Showerthought: feeding the AI my messy handwritten notes worked better than typing them up
I keep a paper notebook for client meeting notes and for 6 months I was retyping everything into a doc so I could search it later. Last month I got lazy after a long day in Cleveland and just took phone photos of 14 pages of my bad handwriting and dropped them into a vision model instead of typing. I expected garbage because my scrawl is rough and I cross stuff out a lot. It pulled out the action items almost perfectly and even kept my margin doodles separate from the real notes. The one thing it tripped on was a client name I wrote in all caps with a slanted K, and it guessed wrong twice before I fixed it. Since then I skip the retyping step on busy weeks and just shoot the pages in good light. Has anyone found a way to make it handle crossed out lines without pulling the dead text in?
There's a weird pattern where the lazier version of a task ends up working better than the careful one. Same thing happens with grocery lists, where snapping a photo of the fridge shelf beats writing a neat list, or with recipes where a screenshot of a stained cookbook page beats copying it out. Somewhere along the way we decided clean input equals clean output, and it's just not always true. Messy stuff carries extra clues, like your margin doodles and the layout on the page. For the crossed out lines, my guess is you'd have to prompt it to skip any text with a line through it, since the model can't see the strike the same way you do. Maybe test it on one page you already know the answer to before trusting it on a whole week of notes.