Today is Pi Day! Find your math-nerdy friend* and have them recite the 100 digits of pi they've memorized. When they're done, casually mention that NASA only uses about 15 digits in their calculations and you only need 37 digits to accurately measure the circumference of the visible universe.

I’m still amazed that one can upload a few hundred manuscript pages to Google’s AI Studio and get a somewhat decent overview of what’s inside. I often request it to scan volumes that might not have anything of interest for Brazil references and it does a reasonable (not perfect!) job in seconds.

Super excited to see this news is public. This grant is very important for the future of Canadian history research.

I've been working with my son for past few months to develop an LLM that corrects kid's dyslexic writing without providing an "easy button". After trial and error, we now have a model that matches the performance of Claude Opus 4.6 on a test story. The first two images show the story and corrections

We need to change this conversation and the underlying publish or parish culture it reflects. We don’t need more papers. Could we instead aim for better papers? Claude is very powerful, but what is the point of pumping out 50 paper that the ‘author’ half comprehends?

“No one is going anywhere. Canada will survive the brief moment of honesty.”

Last minute prepping for a workshop on converting HGIS data to Linked Open Data with the LINCS team tomorrow. Here is a visualization of my draft dataset linking Canadian census subdivision between 1851 and 1921 jburnford.github.io/Canada-Histo...

Finishing this 2019 book by Melanie Mitchell. It is mind boggling how much has changed in 7 or 8 years. “Human level language processing remains a distant goal.” We’re not there yet, but we’ve moved a lot faster than she predicted.

4 days ago Idlout endorsed Lewis. Today she crossed to the Liberals. I don’t understand politicians.

I've added in the Princely States, so I think the knowledge graph is now complete. The visualization only shows 25 of the more significant Princely States. You can find the code to create your own version of the database on my GitHub (linked in the thread).

Listening to Youth: Historicising & Challenging Parental Rights Discourse activehistory.ca/blog/2026/03...

My student @historyjacob.bsky.social has shown the new OCR tools work very well with early modern English print text. We use to struggle with the long s and many other issues. This opens up NER on sources that use to be too messy. He also developed a fine tuned NER model for Atlantic commodities.

How are historians rethinking environmental and social history via improved OCR of imperial archives? Join us this Wednesday 3pm UK time to hear from @jimclifford.bsky.social and @historyjacob.bsky.social - registration link below.

I've been working on a foundation knowledge graph of the British Empire off and on for a few months. I think I have a relatively complete and clean draft done and visualized. I'm also sharing the Cypher so you can take and build on this work. jburnford.github.io/text_as_data...

This is true, if “the good life” means “drinking heavily and networking with other sons of the gentry, followed perhaps by an easy job in the Church,” and “the humanities” mean “Latin and theology.”

The idea that America has lost its post-1945 wars by being too ‘soft’ is deeply embedded in a certain constituency of US minds. During Vietnam the US dropped an equivalent of 100 Hiroshima bombs on SE Asia. Softness wasn’t the issue.

Maybe this speaks to my social and social media networks, but the NDP leadership race isn’t breaking through. I’ve had zero conversations about the race or who should win. I left the party years ago, but I’m still a little surprised how little I’m hearing compared to the last leadership race.

Join the Lancaster-Manchester Environmental #DH Seminar on March 11 @ 3pm UK (online) for a talk by @jimclifford.bsky.social & @historyjacob.bsky.social: "Solving OCR: Using olmOCR to Follow Commodities across the British World" www.eventbrite.co.uk/e/solving-oc... #dhist #ocr #envhist 🗃️

I'm working on a text mining/knowledge graph project using the Indian Office Lists and Colonial Office Lists are the source material. I've extracted the biographies of Indian Office officials from the 1937 list into a knowledge graph visualized here: jburnford.github.io/indianoffice...

My biannual reminder, if we don’t agree to peer review requests, the whole system collapses. And, faculty in at research universities with reasonable teaching loads, need to carry most of the weight. If you or your students want to publish, you need to say yes, even when you are busy.

This feels like a pitch for a Dorchester Review article.

This problem also shows up when the answer isn’t known. This is where academics and students need to be very careful. Gemini will write a full essay to answer your question whether it finds new evidence pointing to the truth or comes up empty and instead piece a compelling but wrong answer.

This is an important measure of LLMs. The underlying engineering makes them want to answer all questions. But what if the question is nonsense? The examples are extreme, but you could seem subtle misunderstanding in questions steering a discussion off the rails. github.com/petergpt/bul...

New working paper out: "Open Maps: New Research Directions and Workflows for Digitized Historical Cartographic Material" led by Vincent Baptist & Jules Schoonman (TU Delft) #dh #maps openmapsmeeting.nl/publications...

