There is a long list of large language model AIs. The names I recognize that my fellow researchers on Substack use are: GPT, Claude, Perplexity, and Gemini. I wonder if all four work equally as well. The article you linked mentioned Claude, but it also mentioned they were testing the API version of Gemini. I'll have to see what kind of results I'd get using the most current layman's version. The programs are improving so quickly, I can see I'm going to have to revisit them often.
I tried using OpenTranscribe. I was struggling, but managed to get to the pricing section. I'm not sure why it is charging to transcribe using Gemini 3 when I can transcribe directly with Gemini 3 for free. Perhaps because they format the transcription to appear by line. Have you used it?
Hi Deborah. No I haven't tried it yet. I had assumed it was covered by my Genealogy Assistant sub (which I use across Ancestry, MyHeritage etc.) but it appears perhaps not?. I usually use - https://transcribehistory.com/ - because it mostly performs well for me. I only go looking for something else if it struggles with a document which isn't very often with the sort of docs I am using it for.
I was inspired to test Transkribus after viewing the RootsTech presentation. I see I’m going to have to do it again with OpenTranscribe. Thanks for the heads up.
Hi, I tried transkribus a year ago for my great grandmother’s illegible scrawl and it failed miserably. But then so have most humans! I tried it on easier documents with pretty poor results. There is a process that it can learn but I did not want to embark on what occurred (possibly unfairly) as a degree course to train it. I might look at it again in the future more seriously.
My son told me ChatGPT (the free version) did better in my test with Transkribus because it was trained with a "large language model." Transkribus doesn't have the resources to train its AI in the same way. I wondered if the ability for me to train the AI would have helped, but from your experience, it sounds like I'd do better to stick with the big names.
Gemini is possibly better https://generativehistory.substack.com/p/gemini-3-solves-handwriting-recognition
There is a long list of large language model AIs. The names I recognize that my fellow researchers on Substack use are: GPT, Claude, Perplexity, and Gemini. I wonder if all four work equally as well. The article you linked mentioned Claude, but it also mentioned they were testing the API version of Gemini. I'll have to see what kind of results I'd get using the most current layman's version. The programs are improving so quickly, I can see I'm going to have to revisit them often.
Thanks Deborah. Have you checked this out - https://substack.com/@genealogyassistant/note/p-181232301?utm_source=notes-share-action&r=1n71wo
I tried using OpenTranscribe. I was struggling, but managed to get to the pricing section. I'm not sure why it is charging to transcribe using Gemini 3 when I can transcribe directly with Gemini 3 for free. Perhaps because they format the transcription to appear by line. Have you used it?
Hi Deborah. No I haven't tried it yet. I had assumed it was covered by my Genealogy Assistant sub (which I use across Ancestry, MyHeritage etc.) but it appears perhaps not?. I usually use - https://transcribehistory.com/ - because it mostly performs well for me. I only go looking for something else if it struggles with a document which isn't very often with the sort of docs I am using it for.
I tried transcribehistory. For English, it looks like the winner.
That great! It rarely lets me down. A few corrections needed sometimes but, on the whole, I find it really good.
I was inspired to test Transkribus after viewing the RootsTech presentation. I see I’m going to have to do it again with OpenTranscribe. Thanks for the heads up.
Hi, I tried transkribus a year ago for my great grandmother’s illegible scrawl and it failed miserably. But then so have most humans! I tried it on easier documents with pretty poor results. There is a process that it can learn but I did not want to embark on what occurred (possibly unfairly) as a degree course to train it. I might look at it again in the future more seriously.
My son told me ChatGPT (the free version) did better in my test with Transkribus because it was trained with a "large language model." Transkribus doesn't have the resources to train its AI in the same way. I wondered if the ability for me to train the AI would have helped, but from your experience, it sounds like I'd do better to stick with the big names.
These things are so unreliable that you have to know what’s written to check it! That surely defeats the purpose.