter.bsky.social, Price Lab DH Specialist, & @fredner.org (with assistance from David McClure & the @stanfordlitlab.bsky.social) built a relational database of all 464 authors & 3,374 works published across the ten editions of the Norton Anthology of American Literature.

@marketplaceapm.bsky.social asked me to talk filing cabinets for their new series on short takes on overlooked tech www.marketplace.org/2024/12/09/f...

We just updated the AI for Humanists guide to model selection to include Llama 3.3, and a recommended best cost/capability tradeoff, llama 3.1 8B. What have you tried, and what would you suggest? aiforhumanists.com/guides/models/

Hiroshi Yoshimura - A•I•R (Air in Resort) archive.org/details/hiro... via @archive.org

“Can’t you just digitise it all!” we’re asked. Sadly not my friend, because resources. #Decisions are made based on likely user-interest & fragility of material. Here's our digital collections site with c.15,000 images: archiveandlibrary.rcsed.ac.uk/special-coll... #EYADecisions #ExploreYourArchive

New! "Infrastructure Futures for Digital Cultural Heritage". How can we best design infrastructures for digital preservation, access, analysis, & dissemination of heritage? Would speculative design help? Can we imagine other futures? With @jenr.bsky.social, Phil Sheail, & Cate Schofield. Do share!🙏

My group just finished up a new paper that I'm excited to get out into the world: "On Classification with Large Language Models in Cultural Analytics" (to be published at CHR): github.com/bamman-group...

Published: The book "Training the Archive" – with a number of great interviews and essays on the potential and dangers of AI in art and culture.