By Shawn Graham
Continuing from the last experiment and ruminating on Drew Breunig’s taxonomy of use cases for AI (in short: gods, interns, cogs, and toys) as well as ‘homecooked software‘ I undertook to use Anthrophic’s ‘Claude’ model to see what I could do. I’m building a cog via Claude’s ‘artifacts’ feature which allows you to iterate on small code snippets, html, etc – Simon Willison does this a lot so blame him.
Anyway, the result is a web-app that I called ‘handwriter’ – which is a terrible name – but permits someone to pass images and multi-page pdfs through the Gemini vision model and get transcribed text back. It handles the prompting and image manipulation.
You can find it at https://github.com/shawngraham/handwriter
Azure’s transcription, as reported in Jeff Blackadar’s Programming Historian lesson:
DECEMBRE
28 VENDREDI. Ss Innocents
362-3
clear and cold, - lovely out.
Visit from mme Thomas D
five daughters from
PontarleĆr.
your Doctor, Major merletti
arrived- good fellow
Not orders to go with Capt
Marrison Vit Road and 200
men to another part of
France
prote Inier
Sittley wip mess account
Cash so francs you the mouth
Gemini’s transcription, via my little app:
DECEMBRE
28 VENDREDI. Ss Innocents 362-3
Clear and cold - lovely out.
Visit from Mme Thomas &
five daughters from
Pontarlier.
New Doctor, Major Merritt
arrived - good fellow.
Got orders to go with Capt.
Morrison (Lt-Hod) and 200
men to southern part of
France.
Wrote Dyer my account -
Settled up mess a/c for the month
Cost 30 francs for the month