Making progress in my efforts to parse all of the articles from the first 8 editions of the Encyclopaedia Britannica. 120,039,792 words; 114,359 articles; 133 volumes. jburnford.github.io/early_encycl...

There a big problems with all of these companies, but OpenAI is competing hard to rank as the worst AI lab. www.nytimes.com/2026/02/27/t...

I remember when Canada couldn’t afford to act because of China…

The long-term concern here, to be clear, is not just about researchers, but about enclosure of the knowledge commons. Schools are paying tech companies for tools that were trained on their own books—while tying their own hands to prevent internal development of similar tools. It's self-destructive.

We're in a strange situation rn where Google can train freely on books from university libraries—but researchers *at* universities have limited access. I'm optimistic this can be fixed, but if you're in admin or working at a foundation, please know: univs are failing here & resources are needed.

For ex., one great “Critical AI” course in lit. would be “Can AI Write a Suspenseful Story?” Students work together to answer the question. Likely “no,” but they need to get there rigorously. They’ll learn to use AI (through an API!), and also learn skepticism that lasts because it was earned. +

I think AI having mostly (not entirely) very bad critics is a real problem because it means we’ll get political action focused on things that probably don’t matter that much in deferring it’s very real harms.

Ezra is very good here. He calls bullshit and really pushes the Anthropic employee. TLDR: the political system is broken and it is hard to imagine good public policy to confront AI safety or job market disruption.

Claude Code has improved a lot in the past few months. I think Ezra's framing is correct here. We need to start thinking about public policy for major disruptions in the job market. And universities are not ready for the disruptions coming out way. youtu.be/lIJelwO8yHQ?...

As a centrist…

Just starting Beckert’s new book

Google is mad at Chinese AI labs for stealing their intellectual property to build their Gen AI. (Completely different from Google stealing all human knowledge to train AI systems.) youtu.be/W9WB2xbM5sc?...

LAC was already limping along on life support for years and now this.

Just spent two hours talking w/ 30 (likely left-leaning?) doctoral students about the opportunities and perils of AI. Marx was quoted; the phrase “zero-shot” was used; “stochastic parrot” was not. If this complex reality isn’t visible in thinkpieces / social media, we need to make it visible.+

Possibly motivated by recent mini-drama here … Melbourne vs. Minneapolis: a superficial comparison Better in Minneapolis: Lakes, river, what you can get at Target, fall colors, snow, house prices Better in Melbourne: Coffee, ocean proximity, sporting events, weather, transit, road safety, democracy

My son and I are working on a dyslexic writing app that uses a small language model trained to correct dyslexic spelling but doesn’t have an easy button. It has speech to text for hard words with dictionary look ups to confirm he finds right word. And text to speech to hear the writing.

Why is my draft suddenly 2000 words over the limit?

These days most of my LLM tokens are spent on laptop, with open-weight models. My situation is atypical: I have a lot of simple textual summary-like tasks that don’t need SOTA. But I’m sharing just to say: SOTA can eventually overshoot need for some (small but growing) fraction of tasks. +

Rereading this today. It is a really important article. I’ve been learning to fine tune and distill open weight models for historical research.

I’ve reached the abandoning MS Word for VS Code with Quarto Markdown stage in my digital humanities career. Claude Code is an exceptional copy editor for dyslexic academics (and follows instructions to only copy edit without rewriting).

My MA student Jacob Polay released a new NER model focused on Early Modern English/Atlantic History. He is getting better results than paid LLMs and much better than spaCy NER, and the model can run on a 4GB VRAM laptop. Jacob started coding with Claude in September. github.com/polayj/early...

What if you could create a semantic search engine using clean ocr for your key primary source collection in a few hours without writing any code? And then for fun you could build and knowledge graph and combine them Into a GraphRAG the next day (a few month after learning all that jargon).

among everything else this week dunno if you heard but coding is over. huge opp for humanities folks. should be megaphoning 'learning to think never goes out of style'

This is absolutely how we should pitch coding agents to humanists. Strictly speaking, what mattered about code was never arbitrary syntax, but clearly specifying data structures and processes. And coding agents still force/teach you to do that part. But … shhh! Let’s get ppl on board.

Playing with a global climate dataset created using ship logs. I wonder if we can use it to study global trade and extractivism? jburnford.github.io/CLIWOC_work/...

I am accepting applications for a postdoc to work with me at the University of Toronto Munk School of Global Affairs and Public Policy. Area of Research: Democratic Institutions, Public Engagement, American History Due by February 23rd munkschool.utoronto.ca/current-oppo...