I was using it to check if certain error codes is listed in a rice cooker manual. Interestingly, it wasn't listed. But notebookLM didn't find the part documenting about error codes I wanted to look for.
Just like its Gemini family in Google drive, notebookLM has partial OCR. It primarily relies on chunking the selectable text, so the key to obtain higher quality content is relying on making the layout easier for chunker to properly select text
Gemini often boosts it has largest token limitation among all LLM model, and that indeed has certain level of improvement when conducting long interpretation. But pursuing higher token limit for model is not something we can do or should do, as a technical writer, it will be wiser to ensure all pieces are chunk properly.
A typical Japanese manual is often filled with combined tables and pictures (like illustration or chart with figures). It may be good and easy to comprehend for human, but it's challenging for AI and current parser.
留言
張貼留言
Please leave your comments. I will read them in mailbox and reply.