Sign in

MarineLives

@marinelives.bsky.social
2K followers 739 following 896 posts

Lives touched by the marine, 1574-1688. A volunteer led collaboration publishing to the Commons. See also: generativelives.substack.com

PostsRepliesMedia
MarineLives @marinelives.bsky.social · 29/04/2026
I have been working with a team of five historians to develop a governed research environment for historical research. We are calling our research environment The Historian’s Desktop and propose to launch it into the Commons in early September 2026. generativelives.substack.com/p/the-histor...
generativelives.substack.com
The Historian's Desktop
A governed research environment for historical research. Designed by historians for historians
010
MarineLives @marinelives.bsky.social · 06/03/2026
@adamchapman.bsky.social @justincolson.bsky.social @mpdavies.bsky.social @melissaterras.bsky.social @earlymodernjohn.bsky.social @ihrhistorylab.bsky.social @ihrbritainseminar.bsky.social
000
MarineLives @marinelives.bsky.social · 06/03/2026
I have been history vibe coding now for four months. I am offering to coach and support any of you who want to take the plunge with a first history vibe coding project. generativelives.substack.com/p/history-vi... @timhitchcock.bsky.social @tedunderwood.com
generativelives.substack.com
History vibe coding
An offer to coach any and all historians wanting to give it a try
000
MarineLives @marinelives.bsky.social · 27/02/2026
May be of interest @dslc.io @historytoday.com @royalhistsoc.org @historians.org
020
MarineLives @marinelives.bsky.social · 26/02/2026
May be of interest @tedunderwood.com @timhitchcock.bsky.social @ihr.bsky.social @jwbaker.bsky.social @adamcrymble.bsky.social
010
MarineLives @marinelives.bsky.social · 26/02/2026
Claude Cowork has huge potential for historians. The next session of the ai + history collaboratory (March 31st 2026) will be a hands on session exploring how to install Cowork on your desktop and to write and test skill files to assist your research generativelives.substack.com/p/cowork-for...
Cowork for Historians: skills and workflows
111
MarineLives @marinelives.bsky.social · 24/02/2026
See also github.com/ai-and-histo... and github.com/ai-and-histo...
github.com
010
MarineLives @marinelives.bsky.social · 21/02/2026
I’m sharing the product of five hours work with Claude Cowork spread over a day. I have been using Cowork now for ten days and this is my biggest project to date. The product is a static 5 annotation layer IIIF viewer for Ottoman Turkish. #ottomanturkish generativelives.substack.com/p/an-ottoman...
generativelives.substack.com
An Ottoman Turkish document viewer built with Claude Cowork
I’m sharing the product of five hours work with Claude Cowork spread over a day.
011
MarineLives @marinelives.bsky.social · 17/02/2026
Those were the days!! First public fruit of my work with Claude Cowork. An interactive dashboard of the MarineLives Twitter archive. 13 years of data. 10,535 Tweets. You can check it out on the ai + history collaboratory GitHub site: ai-and-history-collaboratory.github.io/social-media... #history
000
MarineLives @marinelives.bsky.social · 04/02/2026
We now have a logo @will-smiley.bsky.social @ottomanhistory.bsky.social @turkishstudies.bsky.social @andottoman.bsky.social @gtot.bsky.social github.com/ottoman-arch...
github.com
ottoman-archive
ottoman-archive has one repository available. Follow their code on GitHub.
030
MarineLives @marinelives.bsky.social · 04/02/2026
Launching the ottoman-archive GitHub organisation for all historians and linguists interested in learning about a world culture where only 10,000 people can now read its written records github.com/ottoman-arch... @ottomanhistory.bsky.social
github.com
ottoman-archive
ottoman-archive has one repository available. Follow their code on GitHub.
162
MarineLives @marinelives.bsky.social · 28/01/2026
I would also LOVE to find online digital copies of the Ottoman government gazette in Greek and Armenian. Indeed in any language. Maybe you could ask followers to pop into local archives and image a few pages. I believe they exist Takvîm-i Vekâyi Othōmanikos Mēnytōr ...
100
MarineLives @marinelives.bsky.social · 28/01/2026
Appreciated. Top priority at the moment is small amounts of ground truth (5 to 100 lines) paired with images of handwritten text from different periods and genres (bureaucratic; private correspondence; poetry...) and interaction with scholars who have use cases. We will turn stuff round quickly.
100
MarineLives @marinelives.bsky.social · 28/01/2026
I convene an ai + history online collaboratory : github.com/Addaci/ai-an...
github.com
GitHub - Addaci/ai-and-history-collaboratory: online forum with synchronous and asynchronous components for the discussion of the application of ai to historical research
online forum with synchronous and asynchronous components for the discussion of the application of ai to historical research - Addaci/ai-and-history-collaboratory
000
MarineLives @marinelives.bsky.social · 28/01/2026
I am a research oriented public historian. Ex-McKinsey, , Ex-GSK, an entrepreneur multiple times, working for the Commons. Working with a San Francisco SW engineer pro bono and potentially two Ottoman Turkish scholars (Paris; Bonn) + an Armenian musicologist [Prague] www.linkedin.com/in/colin-gre...
linkedin.com
110
MarineLives @marinelives.bsky.social · 28/01/2026
Hence the theme of my Substack "Opening the Ottoman Archive" for gneral non-Ottamanists as well as extending language scope of specialists (Ottoman Turkish + Albanian + Bulgarian + Greek + Armenian + Modern Turkish + [in future EgyptianmArabic] generativelives.substack.com/p/opening-th...
generativelives.substack.com
Opening the Ottoman Archive
You want to do granular research on the Ottoman Empire and its successor states, but don't read Ottoman Turkish. Do you throw up you hands or look for a new set of powerful tools?
