Last year I built an app to translate books. I did layout detection first, then using the layout, I would programmatically craft thousands of prompts to produce a translation.
It worked. It wasn’t perfect, and each translation of a book cost about $5 - $10, but it worked. The main use was for old, even ancient books that no one would care to translate. There is a lot of historical knowledge locked away in books like this.
While it did work, the results weren’t perfect and it did need some hand holding. I didn’t have time to productize it, so it is one of countless prototypes that show me a concept works.
This concept works better than you may think.
Last year I built an app to translate books. I did layout detection first, then using the layout, I would programmatically craft thousands of prompts to produce a translation.
It worked. It wasn’t perfect, and each translation of a book cost about $5 - $10, but it worked. The main use was for old, even ancient books that no one would care to translate. There is a lot of historical knowledge locked away in books like this.
While it did work, the results weren’t perfect and it did need some hand holding. I didn’t have time to productize it, so it is one of countless prototypes that show me a concept works.