When translating books using OpenAI, we faced the following issues:
- Due to the 8,000 output token limit, we had to split books into more than 20 parts.
- For each split text, we had to repeatedly input preliminary prompts (translation guidelines, terminology dictionary, character speech patterns, and other reference data).
- This resulted in significant token waste.
The newly released O1, with its 100,000 token limit - ten times larger than before - brings hope that we can translate almost an entire book at once.
This raises several questions:
- Will the input prompts be consistently and accurately applied across all 100,000 tokens without errors?
- Does O1 understand the book in segments or comprehend it all at once?
- What is the probability of encountering errors that would waste our limited budget? (We don’t have unlimited funds, and errors are costly.) The likelihood of errors seems higher when outputting 100,000 tokens at once.