100
MarineLives @marinelives.bsky.social · 28/01/2026
I think in under a year
210
MarineLives @marinelives.bsky.social · 28/01/2026
In the shortterm I aim to have a CCBY4 GitHub repository with co-created (machine/scholar) skill files which have been tested for a wide range of print & handwriting across genres + a readme. In medium term add an AI assistant. In longer term use MCP or connector t call skills files from LLM
000
MarineLives @marinelives.bsky.social · 28/01/2026
In the past I have specified no modernisation in the transliteration. I wasn't clear this time. Did it modernise? To be clear this is 2 stage. Claude Opus 4.5 is doing transliteration, translation, NER, summarisation at full semantic settings vs Gemini at "camera" type no thinking settings
100
MarineLives @marinelives.bsky.social · 28/01/2026
Where you could help me and my small team would be to generate some interest from say three or four scholars to submit handwritten text images of different genres and scripts C17th to C19th WITH groundtruth (even if just five or six lines) + virgin manuscript images. We will turn around quickly.
110
MarineLives @marinelives.bsky.social · 28/01/2026
Are you surprised at the combined Gemini/Claude performance? Do you find the Named Enity Extraction useful?
110
MarineLives @marinelives.bsky.social · 28/01/2026
Did Gemini get the first line wrong or not?
110
MarineLives @marinelives.bsky.social · 28/01/2026
If I had a digital persoarabic script rendition of image (not just Will's gold standard transliteration and translation) I could try one shotting it against the image, then giving Gemini for visual capture a new image in the same script. Possibly use it to change capture file
010
MarineLives @marinelives.bsky.social · 28/01/2026
Three comments before any feedback from @nabilal-tikriti.bsky.social This is only the second C17th handwritten Ottoman Turkish text I have tried and the first in this specific script. This is a zero-shot approach (no groundtruth from similar documents paired with images). It can only get better.
110
MarineLives @marinelives.bsky.social · 28/01/2026
It is exactly this input I need to improve the visual capture skills file (and also the semantic processing skills file adapted to the genre and period). Let me share the visual capture instruction file, which was developed for C19th print not C17th handwritten. Gemini is called in Google AI Studio
130
MarineLives @marinelives.bsky.social · 28/01/2026
@will-smiley.bsky.social @nabilal-tikriti.bsky.social I gave Will's transliteration and translation to Claude and asked for forensic analysis
110
MarineLives @marinelives.bsky.social · 28/01/2026
(3) Here are the sumamry instruction and files I gave Claude to do the semantic analysis and Claude's Socratic comments afterwards to accompany the structured JSON model output
010
MarineLives @marinelives.bsky.social · 28/01/2026
(2) Here is the semantic processing of the visual capture of the perso-arabic 5 lines (which no doubt could be improved). Important to understand I do NOT read Ottoman Turkish and am approaching this as a computational linguistics/HTR problem
210
MarineLives @marinelives.bsky.social · 28/01/2026
(1) In the spirit of complete honesty let me post the visual capture (using Gemini). It struggles with line 2. I will then post an extract from the skills file I have used for semantic processing using Claude Opus 4.5. Then series of pastes of the output (pt2). I can rerun the capture and improve
110
MarineLives @marinelives.bsky.social · 28/01/2026
Give me thirty minutes
100
MarineLives @marinelives.bsky.social · 28/01/2026
Can you put it back? I am interested in the manuscript image. I want to try to machine transcribe it for @nabilal-tikriti.bsky.social . Or Nabil DM me and I will give it a try
130
MarineLives @marinelives.bsky.social · 28/01/2026
The best online Ottoman Turkish (and some other Ottoman language) digital resources I have found with remote access in Turkey with a very searchable set of archival metadata is ISAM. For example: arsiv.isam.org.tr/index.php/hu...
ISAM: Fund HHP - Huseyin Hilmi Pasha DocumentsOttoman Turkish HTR screenshot
020
Reposted by MarineLives
Adam Chapman @adamchapman.bsky.social · 26/01/2026
Our new short courses are a great way to learn about using different kinds of sources for researching and writing history. Whether you're researching one place, or something broader, this is the ideal introduction to using the built environment in history! #Skystorians
0104
MarineLives @marinelives.bsky.social · 20/01/2026
Hi Zoe. You may be interested in the work we are doing in an ai + history collaboratory I am convening. You would be welcome to join a session as an observer. github.com/Addaci/ai-an...
github.com
GitHub - Addaci/ai-and-history-collaboratory: online forum with synchronous and asynchronous components for the discussion of the application of ai to historical research
online forum with synchronous and asynchronous components for the discussion of the application of ai to historical research - Addaci/ai-and-history-collaboratory
020
MarineLives @marinelives.bsky.social · 20/01/2026
Try using Claude in Project mode. You can save sizeable project files (PDFs, markdown files etc) into a self-organising RAG whih Claude will call in addition to all past conversations in that Project. See for example my Claude"Ottoman Turkish HTR" project.
