Retry transcription button?

[Logan_Buffa]

Fairly often, the model gets a little confused and starts spitting out repeated messages or lines of empty text. I’m not fully sure why–initially I thought I was overloading Leo by putting in a new document to transcribe before Leo finished the previous one. I’m less sure of that now, but I think that it’s still true for SOME cases. Here’s a few examples.

I think that a good way to remedy this is by adding a feature that allows us to request the AI to try again. This would solve issues associated with credit waste while also keeping operations smooth by not requiring y’all to restore missed credits, etc.

[Jon]

Thanks for this Logan. The problem doesn’t relate to overwhelming Leo by asking it to transcribe many documents at once. It’s an issue with the model’s ability to transcribe that particular image. Unfortunately, asking it to try again with the same image won’t help. Please see here for my best attempt at explaining why these “hallucinations” tend to happen at the end of transcripts and steps you can take to reduce the chance of it happening.