000
MarineLives @marinelives.bsky.social · 20/01/2026
@christorical.bsky.social You might want to take a look at the monthly topic agenda of the ai + history collaboratory I convene. In December 2025 we built a Socratic Research Agent. github.com/Addaci/ai-an...
github.com
GitHub - Addaci/ai-and-history-collaboratory: online forum with synchronous and asynchronous components for the discussion of the application of ai to historical research
online forum with synchronous and asynchronous components for the discussion of the application of ai to historical research - Addaci/ai-and-history-collaboratory
000
MarineLives @marinelives.bsky.social · 10/01/2026
We are not there yet. But on C19th and early C20th movable type and lithographic Ottoman Turkish it is pretty good. Working on improving handwritten manuscript recognition
010
MarineLives @marinelives.bsky.social · 09/01/2026
This is how I work on Ottoman Turkish with Claude (and Gemini) to visually capture, transliterate, translate, extract named entitities, and contextualise C19th and early C20th Ottoman Turkish documents. generativelives.substack.com/p/opening-th... #Ottomanhistory @ottomanhistory.bsky.social
110
MarineLives @marinelives.bsky.social · 06/01/2026
(3) My personal (non-Ottoman Turkish expert) conclusion is, the Gemini/Claude work was a decent start, but signifcant adaption is needed for the mid-C17th script and some pre-processing is needed to deal with the slanted text. The source URL for the document is here memorients.com/articles/dis...
010
MarineLives @marinelives.bsky.social · 06/01/2026
(2) I had Claude then analyse the Named Eniities it had extracted from the text (with varying degreees of context) and compare them with external data to test how semantically sound the work felt
110
MarineLives @marinelives.bsky.social · 06/01/2026
Once you have taken a look at the article I can show you how my approach managed to analyse the following image and come up with the following summary. It is dodgy, because the document is mid-C17th and I am applying mid-C19th assumptions, & Gemini struggled with lineation but it can all be improved
110
MarineLives @marinelives.bsky.social · 06/01/2026
We could also try some well structured Ottoman naval manuscripts from C17th or C18th. I/we will need to adapt one or more HTR skills markdown files for period and genre. Intrigued how good results will be. Doable
110
MarineLives @marinelives.bsky.social · 06/01/2026
Many thanks Dr Sen. I would be delighted to send you further documentation, and if you have test printed (or well structured bureaucratic handwritten text) similar to the following documents we could test them together @pseudo-isidore.bsky.social @justagalandadog.bsky.social #Ottomanhistory
110
MarineLives @marinelives.bsky.social · 06/01/2026
May be of interest #Ottomanhistory @guel-sen.bsky.social @sariyannis.bsky.social @elifbengusuarik.bsky.social
151
MarineLives @marinelives.bsky.social · 05/01/2026
Professor Mark Humphries wrote an excellent review of English language processing by Gemini in a historical research setting: generativehistory.substack.com/p/gemini-3-s...
generativehistory.substack.com
Gemini 3 Solves Handwriting Recognition and it’s a Bitter Lesson
Testing shows that Gemini 3 has effectively solved handwriting on English texts, one of the oldest problems in AI, achieving expert human levels of performance.
020
MarineLives @marinelives.bsky.social · 05/01/2026
For my first take on Gemini (in context of processing Armenian script) see: generativelives.substack.com/p/a-new-lens...
generativelives.substack.com
A New Lens into the Archive
You are in an archive. You find a document in a language you don't understand. You take a photo, input it into Gemini 3 Pro. 60 seconds later you have a transcription, transliteration, and translation
120
MarineLives @marinelives.bsky.social · 05/01/2026
Google's latest model Gemini 3 Pro is particularly useful because of its excellent pixel level visual grounding. This enables visual capture of RTL complex scripts and subsequent semantic processing. Once you have the transliteration and translation from the accurate script then NER & LOD
110
MarineLives @marinelives.bsky.social · 05/01/2026
Interesting to see the framing in the link you sent me very much comparative and in the Balkan (and other Anatolian/Near Eastern) language context, which matches how I have been thinking about "Accessing the Ottoman Archive" linguistically and culturally
100
MarineLives @marinelives.bsky.social · 05/01/2026
Thank you. I will follow up with this suggestion. I am just at the start of outreach. I am not going to rush this for all the reasons you mention - just because I believe it works doesn't mean othrs do. Ther ehas to be personal discovery
000
MarineLives @marinelives.bsky.social · 05/01/2026
Had an expression of interest today from a Turkish scholar working in economics and econometrics interested in extracting Ottoman Turkish data from C14th handwritten Ottoman documents. A challenge. Waiting to see some images of documents. May not be the best to start with, but encouraging
130